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, struct nlattr *tab) 393 { 394 struct qdisc_rate_table *rtab; 395 396 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 397 nla_len(tab) != TC_RTAB_SIZE) 398 return NULL; 399 400 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 401 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 402 !memcmp(&rtab->data, nla_data(tab), 1024)) { 403 rtab->refcnt++; 404 return rtab; 405 } 406 } 407 408 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 409 if (rtab) { 410 rtab->rate = *r; 411 rtab->refcnt = 1; 412 memcpy(rtab->data, nla_data(tab), 1024); 413 if (r->linklayer == TC_LINKLAYER_UNAWARE) 414 r->linklayer = __detect_linklayer(r, rtab->data); 415 rtab->next = qdisc_rtab_list; 416 qdisc_rtab_list = rtab; 417 } 418 return rtab; 419 } 420 EXPORT_SYMBOL(qdisc_get_rtab); 421 422 void qdisc_put_rtab(struct qdisc_rate_table *tab) 423 { 424 struct qdisc_rate_table *rtab, **rtabp; 425 426 if (!tab || --tab->refcnt) 427 return; 428 429 for (rtabp = &qdisc_rtab_list; 430 (rtab = *rtabp) != NULL; 431 rtabp = &rtab->next) { 432 if (rtab == tab) { 433 *rtabp = rtab->next; 434 kfree(rtab); 435 return; 436 } 437 } 438 } 439 EXPORT_SYMBOL(qdisc_put_rtab); 440 441 static LIST_HEAD(qdisc_stab_list); 442 static DEFINE_SPINLOCK(qdisc_stab_lock); 443 444 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 445 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 446 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 447 }; 448 449 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 450 { 451 struct nlattr *tb[TCA_STAB_MAX + 1]; 452 struct qdisc_size_table *stab; 453 struct tc_sizespec *s; 454 unsigned int tsize = 0; 455 u16 *tab = NULL; 456 int err; 457 458 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); 459 if (err < 0) 460 return ERR_PTR(err); 461 if (!tb[TCA_STAB_BASE]) 462 return ERR_PTR(-EINVAL); 463 464 s = nla_data(tb[TCA_STAB_BASE]); 465 466 if (s->tsize > 0) { 467 if (!tb[TCA_STAB_DATA]) 468 return ERR_PTR(-EINVAL); 469 tab = nla_data(tb[TCA_STAB_DATA]); 470 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 471 } 472 473 if (tsize != s->tsize || (!tab && tsize > 0)) 474 return ERR_PTR(-EINVAL); 475 476 spin_lock(&qdisc_stab_lock); 477 478 list_for_each_entry(stab, &qdisc_stab_list, list) { 479 if (memcmp(&stab->szopts, s, sizeof(*s))) 480 continue; 481 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 482 continue; 483 stab->refcnt++; 484 spin_unlock(&qdisc_stab_lock); 485 return stab; 486 } 487 488 spin_unlock(&qdisc_stab_lock); 489 490 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 491 if (!stab) 492 return ERR_PTR(-ENOMEM); 493 494 stab->refcnt = 1; 495 stab->szopts = *s; 496 if (tsize > 0) 497 memcpy(stab->data, tab, tsize * sizeof(u16)); 498 499 spin_lock(&qdisc_stab_lock); 500 list_add_tail(&stab->list, &qdisc_stab_list); 501 spin_unlock(&qdisc_stab_lock); 502 503 return stab; 504 } 505 506 static void stab_kfree_rcu(struct rcu_head *head) 507 { 508 kfree(container_of(head, struct qdisc_size_table, rcu)); 509 } 510 511 void qdisc_put_stab(struct qdisc_size_table *tab) 512 { 513 if (!tab) 514 return; 515 516 spin_lock(&qdisc_stab_lock); 517 518 if (--tab->refcnt == 0) { 519 list_del(&tab->list); 520 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 521 } 522 523 spin_unlock(&qdisc_stab_lock); 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, const struct qdisc_size_table *stab) 545 { 546 int pkt_len, slot; 547 548 pkt_len = skb->len + stab->szopts.overhead; 549 if (unlikely(!stab->szopts.tsize)) 550 goto out; 551 552 slot = pkt_len + stab->szopts.cell_align; 553 if (unlikely(slot < 0)) 554 slot = 0; 555 556 slot >>= stab->szopts.cell_log; 557 if (likely(slot < stab->szopts.tsize)) 558 pkt_len = stab->data[slot]; 559 else 560 pkt_len = stab->data[stab->szopts.tsize - 1] * 561 (slot / stab->szopts.tsize) + 562 stab->data[slot % stab->szopts.tsize]; 563 564 pkt_len <<= stab->szopts.size_log; 565 out: 566 if (unlikely(pkt_len < 1)) 567 pkt_len = 1; 568 qdisc_skb_cb(skb)->pkt_len = pkt_len; 569 } 570 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 571 572 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 573 { 574 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 575 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 576 txt, qdisc->ops->id, qdisc->handle >> 16); 577 qdisc->flags |= TCQ_F_WARN_NONWC; 578 } 579 } 580 EXPORT_SYMBOL(qdisc_warn_nonwc); 581 582 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 583 { 584 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 585 timer); 586 587 rcu_read_lock(); 588 __netif_schedule(qdisc_root(wd->qdisc)); 589 rcu_read_unlock(); 590 591 return HRTIMER_NORESTART; 592 } 593 594 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 595 { 596 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 597 wd->timer.function = qdisc_watchdog; 598 wd->qdisc = qdisc; 599 } 600 EXPORT_SYMBOL(qdisc_watchdog_init); 601 602 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires) 603 { 604 if (test_bit(__QDISC_STATE_DEACTIVATED, 605 &qdisc_root_sleeping(wd->qdisc)->state)) 606 return; 607 608 if (wd->last_expires == expires) 609 return; 610 611 wd->last_expires = expires; 612 hrtimer_start(&wd->timer, 613 ns_to_ktime(expires), 614 HRTIMER_MODE_ABS_PINNED); 615 } 616 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns); 617 618 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 619 { 620 hrtimer_cancel(&wd->timer); 621 } 622 EXPORT_SYMBOL(qdisc_watchdog_cancel); 623 624 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 625 { 626 unsigned int size = n * sizeof(struct hlist_head), i; 627 struct hlist_head *h; 628 629 if (size <= PAGE_SIZE) 630 h = kmalloc(size, GFP_KERNEL); 631 else 632 h = (struct hlist_head *) 633 __get_free_pages(GFP_KERNEL, get_order(size)); 634 635 if (h != NULL) { 636 for (i = 0; i < n; i++) 637 INIT_HLIST_HEAD(&h[i]); 638 } 639 return h; 640 } 641 642 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 643 { 644 unsigned int size = n * sizeof(struct hlist_head); 645 646 if (size <= PAGE_SIZE) 647 kfree(h); 648 else 649 free_pages((unsigned long)h, get_order(size)); 650 } 651 652 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 653 { 654 struct Qdisc_class_common *cl; 655 struct hlist_node *next; 656 struct hlist_head *nhash, *ohash; 657 unsigned int nsize, nmask, osize; 658 unsigned int i, h; 659 660 /* Rehash when load factor exceeds 0.75 */ 661 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 662 return; 663 nsize = clhash->hashsize * 2; 664 nmask = nsize - 1; 665 nhash = qdisc_class_hash_alloc(nsize); 666 if (nhash == NULL) 667 return; 668 669 ohash = clhash->hash; 670 osize = clhash->hashsize; 671 672 sch_tree_lock(sch); 673 for (i = 0; i < osize; i++) { 674 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 675 h = qdisc_class_hash(cl->classid, nmask); 676 hlist_add_head(&cl->hnode, &nhash[h]); 677 } 678 } 679 clhash->hash = nhash; 680 clhash->hashsize = nsize; 681 clhash->hashmask = nmask; 682 sch_tree_unlock(sch); 683 684 qdisc_class_hash_free(ohash, osize); 685 } 686 EXPORT_SYMBOL(qdisc_class_hash_grow); 687 688 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 689 { 690 unsigned int size = 4; 691 692 clhash->hash = qdisc_class_hash_alloc(size); 693 if (clhash->hash == NULL) 694 return -ENOMEM; 695 clhash->hashsize = size; 696 clhash->hashmask = size - 1; 697 clhash->hashelems = 0; 698 return 0; 699 } 700 EXPORT_SYMBOL(qdisc_class_hash_init); 701 702 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 703 { 704 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 705 } 706 EXPORT_SYMBOL(qdisc_class_hash_destroy); 707 708 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 709 struct Qdisc_class_common *cl) 710 { 711 unsigned int h; 712 713 INIT_HLIST_NODE(&cl->hnode); 714 h = qdisc_class_hash(cl->classid, clhash->hashmask); 715 hlist_add_head(&cl->hnode, &clhash->hash[h]); 716 clhash->hashelems++; 717 } 718 EXPORT_SYMBOL(qdisc_class_hash_insert); 719 720 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 721 struct Qdisc_class_common *cl) 722 { 723 hlist_del(&cl->hnode); 724 clhash->hashelems--; 725 } 726 EXPORT_SYMBOL(qdisc_class_hash_remove); 727 728 /* Allocate an unique handle from space managed by kernel 729 * Possible range is [8000-FFFF]:0000 (0x8000 values) 730 */ 731 static u32 qdisc_alloc_handle(struct net_device *dev) 732 { 733 int i = 0x8000; 734 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 735 736 do { 737 autohandle += TC_H_MAKE(0x10000U, 0); 738 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 739 autohandle = TC_H_MAKE(0x80000000U, 0); 740 if (!qdisc_lookup(dev, autohandle)) 741 return autohandle; 742 cond_resched(); 743 } while (--i > 0); 744 745 return 0; 746 } 747 748 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 749 unsigned int len) 750 { 751 const struct Qdisc_class_ops *cops; 752 unsigned long cl; 753 u32 parentid; 754 int drops; 755 756 if (n == 0 && len == 0) 757 return; 758 drops = max_t(int, n, 0); 759 rcu_read_lock(); 760 while ((parentid = sch->parent)) { 761 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 762 break; 763 764 if (sch->flags & TCQ_F_NOPARENT) 765 break; 766 /* TODO: perform the search on a per txq basis */ 767 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 768 if (sch == NULL) { 769 WARN_ON_ONCE(parentid != TC_H_ROOT); 770 break; 771 } 772 cops = sch->ops->cl_ops; 773 if (cops->qlen_notify) { 774 cl = cops->get(sch, parentid); 775 cops->qlen_notify(sch, cl); 776 cops->put(sch, cl); 777 } 778 sch->q.qlen -= n; 779 sch->qstats.backlog -= len; 780 __qdisc_qstats_drop(sch, drops); 781 } 782 rcu_read_unlock(); 783 } 784 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 785 786 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 787 struct nlmsghdr *n, u32 clid, 788 struct Qdisc *old, struct Qdisc *new) 789 { 790 if (new || old) 791 qdisc_notify(net, skb, n, clid, old, new); 792 793 if (old) 794 qdisc_destroy(old); 795 } 796 797 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 798 * to device "dev". 799 * 800 * When appropriate send a netlink notification using 'skb' 801 * and "n". 802 * 803 * On success, destroy old qdisc. 804 */ 805 806 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 807 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 808 struct Qdisc *new, struct Qdisc *old) 809 { 810 struct Qdisc *q = old; 811 struct net *net = dev_net(dev); 812 int err = 0; 813 814 if (parent == NULL) { 815 unsigned int i, num_q, ingress; 816 817 ingress = 0; 818 num_q = dev->num_tx_queues; 819 if ((q && q->flags & TCQ_F_INGRESS) || 820 (new && new->flags & TCQ_F_INGRESS)) { 821 num_q = 1; 822 ingress = 1; 823 if (!dev_ingress_queue(dev)) 824 return -ENOENT; 825 } 826 827 if (dev->flags & IFF_UP) 828 dev_deactivate(dev); 829 830 if (new && new->ops->attach) 831 goto skip; 832 833 for (i = 0; i < num_q; i++) { 834 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 835 836 if (!ingress) 837 dev_queue = netdev_get_tx_queue(dev, i); 838 839 old = dev_graft_qdisc(dev_queue, new); 840 if (new && i > 0) 841 atomic_inc(&new->refcnt); 842 843 if (!ingress) 844 qdisc_destroy(old); 845 } 846 847 skip: 848 if (!ingress) { 849 notify_and_destroy(net, skb, n, classid, 850 dev->qdisc, new); 851 if (new && !new->ops->attach) 852 atomic_inc(&new->refcnt); 853 dev->qdisc = new ? : &noop_qdisc; 854 855 if (new && new->ops->attach) 856 new->ops->attach(new); 857 } else { 858 notify_and_destroy(net, skb, n, classid, old, new); 859 } 860 861 if (dev->flags & IFF_UP) 862 dev_activate(dev); 863 } else { 864 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 865 866 err = -EOPNOTSUPP; 867 if (cops && cops->graft) { 868 unsigned long cl = cops->get(parent, classid); 869 if (cl) { 870 err = cops->graft(parent, cl, new, &old); 871 cops->put(parent, cl); 872 } else 873 err = -ENOENT; 874 } 875 if (!err) 876 notify_and_destroy(net, skb, n, classid, old, new); 877 } 878 return err; 879 } 880 881 /* lockdep annotation is needed for ingress; egress gets it only for name */ 882 static struct lock_class_key qdisc_tx_lock; 883 static struct lock_class_key qdisc_rx_lock; 884 885 /* 886 Allocate and initialize new qdisc. 887 888 Parameters are passed via opt. 889 */ 890 891 static struct Qdisc * 892 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, 893 struct Qdisc *p, u32 parent, u32 handle, 894 struct nlattr **tca, int *errp) 895 { 896 int err; 897 struct nlattr *kind = tca[TCA_KIND]; 898 struct Qdisc *sch; 899 struct Qdisc_ops *ops; 900 struct qdisc_size_table *stab; 901 902 ops = qdisc_lookup_ops(kind); 903 #ifdef CONFIG_MODULES 904 if (ops == NULL && kind != NULL) { 905 char name[IFNAMSIZ]; 906 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 907 /* We dropped the RTNL semaphore in order to 908 * perform the module load. So, even if we 909 * succeeded in loading the module we have to 910 * tell the caller to replay the request. We 911 * indicate this using -EAGAIN. 912 * We replay the request because the device may 913 * go away in the mean time. 914 */ 915 rtnl_unlock(); 916 request_module("sch_%s", name); 917 rtnl_lock(); 918 ops = qdisc_lookup_ops(kind); 919 if (ops != NULL) { 920 /* We will try again qdisc_lookup_ops, 921 * so don't keep a reference. 922 */ 923 module_put(ops->owner); 924 err = -EAGAIN; 925 goto err_out; 926 } 927 } 928 } 929 #endif 930 931 err = -ENOENT; 932 if (ops == NULL) 933 goto err_out; 934 935 sch = qdisc_alloc(dev_queue, ops); 936 if (IS_ERR(sch)) { 937 err = PTR_ERR(sch); 938 goto err_out2; 939 } 940 941 sch->parent = parent; 942 943 if (handle == TC_H_INGRESS) { 944 sch->flags |= TCQ_F_INGRESS; 945 handle = TC_H_MAKE(TC_H_INGRESS, 0); 946 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 947 } else { 948 if (handle == 0) { 949 handle = qdisc_alloc_handle(dev); 950 err = -ENOMEM; 951 if (handle == 0) 952 goto err_out3; 953 } 954 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 955 if (!netif_is_multiqueue(dev)) 956 sch->flags |= TCQ_F_ONETXQUEUE; 957 } 958 959 sch->handle = handle; 960 961 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 962 if (qdisc_is_percpu_stats(sch)) { 963 sch->cpu_bstats = 964 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 965 if (!sch->cpu_bstats) 966 goto err_out4; 967 968 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 969 if (!sch->cpu_qstats) 970 goto err_out4; 971 } 972 973 if (tca[TCA_STAB]) { 974 stab = qdisc_get_stab(tca[TCA_STAB]); 975 if (IS_ERR(stab)) { 976 err = PTR_ERR(stab); 977 goto err_out4; 978 } 979 rcu_assign_pointer(sch->stab, stab); 980 } 981 if (tca[TCA_RATE]) { 982 seqcount_t *running; 983 984 err = -EOPNOTSUPP; 985 if (sch->flags & TCQ_F_MQROOT) 986 goto err_out4; 987 988 if ((sch->parent != TC_H_ROOT) && 989 !(sch->flags & TCQ_F_INGRESS) && 990 (!p || !(p->flags & TCQ_F_MQROOT))) 991 running = qdisc_root_sleeping_running(sch); 992 else 993 running = &sch->running; 994 995 err = gen_new_estimator(&sch->bstats, 996 sch->cpu_bstats, 997 &sch->rate_est, 998 NULL, 999 running, 1000 tca[TCA_RATE]); 1001 if (err) 1002 goto err_out4; 1003 } 1004 1005 qdisc_hash_add(sch); 1006 1007 return sch; 1008 } 1009 err_out3: 1010 dev_put(dev); 1011 kfree((char *) sch - sch->padded); 1012 err_out2: 1013 module_put(ops->owner); 1014 err_out: 1015 *errp = err; 1016 return NULL; 1017 1018 err_out4: 1019 free_percpu(sch->cpu_bstats); 1020 free_percpu(sch->cpu_qstats); 1021 /* 1022 * Any broken qdiscs that would require a ops->reset() here? 1023 * The qdisc was never in action so it shouldn't be necessary. 1024 */ 1025 qdisc_put_stab(rtnl_dereference(sch->stab)); 1026 if (ops->destroy) 1027 ops->destroy(sch); 1028 goto err_out3; 1029 } 1030 1031 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1032 { 1033 struct qdisc_size_table *ostab, *stab = NULL; 1034 int err = 0; 1035 1036 if (tca[TCA_OPTIONS]) { 1037 if (sch->ops->change == NULL) 1038 return -EINVAL; 1039 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1040 if (err) 1041 return err; 1042 } 1043 1044 if (tca[TCA_STAB]) { 1045 stab = qdisc_get_stab(tca[TCA_STAB]); 1046 if (IS_ERR(stab)) 1047 return PTR_ERR(stab); 1048 } 1049 1050 ostab = rtnl_dereference(sch->stab); 1051 rcu_assign_pointer(sch->stab, stab); 1052 qdisc_put_stab(ostab); 1053 1054 if (tca[TCA_RATE]) { 1055 /* NB: ignores errors from replace_estimator 1056 because change can't be undone. */ 1057 if (sch->flags & TCQ_F_MQROOT) 1058 goto out; 1059 gen_replace_estimator(&sch->bstats, 1060 sch->cpu_bstats, 1061 &sch->rate_est, 1062 NULL, 1063 qdisc_root_sleeping_running(sch), 1064 tca[TCA_RATE]); 1065 } 1066 out: 1067 return 0; 1068 } 1069 1070 struct check_loop_arg { 1071 struct qdisc_walker w; 1072 struct Qdisc *p; 1073 int depth; 1074 }; 1075 1076 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); 1077 1078 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1079 { 1080 struct check_loop_arg arg; 1081 1082 if (q->ops->cl_ops == NULL) 1083 return 0; 1084 1085 arg.w.stop = arg.w.skip = arg.w.count = 0; 1086 arg.w.fn = check_loop_fn; 1087 arg.depth = depth; 1088 arg.p = p; 1089 q->ops->cl_ops->walk(q, &arg.w); 1090 return arg.w.stop ? -ELOOP : 0; 1091 } 1092 1093 static int 1094 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1095 { 1096 struct Qdisc *leaf; 1097 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1098 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1099 1100 leaf = cops->leaf(q, cl); 1101 if (leaf) { 1102 if (leaf == arg->p || arg->depth > 7) 1103 return -ELOOP; 1104 return check_loop(leaf, arg->p, arg->depth + 1); 1105 } 1106 return 0; 1107 } 1108 1109 /* 1110 * Delete/get qdisc. 1111 */ 1112 1113 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1114 { 1115 struct net *net = sock_net(skb->sk); 1116 struct tcmsg *tcm = nlmsg_data(n); 1117 struct nlattr *tca[TCA_MAX + 1]; 1118 struct net_device *dev; 1119 u32 clid; 1120 struct Qdisc *q = NULL; 1121 struct Qdisc *p = NULL; 1122 int err; 1123 1124 if ((n->nlmsg_type != RTM_GETQDISC) && 1125 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1126 return -EPERM; 1127 1128 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1129 if (err < 0) 1130 return err; 1131 1132 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1133 if (!dev) 1134 return -ENODEV; 1135 1136 clid = tcm->tcm_parent; 1137 if (clid) { 1138 if (clid != TC_H_ROOT) { 1139 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1140 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1141 if (!p) 1142 return -ENOENT; 1143 q = qdisc_leaf(p, clid); 1144 } else if (dev_ingress_queue(dev)) { 1145 q = dev_ingress_queue(dev)->qdisc_sleeping; 1146 } 1147 } else { 1148 q = dev->qdisc; 1149 } 1150 if (!q) 1151 return -ENOENT; 1152 1153 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1154 return -EINVAL; 1155 } else { 1156 q = qdisc_lookup(dev, tcm->tcm_handle); 1157 if (!q) 1158 return -ENOENT; 1159 } 1160 1161 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1162 return -EINVAL; 1163 1164 if (n->nlmsg_type == RTM_DELQDISC) { 1165 if (!clid) 1166 return -EINVAL; 1167 if (q->handle == 0) 1168 return -ENOENT; 1169 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1170 if (err != 0) 1171 return err; 1172 } else { 1173 qdisc_notify(net, skb, n, clid, NULL, q); 1174 } 1175 return 0; 1176 } 1177 1178 /* 1179 * Create/change qdisc. 1180 */ 1181 1182 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1183 { 1184 struct net *net = sock_net(skb->sk); 1185 struct tcmsg *tcm; 1186 struct nlattr *tca[TCA_MAX + 1]; 1187 struct net_device *dev; 1188 u32 clid; 1189 struct Qdisc *q, *p; 1190 int err; 1191 1192 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1193 return -EPERM; 1194 1195 replay: 1196 /* Reinit, just in case something touches this. */ 1197 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1198 if (err < 0) 1199 return err; 1200 1201 tcm = nlmsg_data(n); 1202 clid = tcm->tcm_parent; 1203 q = p = NULL; 1204 1205 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1206 if (!dev) 1207 return -ENODEV; 1208 1209 1210 if (clid) { 1211 if (clid != TC_H_ROOT) { 1212 if (clid != TC_H_INGRESS) { 1213 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1214 if (!p) 1215 return -ENOENT; 1216 q = qdisc_leaf(p, clid); 1217 } else if (dev_ingress_queue_create(dev)) { 1218 q = dev_ingress_queue(dev)->qdisc_sleeping; 1219 } 1220 } else { 1221 q = dev->qdisc; 1222 } 1223 1224 /* It may be default qdisc, ignore it */ 1225 if (q && q->handle == 0) 1226 q = NULL; 1227 1228 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1229 if (tcm->tcm_handle) { 1230 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1231 return -EEXIST; 1232 if (TC_H_MIN(tcm->tcm_handle)) 1233 return -EINVAL; 1234 q = qdisc_lookup(dev, tcm->tcm_handle); 1235 if (!q) 1236 goto create_n_graft; 1237 if (n->nlmsg_flags & NLM_F_EXCL) 1238 return -EEXIST; 1239 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1240 return -EINVAL; 1241 if (q == p || 1242 (p && check_loop(q, p, 0))) 1243 return -ELOOP; 1244 atomic_inc(&q->refcnt); 1245 goto graft; 1246 } else { 1247 if (!q) 1248 goto create_n_graft; 1249 1250 /* This magic test requires explanation. 1251 * 1252 * We know, that some child q is already 1253 * attached to this parent and have choice: 1254 * either to change it or to create/graft new one. 1255 * 1256 * 1. We are allowed to create/graft only 1257 * if CREATE and REPLACE flags are set. 1258 * 1259 * 2. If EXCL is set, requestor wanted to say, 1260 * that qdisc tcm_handle is not expected 1261 * to exist, so that we choose create/graft too. 1262 * 1263 * 3. The last case is when no flags are set. 1264 * Alas, it is sort of hole in API, we 1265 * cannot decide what to do unambiguously. 1266 * For now we select create/graft, if 1267 * user gave KIND, which does not match existing. 1268 */ 1269 if ((n->nlmsg_flags & NLM_F_CREATE) && 1270 (n->nlmsg_flags & NLM_F_REPLACE) && 1271 ((n->nlmsg_flags & NLM_F_EXCL) || 1272 (tca[TCA_KIND] && 1273 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1274 goto create_n_graft; 1275 } 1276 } 1277 } else { 1278 if (!tcm->tcm_handle) 1279 return -EINVAL; 1280 q = qdisc_lookup(dev, tcm->tcm_handle); 1281 } 1282 1283 /* Change qdisc parameters */ 1284 if (q == NULL) 1285 return -ENOENT; 1286 if (n->nlmsg_flags & NLM_F_EXCL) 1287 return -EEXIST; 1288 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1289 return -EINVAL; 1290 err = qdisc_change(q, tca); 1291 if (err == 0) 1292 qdisc_notify(net, skb, n, clid, NULL, q); 1293 return err; 1294 1295 create_n_graft: 1296 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1297 return -ENOENT; 1298 if (clid == TC_H_INGRESS) { 1299 if (dev_ingress_queue(dev)) 1300 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1301 tcm->tcm_parent, tcm->tcm_parent, 1302 tca, &err); 1303 else 1304 err = -ENOENT; 1305 } else { 1306 struct netdev_queue *dev_queue; 1307 1308 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1309 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1310 else if (p) 1311 dev_queue = p->dev_queue; 1312 else 1313 dev_queue = netdev_get_tx_queue(dev, 0); 1314 1315 q = qdisc_create(dev, dev_queue, p, 1316 tcm->tcm_parent, tcm->tcm_handle, 1317 tca, &err); 1318 } 1319 if (q == NULL) { 1320 if (err == -EAGAIN) 1321 goto replay; 1322 return err; 1323 } 1324 1325 graft: 1326 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1327 if (err) { 1328 if (q) 1329 qdisc_destroy(q); 1330 return err; 1331 } 1332 1333 return 0; 1334 } 1335 1336 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1337 u32 portid, u32 seq, u16 flags, int event) 1338 { 1339 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 1340 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 1341 struct tcmsg *tcm; 1342 struct nlmsghdr *nlh; 1343 unsigned char *b = skb_tail_pointer(skb); 1344 struct gnet_dump d; 1345 struct qdisc_size_table *stab; 1346 __u32 qlen; 1347 1348 cond_resched(); 1349 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1350 if (!nlh) 1351 goto out_nlmsg_trim; 1352 tcm = nlmsg_data(nlh); 1353 tcm->tcm_family = AF_UNSPEC; 1354 tcm->tcm__pad1 = 0; 1355 tcm->tcm__pad2 = 0; 1356 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1357 tcm->tcm_parent = clid; 1358 tcm->tcm_handle = q->handle; 1359 tcm->tcm_info = atomic_read(&q->refcnt); 1360 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1361 goto nla_put_failure; 1362 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1363 goto nla_put_failure; 1364 qlen = q->q.qlen; 1365 1366 stab = rtnl_dereference(q->stab); 1367 if (stab && qdisc_dump_stab(skb, stab) < 0) 1368 goto nla_put_failure; 1369 1370 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1371 NULL, &d, TCA_PAD) < 0) 1372 goto nla_put_failure; 1373 1374 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1375 goto nla_put_failure; 1376 1377 if (qdisc_is_percpu_stats(q)) { 1378 cpu_bstats = q->cpu_bstats; 1379 cpu_qstats = q->cpu_qstats; 1380 } 1381 1382 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 1383 &d, cpu_bstats, &q->bstats) < 0 || 1384 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 || 1385 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 1386 goto nla_put_failure; 1387 1388 if (gnet_stats_finish_copy(&d) < 0) 1389 goto nla_put_failure; 1390 1391 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1392 return skb->len; 1393 1394 out_nlmsg_trim: 1395 nla_put_failure: 1396 nlmsg_trim(skb, b); 1397 return -1; 1398 } 1399 1400 static bool tc_qdisc_dump_ignore(struct Qdisc *q) 1401 { 1402 return (q->flags & TCQ_F_BUILTIN) ? true : false; 1403 } 1404 1405 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1406 struct nlmsghdr *n, u32 clid, 1407 struct Qdisc *old, struct Qdisc *new) 1408 { 1409 struct sk_buff *skb; 1410 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1411 1412 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1413 if (!skb) 1414 return -ENOBUFS; 1415 1416 if (old && !tc_qdisc_dump_ignore(old)) { 1417 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1418 0, RTM_DELQDISC) < 0) 1419 goto err_out; 1420 } 1421 if (new && !tc_qdisc_dump_ignore(new)) { 1422 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1423 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1424 goto err_out; 1425 } 1426 1427 if (skb->len) 1428 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1429 n->nlmsg_flags & NLM_F_ECHO); 1430 1431 err_out: 1432 kfree_skb(skb); 1433 return -EINVAL; 1434 } 1435 1436 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1437 struct netlink_callback *cb, 1438 int *q_idx_p, int s_q_idx, bool recur) 1439 { 1440 int ret = 0, q_idx = *q_idx_p; 1441 struct Qdisc *q; 1442 int b; 1443 1444 if (!root) 1445 return 0; 1446 1447 q = root; 1448 if (q_idx < s_q_idx) { 1449 q_idx++; 1450 } else { 1451 if (!tc_qdisc_dump_ignore(q) && 1452 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1453 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1454 goto done; 1455 q_idx++; 1456 } 1457 1458 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1459 * itself has already been dumped. 1460 * 1461 * If we've already dumped the top-level (ingress) qdisc above and the global 1462 * qdisc hashtable, we don't want to hit it again 1463 */ 1464 if (!qdisc_dev(root) || !recur) 1465 goto out; 1466 1467 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1468 if (q_idx < s_q_idx) { 1469 q_idx++; 1470 continue; 1471 } 1472 if (!tc_qdisc_dump_ignore(q) && 1473 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1474 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1475 goto done; 1476 q_idx++; 1477 } 1478 1479 out: 1480 *q_idx_p = q_idx; 1481 return ret; 1482 done: 1483 ret = -1; 1484 goto out; 1485 } 1486 1487 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1488 { 1489 struct net *net = sock_net(skb->sk); 1490 int idx, q_idx; 1491 int s_idx, s_q_idx; 1492 struct net_device *dev; 1493 1494 s_idx = cb->args[0]; 1495 s_q_idx = q_idx = cb->args[1]; 1496 1497 idx = 0; 1498 ASSERT_RTNL(); 1499 for_each_netdev(net, dev) { 1500 struct netdev_queue *dev_queue; 1501 1502 if (idx < s_idx) 1503 goto cont; 1504 if (idx > s_idx) 1505 s_q_idx = 0; 1506 q_idx = 0; 1507 1508 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx, true) < 0) 1509 goto done; 1510 1511 dev_queue = dev_ingress_queue(dev); 1512 if (dev_queue && 1513 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1514 &q_idx, s_q_idx, false) < 0) 1515 goto done; 1516 1517 cont: 1518 idx++; 1519 } 1520 1521 done: 1522 cb->args[0] = idx; 1523 cb->args[1] = q_idx; 1524 1525 return skb->len; 1526 } 1527 1528 1529 1530 /************************************************ 1531 * Traffic classes manipulation. * 1532 ************************************************/ 1533 1534 1535 1536 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n) 1537 { 1538 struct net *net = sock_net(skb->sk); 1539 struct tcmsg *tcm = nlmsg_data(n); 1540 struct nlattr *tca[TCA_MAX + 1]; 1541 struct net_device *dev; 1542 struct Qdisc *q = NULL; 1543 const struct Qdisc_class_ops *cops; 1544 unsigned long cl = 0; 1545 unsigned long new_cl; 1546 u32 portid; 1547 u32 clid; 1548 u32 qid; 1549 int err; 1550 1551 if ((n->nlmsg_type != RTM_GETTCLASS) && 1552 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1553 return -EPERM; 1554 1555 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1556 if (err < 0) 1557 return err; 1558 1559 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1560 if (!dev) 1561 return -ENODEV; 1562 1563 /* 1564 parent == TC_H_UNSPEC - unspecified parent. 1565 parent == TC_H_ROOT - class is root, which has no parent. 1566 parent == X:0 - parent is root class. 1567 parent == X:Y - parent is a node in hierarchy. 1568 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1569 1570 handle == 0:0 - generate handle from kernel pool. 1571 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1572 handle == X:Y - clear. 1573 handle == X:0 - root class. 1574 */ 1575 1576 /* Step 1. Determine qdisc handle X:0 */ 1577 1578 portid = tcm->tcm_parent; 1579 clid = tcm->tcm_handle; 1580 qid = TC_H_MAJ(clid); 1581 1582 if (portid != TC_H_ROOT) { 1583 u32 qid1 = TC_H_MAJ(portid); 1584 1585 if (qid && qid1) { 1586 /* If both majors are known, they must be identical. */ 1587 if (qid != qid1) 1588 return -EINVAL; 1589 } else if (qid1) { 1590 qid = qid1; 1591 } else if (qid == 0) 1592 qid = dev->qdisc->handle; 1593 1594 /* Now qid is genuine qdisc handle consistent 1595 * both with parent and child. 1596 * 1597 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1598 */ 1599 if (portid) 1600 portid = TC_H_MAKE(qid, portid); 1601 } else { 1602 if (qid == 0) 1603 qid = dev->qdisc->handle; 1604 } 1605 1606 /* OK. Locate qdisc */ 1607 q = qdisc_lookup(dev, qid); 1608 if (!q) 1609 return -ENOENT; 1610 1611 /* An check that it supports classes */ 1612 cops = q->ops->cl_ops; 1613 if (cops == NULL) 1614 return -EINVAL; 1615 1616 /* Now try to get class */ 1617 if (clid == 0) { 1618 if (portid == TC_H_ROOT) 1619 clid = qid; 1620 } else 1621 clid = TC_H_MAKE(qid, clid); 1622 1623 if (clid) 1624 cl = cops->get(q, clid); 1625 1626 if (cl == 0) { 1627 err = -ENOENT; 1628 if (n->nlmsg_type != RTM_NEWTCLASS || 1629 !(n->nlmsg_flags & NLM_F_CREATE)) 1630 goto out; 1631 } else { 1632 switch (n->nlmsg_type) { 1633 case RTM_NEWTCLASS: 1634 err = -EEXIST; 1635 if (n->nlmsg_flags & NLM_F_EXCL) 1636 goto out; 1637 break; 1638 case RTM_DELTCLASS: 1639 err = -EOPNOTSUPP; 1640 if (cops->delete) 1641 err = cops->delete(q, cl); 1642 if (err == 0) 1643 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS); 1644 goto out; 1645 case RTM_GETTCLASS: 1646 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1647 goto out; 1648 default: 1649 err = -EINVAL; 1650 goto out; 1651 } 1652 } 1653 1654 new_cl = cl; 1655 err = -EOPNOTSUPP; 1656 if (cops->change) 1657 err = cops->change(q, clid, portid, tca, &new_cl); 1658 if (err == 0) 1659 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1660 1661 out: 1662 if (cl) 1663 cops->put(q, cl); 1664 1665 return err; 1666 } 1667 1668 1669 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1670 unsigned long cl, 1671 u32 portid, u32 seq, u16 flags, int event) 1672 { 1673 struct tcmsg *tcm; 1674 struct nlmsghdr *nlh; 1675 unsigned char *b = skb_tail_pointer(skb); 1676 struct gnet_dump d; 1677 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1678 1679 cond_resched(); 1680 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1681 if (!nlh) 1682 goto out_nlmsg_trim; 1683 tcm = nlmsg_data(nlh); 1684 tcm->tcm_family = AF_UNSPEC; 1685 tcm->tcm__pad1 = 0; 1686 tcm->tcm__pad2 = 0; 1687 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1688 tcm->tcm_parent = q->handle; 1689 tcm->tcm_handle = q->handle; 1690 tcm->tcm_info = 0; 1691 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1692 goto nla_put_failure; 1693 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1694 goto nla_put_failure; 1695 1696 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1697 NULL, &d, TCA_PAD) < 0) 1698 goto nla_put_failure; 1699 1700 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1701 goto nla_put_failure; 1702 1703 if (gnet_stats_finish_copy(&d) < 0) 1704 goto nla_put_failure; 1705 1706 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1707 return skb->len; 1708 1709 out_nlmsg_trim: 1710 nla_put_failure: 1711 nlmsg_trim(skb, b); 1712 return -1; 1713 } 1714 1715 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1716 struct nlmsghdr *n, struct Qdisc *q, 1717 unsigned long cl, int event) 1718 { 1719 struct sk_buff *skb; 1720 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1721 1722 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1723 if (!skb) 1724 return -ENOBUFS; 1725 1726 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1727 kfree_skb(skb); 1728 return -EINVAL; 1729 } 1730 1731 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1732 n->nlmsg_flags & NLM_F_ECHO); 1733 } 1734 1735 struct qdisc_dump_args { 1736 struct qdisc_walker w; 1737 struct sk_buff *skb; 1738 struct netlink_callback *cb; 1739 }; 1740 1741 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) 1742 { 1743 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1744 1745 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1746 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); 1747 } 1748 1749 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1750 struct tcmsg *tcm, struct netlink_callback *cb, 1751 int *t_p, int s_t) 1752 { 1753 struct qdisc_dump_args arg; 1754 1755 if (tc_qdisc_dump_ignore(q) || 1756 *t_p < s_t || !q->ops->cl_ops || 1757 (tcm->tcm_parent && 1758 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1759 (*t_p)++; 1760 return 0; 1761 } 1762 if (*t_p > s_t) 1763 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1764 arg.w.fn = qdisc_class_dump; 1765 arg.skb = skb; 1766 arg.cb = cb; 1767 arg.w.stop = 0; 1768 arg.w.skip = cb->args[1]; 1769 arg.w.count = 0; 1770 q->ops->cl_ops->walk(q, &arg.w); 1771 cb->args[1] = arg.w.count; 1772 if (arg.w.stop) 1773 return -1; 1774 (*t_p)++; 1775 return 0; 1776 } 1777 1778 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1779 struct tcmsg *tcm, struct netlink_callback *cb, 1780 int *t_p, int s_t) 1781 { 1782 struct Qdisc *q; 1783 int b; 1784 1785 if (!root) 1786 return 0; 1787 1788 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1789 return -1; 1790 1791 if (!qdisc_dev(root)) 1792 return 0; 1793 1794 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1795 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1796 return -1; 1797 } 1798 1799 return 0; 1800 } 1801 1802 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1803 { 1804 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1805 struct net *net = sock_net(skb->sk); 1806 struct netdev_queue *dev_queue; 1807 struct net_device *dev; 1808 int t, s_t; 1809 1810 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1811 return 0; 1812 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1813 if (!dev) 1814 return 0; 1815 1816 s_t = cb->args[0]; 1817 t = 0; 1818 1819 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1820 goto done; 1821 1822 dev_queue = dev_ingress_queue(dev); 1823 if (dev_queue && 1824 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1825 &t, s_t) < 0) 1826 goto done; 1827 1828 done: 1829 cb->args[0] = t; 1830 1831 dev_put(dev); 1832 return skb->len; 1833 } 1834 1835 /* Main classifier routine: scans classifier chain attached 1836 * to this qdisc, (optionally) tests for protocol and asks 1837 * specific classifiers. 1838 */ 1839 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp, 1840 struct tcf_result *res, bool compat_mode) 1841 { 1842 __be16 protocol = tc_skb_protocol(skb); 1843 #ifdef CONFIG_NET_CLS_ACT 1844 const struct tcf_proto *old_tp = tp; 1845 int limit = 0; 1846 1847 reclassify: 1848 #endif 1849 for (; tp; tp = rcu_dereference_bh(tp->next)) { 1850 int err; 1851 1852 if (tp->protocol != protocol && 1853 tp->protocol != htons(ETH_P_ALL)) 1854 continue; 1855 1856 err = tp->classify(skb, tp, res); 1857 #ifdef CONFIG_NET_CLS_ACT 1858 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) 1859 goto reset; 1860 #endif 1861 if (err >= 0) 1862 return err; 1863 } 1864 1865 return TC_ACT_UNSPEC; /* signal: continue lookup */ 1866 #ifdef CONFIG_NET_CLS_ACT 1867 reset: 1868 if (unlikely(limit++ >= MAX_REC_LOOP)) { 1869 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 1870 tp->q->ops->id, tp->prio & 0xffff, 1871 ntohs(tp->protocol)); 1872 return TC_ACT_SHOT; 1873 } 1874 1875 tp = old_tp; 1876 protocol = tc_skb_protocol(skb); 1877 goto reclassify; 1878 #endif 1879 } 1880 EXPORT_SYMBOL(tc_classify); 1881 1882 bool tcf_destroy(struct tcf_proto *tp, bool force) 1883 { 1884 if (tp->ops->destroy(tp, force)) { 1885 module_put(tp->ops->owner); 1886 kfree_rcu(tp, rcu); 1887 return true; 1888 } 1889 1890 return false; 1891 } 1892 1893 void tcf_destroy_chain(struct tcf_proto __rcu **fl) 1894 { 1895 struct tcf_proto *tp; 1896 1897 while ((tp = rtnl_dereference(*fl)) != NULL) { 1898 RCU_INIT_POINTER(*fl, tp->next); 1899 tcf_destroy(tp, true); 1900 } 1901 } 1902 EXPORT_SYMBOL(tcf_destroy_chain); 1903 1904 #ifdef CONFIG_PROC_FS 1905 static int psched_show(struct seq_file *seq, void *v) 1906 { 1907 seq_printf(seq, "%08x %08x %08x %08x\n", 1908 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1909 1000000, 1910 (u32)NSEC_PER_SEC / hrtimer_resolution); 1911 1912 return 0; 1913 } 1914 1915 static int psched_open(struct inode *inode, struct file *file) 1916 { 1917 return single_open(file, psched_show, NULL); 1918 } 1919 1920 static const struct file_operations psched_fops = { 1921 .owner = THIS_MODULE, 1922 .open = psched_open, 1923 .read = seq_read, 1924 .llseek = seq_lseek, 1925 .release = single_release, 1926 }; 1927 1928 static int __net_init psched_net_init(struct net *net) 1929 { 1930 struct proc_dir_entry *e; 1931 1932 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1933 if (e == NULL) 1934 return -ENOMEM; 1935 1936 return 0; 1937 } 1938 1939 static void __net_exit psched_net_exit(struct net *net) 1940 { 1941 remove_proc_entry("psched", net->proc_net); 1942 } 1943 #else 1944 static int __net_init psched_net_init(struct net *net) 1945 { 1946 return 0; 1947 } 1948 1949 static void __net_exit psched_net_exit(struct net *net) 1950 { 1951 } 1952 #endif 1953 1954 static struct pernet_operations psched_net_ops = { 1955 .init = psched_net_init, 1956 .exit = psched_net_exit, 1957 }; 1958 1959 static int __init pktsched_init(void) 1960 { 1961 int err; 1962 1963 err = register_pernet_subsys(&psched_net_ops); 1964 if (err) { 1965 pr_err("pktsched_init: " 1966 "cannot initialize per netns operations\n"); 1967 return err; 1968 } 1969 1970 register_qdisc(&pfifo_fast_ops); 1971 register_qdisc(&pfifo_qdisc_ops); 1972 register_qdisc(&bfifo_qdisc_ops); 1973 register_qdisc(&pfifo_head_drop_qdisc_ops); 1974 register_qdisc(&mq_qdisc_ops); 1975 register_qdisc(&noqueue_qdisc_ops); 1976 1977 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); 1978 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); 1979 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL); 1980 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); 1981 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); 1982 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL); 1983 1984 return 0; 1985 } 1986 1987 subsys_initcall(pktsched_init); 1988