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 33 #include <net/net_namespace.h> 34 #include <net/sock.h> 35 #include <net/netlink.h> 36 #include <net/pkt_sched.h> 37 38 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 39 struct nlmsghdr *n, u32 clid, 40 struct Qdisc *old, struct Qdisc *new); 41 static int tclass_notify(struct net *net, struct sk_buff *oskb, 42 struct nlmsghdr *n, struct Qdisc *q, 43 unsigned long cl, int event); 44 45 /* 46 47 Short review. 48 ------------- 49 50 This file consists of two interrelated parts: 51 52 1. queueing disciplines manager frontend. 53 2. traffic classes manager frontend. 54 55 Generally, queueing discipline ("qdisc") is a black box, 56 which is able to enqueue packets and to dequeue them (when 57 device is ready to send something) in order and at times 58 determined by algorithm hidden in it. 59 60 qdisc's are divided to two categories: 61 - "queues", which have no internal structure visible from outside. 62 - "schedulers", which split all the packets to "traffic classes", 63 using "packet classifiers" (look at cls_api.c) 64 65 In turn, classes may have child qdiscs (as rule, queues) 66 attached to them etc. etc. etc. 67 68 The goal of the routines in this file is to translate 69 information supplied by user in the form of handles 70 to more intelligible for kernel form, to make some sanity 71 checks and part of work, which is common to all qdiscs 72 and to provide rtnetlink notifications. 73 74 All real intelligent work is done inside qdisc modules. 75 76 77 78 Every discipline has two major routines: enqueue and dequeue. 79 80 ---dequeue 81 82 dequeue usually returns a skb to send. It is allowed to return NULL, 83 but it does not mean that queue is empty, it just means that 84 discipline does not want to send anything this time. 85 Queue is really empty if q->q.qlen == 0. 86 For complicated disciplines with multiple queues q->q is not 87 real packet queue, but however q->q.qlen must be valid. 88 89 ---enqueue 90 91 enqueue returns 0, if packet was enqueued successfully. 92 If packet (this one or another one) was dropped, it returns 93 not zero error code. 94 NET_XMIT_DROP - this packet dropped 95 Expected action: do not backoff, but wait until queue will clear. 96 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 97 Expected action: backoff or ignore 98 NET_XMIT_POLICED - dropped by police. 99 Expected action: backoff or error to real-time apps. 100 101 Auxiliary routines: 102 103 ---peek 104 105 like dequeue but without removing a packet from the queue 106 107 ---reset 108 109 returns qdisc to initial state: purge all buffers, clear all 110 timers, counters (except for statistics) etc. 111 112 ---init 113 114 initializes newly created qdisc. 115 116 ---destroy 117 118 destroys resources allocated by init and during lifetime of qdisc. 119 120 ---change 121 122 changes qdisc parameters. 123 */ 124 125 /* Protects list of registered TC modules. It is pure SMP lock. */ 126 static DEFINE_RWLOCK(qdisc_mod_lock); 127 128 129 /************************************************ 130 * Queueing disciplines manipulation. * 131 ************************************************/ 132 133 134 /* The list of all installed queueing disciplines. */ 135 136 static struct Qdisc_ops *qdisc_base; 137 138 /* Register/unregister queueing discipline */ 139 140 int register_qdisc(struct Qdisc_ops *qops) 141 { 142 struct Qdisc_ops *q, **qp; 143 int rc = -EEXIST; 144 145 write_lock(&qdisc_mod_lock); 146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 147 if (!strcmp(qops->id, q->id)) 148 goto out; 149 150 if (qops->enqueue == NULL) 151 qops->enqueue = noop_qdisc_ops.enqueue; 152 if (qops->peek == NULL) { 153 if (qops->dequeue == NULL) 154 qops->peek = noop_qdisc_ops.peek; 155 else 156 goto out_einval; 157 } 158 if (qops->dequeue == NULL) 159 qops->dequeue = noop_qdisc_ops.dequeue; 160 161 if (qops->cl_ops) { 162 const struct Qdisc_class_ops *cops = qops->cl_ops; 163 164 if (!(cops->get && cops->put && cops->walk && cops->leaf)) 165 goto out_einval; 166 167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf)) 168 goto out_einval; 169 } 170 171 qops->next = NULL; 172 *qp = qops; 173 rc = 0; 174 out: 175 write_unlock(&qdisc_mod_lock); 176 return rc; 177 178 out_einval: 179 rc = -EINVAL; 180 goto out; 181 } 182 EXPORT_SYMBOL(register_qdisc); 183 184 int unregister_qdisc(struct Qdisc_ops *qops) 185 { 186 struct Qdisc_ops *q, **qp; 187 int err = -ENOENT; 188 189 write_lock(&qdisc_mod_lock); 190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 191 if (q == qops) 192 break; 193 if (q) { 194 *qp = q->next; 195 q->next = NULL; 196 err = 0; 197 } 198 write_unlock(&qdisc_mod_lock); 199 return err; 200 } 201 EXPORT_SYMBOL(unregister_qdisc); 202 203 /* Get default qdisc if not otherwise specified */ 204 void qdisc_get_default(char *name, size_t len) 205 { 206 read_lock(&qdisc_mod_lock); 207 strlcpy(name, default_qdisc_ops->id, len); 208 read_unlock(&qdisc_mod_lock); 209 } 210 211 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 212 { 213 struct Qdisc_ops *q = NULL; 214 215 for (q = qdisc_base; q; q = q->next) { 216 if (!strcmp(name, q->id)) { 217 if (!try_module_get(q->owner)) 218 q = NULL; 219 break; 220 } 221 } 222 223 return q; 224 } 225 226 /* Set new default qdisc to use */ 227 int qdisc_set_default(const char *name) 228 { 229 const struct Qdisc_ops *ops; 230 231 if (!capable(CAP_NET_ADMIN)) 232 return -EPERM; 233 234 write_lock(&qdisc_mod_lock); 235 ops = qdisc_lookup_default(name); 236 if (!ops) { 237 /* Not found, drop lock and try to load module */ 238 write_unlock(&qdisc_mod_lock); 239 request_module("sch_%s", name); 240 write_lock(&qdisc_mod_lock); 241 242 ops = qdisc_lookup_default(name); 243 } 244 245 if (ops) { 246 /* Set new default */ 247 module_put(default_qdisc_ops->owner); 248 default_qdisc_ops = ops; 249 } 250 write_unlock(&qdisc_mod_lock); 251 252 return ops ? 0 : -ENOENT; 253 } 254 255 /* We know handle. Find qdisc among all qdisc's attached to device 256 * (root qdisc, all its children, children of children etc.) 257 * Note: caller either uses rtnl or rcu_read_lock() 258 */ 259 260 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 261 { 262 struct Qdisc *q; 263 264 if (!(root->flags & TCQ_F_BUILTIN) && 265 root->handle == handle) 266 return root; 267 268 list_for_each_entry_rcu(q, &root->list, list) { 269 if (q->handle == handle) 270 return q; 271 } 272 return NULL; 273 } 274 275 void qdisc_list_add(struct Qdisc *q) 276 { 277 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 278 struct Qdisc *root = qdisc_dev(q)->qdisc; 279 280 WARN_ON_ONCE(root == &noop_qdisc); 281 ASSERT_RTNL(); 282 list_add_tail_rcu(&q->list, &root->list); 283 } 284 } 285 EXPORT_SYMBOL(qdisc_list_add); 286 287 void qdisc_list_del(struct Qdisc *q) 288 { 289 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 290 ASSERT_RTNL(); 291 list_del_rcu(&q->list); 292 } 293 } 294 EXPORT_SYMBOL(qdisc_list_del); 295 296 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 297 { 298 struct Qdisc *q; 299 300 q = qdisc_match_from_root(dev->qdisc, handle); 301 if (q) 302 goto out; 303 304 if (dev_ingress_queue(dev)) 305 q = qdisc_match_from_root( 306 dev_ingress_queue(dev)->qdisc_sleeping, 307 handle); 308 out: 309 return q; 310 } 311 312 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 313 { 314 unsigned long cl; 315 struct Qdisc *leaf; 316 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 317 318 if (cops == NULL) 319 return NULL; 320 cl = cops->get(p, classid); 321 322 if (cl == 0) 323 return NULL; 324 leaf = cops->leaf(p, cl); 325 cops->put(p, cl); 326 return leaf; 327 } 328 329 /* Find queueing discipline by name */ 330 331 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 332 { 333 struct Qdisc_ops *q = NULL; 334 335 if (kind) { 336 read_lock(&qdisc_mod_lock); 337 for (q = qdisc_base; q; q = q->next) { 338 if (nla_strcmp(kind, q->id) == 0) { 339 if (!try_module_get(q->owner)) 340 q = NULL; 341 break; 342 } 343 } 344 read_unlock(&qdisc_mod_lock); 345 } 346 return q; 347 } 348 349 /* The linklayer setting were not transferred from iproute2, in older 350 * versions, and the rate tables lookup systems have been dropped in 351 * the kernel. To keep backward compatible with older iproute2 tc 352 * utils, we detect the linklayer setting by detecting if the rate 353 * table were modified. 354 * 355 * For linklayer ATM table entries, the rate table will be aligned to 356 * 48 bytes, thus some table entries will contain the same value. The 357 * mpu (min packet unit) is also encoded into the old rate table, thus 358 * starting from the mpu, we find low and high table entries for 359 * mapping this cell. If these entries contain the same value, when 360 * the rate tables have been modified for linklayer ATM. 361 * 362 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 363 * and then roundup to the next cell, calc the table entry one below, 364 * and compare. 365 */ 366 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 367 { 368 int low = roundup(r->mpu, 48); 369 int high = roundup(low+1, 48); 370 int cell_low = low >> r->cell_log; 371 int cell_high = (high >> r->cell_log) - 1; 372 373 /* rtab is too inaccurate at rates > 100Mbit/s */ 374 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 375 pr_debug("TC linklayer: Giving up ATM detection\n"); 376 return TC_LINKLAYER_ETHERNET; 377 } 378 379 if ((cell_high > cell_low) && (cell_high < 256) 380 && (rtab[cell_low] == rtab[cell_high])) { 381 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 382 cell_low, cell_high, rtab[cell_high]); 383 return TC_LINKLAYER_ATM; 384 } 385 return TC_LINKLAYER_ETHERNET; 386 } 387 388 static struct qdisc_rate_table *qdisc_rtab_list; 389 390 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) 391 { 392 struct qdisc_rate_table *rtab; 393 394 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 395 nla_len(tab) != TC_RTAB_SIZE) 396 return NULL; 397 398 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 399 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 400 !memcmp(&rtab->data, nla_data(tab), 1024)) { 401 rtab->refcnt++; 402 return rtab; 403 } 404 } 405 406 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 407 if (rtab) { 408 rtab->rate = *r; 409 rtab->refcnt = 1; 410 memcpy(rtab->data, nla_data(tab), 1024); 411 if (r->linklayer == TC_LINKLAYER_UNAWARE) 412 r->linklayer = __detect_linklayer(r, rtab->data); 413 rtab->next = qdisc_rtab_list; 414 qdisc_rtab_list = rtab; 415 } 416 return rtab; 417 } 418 EXPORT_SYMBOL(qdisc_get_rtab); 419 420 void qdisc_put_rtab(struct qdisc_rate_table *tab) 421 { 422 struct qdisc_rate_table *rtab, **rtabp; 423 424 if (!tab || --tab->refcnt) 425 return; 426 427 for (rtabp = &qdisc_rtab_list; 428 (rtab = *rtabp) != NULL; 429 rtabp = &rtab->next) { 430 if (rtab == tab) { 431 *rtabp = rtab->next; 432 kfree(rtab); 433 return; 434 } 435 } 436 } 437 EXPORT_SYMBOL(qdisc_put_rtab); 438 439 static LIST_HEAD(qdisc_stab_list); 440 static DEFINE_SPINLOCK(qdisc_stab_lock); 441 442 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 443 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 444 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 445 }; 446 447 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 448 { 449 struct nlattr *tb[TCA_STAB_MAX + 1]; 450 struct qdisc_size_table *stab; 451 struct tc_sizespec *s; 452 unsigned int tsize = 0; 453 u16 *tab = NULL; 454 int err; 455 456 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); 457 if (err < 0) 458 return ERR_PTR(err); 459 if (!tb[TCA_STAB_BASE]) 460 return ERR_PTR(-EINVAL); 461 462 s = nla_data(tb[TCA_STAB_BASE]); 463 464 if (s->tsize > 0) { 465 if (!tb[TCA_STAB_DATA]) 466 return ERR_PTR(-EINVAL); 467 tab = nla_data(tb[TCA_STAB_DATA]); 468 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 469 } 470 471 if (tsize != s->tsize || (!tab && tsize > 0)) 472 return ERR_PTR(-EINVAL); 473 474 spin_lock(&qdisc_stab_lock); 475 476 list_for_each_entry(stab, &qdisc_stab_list, list) { 477 if (memcmp(&stab->szopts, s, sizeof(*s))) 478 continue; 479 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 480 continue; 481 stab->refcnt++; 482 spin_unlock(&qdisc_stab_lock); 483 return stab; 484 } 485 486 spin_unlock(&qdisc_stab_lock); 487 488 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 489 if (!stab) 490 return ERR_PTR(-ENOMEM); 491 492 stab->refcnt = 1; 493 stab->szopts = *s; 494 if (tsize > 0) 495 memcpy(stab->data, tab, tsize * sizeof(u16)); 496 497 spin_lock(&qdisc_stab_lock); 498 list_add_tail(&stab->list, &qdisc_stab_list); 499 spin_unlock(&qdisc_stab_lock); 500 501 return stab; 502 } 503 504 static void stab_kfree_rcu(struct rcu_head *head) 505 { 506 kfree(container_of(head, struct qdisc_size_table, rcu)); 507 } 508 509 void qdisc_put_stab(struct qdisc_size_table *tab) 510 { 511 if (!tab) 512 return; 513 514 spin_lock(&qdisc_stab_lock); 515 516 if (--tab->refcnt == 0) { 517 list_del(&tab->list); 518 call_rcu_bh(&tab->rcu, stab_kfree_rcu); 519 } 520 521 spin_unlock(&qdisc_stab_lock); 522 } 523 EXPORT_SYMBOL(qdisc_put_stab); 524 525 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 526 { 527 struct nlattr *nest; 528 529 nest = nla_nest_start(skb, TCA_STAB); 530 if (nest == NULL) 531 goto nla_put_failure; 532 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 533 goto nla_put_failure; 534 nla_nest_end(skb, nest); 535 536 return skb->len; 537 538 nla_put_failure: 539 return -1; 540 } 541 542 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab) 543 { 544 int pkt_len, slot; 545 546 pkt_len = skb->len + stab->szopts.overhead; 547 if (unlikely(!stab->szopts.tsize)) 548 goto out; 549 550 slot = pkt_len + stab->szopts.cell_align; 551 if (unlikely(slot < 0)) 552 slot = 0; 553 554 slot >>= stab->szopts.cell_log; 555 if (likely(slot < stab->szopts.tsize)) 556 pkt_len = stab->data[slot]; 557 else 558 pkt_len = stab->data[stab->szopts.tsize - 1] * 559 (slot / stab->szopts.tsize) + 560 stab->data[slot % stab->szopts.tsize]; 561 562 pkt_len <<= stab->szopts.size_log; 563 out: 564 if (unlikely(pkt_len < 1)) 565 pkt_len = 1; 566 qdisc_skb_cb(skb)->pkt_len = pkt_len; 567 } 568 EXPORT_SYMBOL(__qdisc_calculate_pkt_len); 569 570 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 571 { 572 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 573 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 574 txt, qdisc->ops->id, qdisc->handle >> 16); 575 qdisc->flags |= TCQ_F_WARN_NONWC; 576 } 577 } 578 EXPORT_SYMBOL(qdisc_warn_nonwc); 579 580 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 581 { 582 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 583 timer); 584 585 rcu_read_lock(); 586 qdisc_unthrottled(wd->qdisc); 587 __netif_schedule(qdisc_root(wd->qdisc)); 588 rcu_read_unlock(); 589 590 return HRTIMER_NORESTART; 591 } 592 593 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 594 { 595 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED); 596 wd->timer.function = qdisc_watchdog; 597 wd->qdisc = qdisc; 598 } 599 EXPORT_SYMBOL(qdisc_watchdog_init); 600 601 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle) 602 { 603 if (test_bit(__QDISC_STATE_DEACTIVATED, 604 &qdisc_root_sleeping(wd->qdisc)->state)) 605 return; 606 607 if (throttle) 608 qdisc_throttled(wd->qdisc); 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 qdisc_unthrottled(wd->qdisc); 624 } 625 EXPORT_SYMBOL(qdisc_watchdog_cancel); 626 627 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 628 { 629 unsigned int size = n * sizeof(struct hlist_head), i; 630 struct hlist_head *h; 631 632 if (size <= PAGE_SIZE) 633 h = kmalloc(size, GFP_KERNEL); 634 else 635 h = (struct hlist_head *) 636 __get_free_pages(GFP_KERNEL, get_order(size)); 637 638 if (h != NULL) { 639 for (i = 0; i < n; i++) 640 INIT_HLIST_HEAD(&h[i]); 641 } 642 return h; 643 } 644 645 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 646 { 647 unsigned int size = n * sizeof(struct hlist_head); 648 649 if (size <= PAGE_SIZE) 650 kfree(h); 651 else 652 free_pages((unsigned long)h, get_order(size)); 653 } 654 655 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 656 { 657 struct Qdisc_class_common *cl; 658 struct hlist_node *next; 659 struct hlist_head *nhash, *ohash; 660 unsigned int nsize, nmask, osize; 661 unsigned int i, h; 662 663 /* Rehash when load factor exceeds 0.75 */ 664 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 665 return; 666 nsize = clhash->hashsize * 2; 667 nmask = nsize - 1; 668 nhash = qdisc_class_hash_alloc(nsize); 669 if (nhash == NULL) 670 return; 671 672 ohash = clhash->hash; 673 osize = clhash->hashsize; 674 675 sch_tree_lock(sch); 676 for (i = 0; i < osize; i++) { 677 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 678 h = qdisc_class_hash(cl->classid, nmask); 679 hlist_add_head(&cl->hnode, &nhash[h]); 680 } 681 } 682 clhash->hash = nhash; 683 clhash->hashsize = nsize; 684 clhash->hashmask = nmask; 685 sch_tree_unlock(sch); 686 687 qdisc_class_hash_free(ohash, osize); 688 } 689 EXPORT_SYMBOL(qdisc_class_hash_grow); 690 691 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 692 { 693 unsigned int size = 4; 694 695 clhash->hash = qdisc_class_hash_alloc(size); 696 if (clhash->hash == NULL) 697 return -ENOMEM; 698 clhash->hashsize = size; 699 clhash->hashmask = size - 1; 700 clhash->hashelems = 0; 701 return 0; 702 } 703 EXPORT_SYMBOL(qdisc_class_hash_init); 704 705 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 706 { 707 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 708 } 709 EXPORT_SYMBOL(qdisc_class_hash_destroy); 710 711 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 712 struct Qdisc_class_common *cl) 713 { 714 unsigned int h; 715 716 INIT_HLIST_NODE(&cl->hnode); 717 h = qdisc_class_hash(cl->classid, clhash->hashmask); 718 hlist_add_head(&cl->hnode, &clhash->hash[h]); 719 clhash->hashelems++; 720 } 721 EXPORT_SYMBOL(qdisc_class_hash_insert); 722 723 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 724 struct Qdisc_class_common *cl) 725 { 726 hlist_del(&cl->hnode); 727 clhash->hashelems--; 728 } 729 EXPORT_SYMBOL(qdisc_class_hash_remove); 730 731 /* Allocate an unique handle from space managed by kernel 732 * Possible range is [8000-FFFF]:0000 (0x8000 values) 733 */ 734 static u32 qdisc_alloc_handle(struct net_device *dev) 735 { 736 int i = 0x8000; 737 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 738 739 do { 740 autohandle += TC_H_MAKE(0x10000U, 0); 741 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 742 autohandle = TC_H_MAKE(0x80000000U, 0); 743 if (!qdisc_lookup(dev, autohandle)) 744 return autohandle; 745 cond_resched(); 746 } while (--i > 0); 747 748 return 0; 749 } 750 751 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n, 752 unsigned int len) 753 { 754 const struct Qdisc_class_ops *cops; 755 unsigned long cl; 756 u32 parentid; 757 int drops; 758 759 if (n == 0 && len == 0) 760 return; 761 drops = max_t(int, n, 0); 762 rcu_read_lock(); 763 while ((parentid = sch->parent)) { 764 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 765 break; 766 767 if (sch->flags & TCQ_F_NOPARENT) 768 break; 769 /* TODO: perform the search on a per txq basis */ 770 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 771 if (sch == NULL) { 772 WARN_ON_ONCE(parentid != TC_H_ROOT); 773 break; 774 } 775 cops = sch->ops->cl_ops; 776 if (cops->qlen_notify) { 777 cl = cops->get(sch, parentid); 778 cops->qlen_notify(sch, cl); 779 cops->put(sch, cl); 780 } 781 sch->q.qlen -= n; 782 sch->qstats.backlog -= len; 783 __qdisc_qstats_drop(sch, drops); 784 } 785 rcu_read_unlock(); 786 } 787 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 788 789 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 790 struct nlmsghdr *n, u32 clid, 791 struct Qdisc *old, struct Qdisc *new) 792 { 793 if (new || old) 794 qdisc_notify(net, skb, n, clid, old, new); 795 796 if (old) 797 qdisc_destroy(old); 798 } 799 800 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 801 * to device "dev". 802 * 803 * When appropriate send a netlink notification using 'skb' 804 * and "n". 805 * 806 * On success, destroy old qdisc. 807 */ 808 809 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 810 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 811 struct Qdisc *new, struct Qdisc *old) 812 { 813 struct Qdisc *q = old; 814 struct net *net = dev_net(dev); 815 int err = 0; 816 817 if (parent == NULL) { 818 unsigned int i, num_q, ingress; 819 820 ingress = 0; 821 num_q = dev->num_tx_queues; 822 if ((q && q->flags & TCQ_F_INGRESS) || 823 (new && new->flags & TCQ_F_INGRESS)) { 824 num_q = 1; 825 ingress = 1; 826 if (!dev_ingress_queue(dev)) 827 return -ENOENT; 828 } 829 830 if (dev->flags & IFF_UP) 831 dev_deactivate(dev); 832 833 if (new && new->ops->attach) 834 goto skip; 835 836 for (i = 0; i < num_q; i++) { 837 struct netdev_queue *dev_queue = dev_ingress_queue(dev); 838 839 if (!ingress) 840 dev_queue = netdev_get_tx_queue(dev, i); 841 842 old = dev_graft_qdisc(dev_queue, new); 843 if (new && i > 0) 844 atomic_inc(&new->refcnt); 845 846 if (!ingress) 847 qdisc_destroy(old); 848 } 849 850 skip: 851 if (!ingress) { 852 notify_and_destroy(net, skb, n, classid, 853 dev->qdisc, new); 854 if (new && !new->ops->attach) 855 atomic_inc(&new->refcnt); 856 dev->qdisc = new ? : &noop_qdisc; 857 858 if (new && new->ops->attach) 859 new->ops->attach(new); 860 } else { 861 notify_and_destroy(net, skb, n, classid, old, new); 862 } 863 864 if (dev->flags & IFF_UP) 865 dev_activate(dev); 866 } else { 867 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 868 869 err = -EOPNOTSUPP; 870 if (cops && cops->graft) { 871 unsigned long cl = cops->get(parent, classid); 872 if (cl) { 873 err = cops->graft(parent, cl, new, &old); 874 cops->put(parent, cl); 875 } else 876 err = -ENOENT; 877 } 878 if (!err) 879 notify_and_destroy(net, skb, n, classid, old, new); 880 } 881 return err; 882 } 883 884 /* lockdep annotation is needed for ingress; egress gets it only for name */ 885 static struct lock_class_key qdisc_tx_lock; 886 static struct lock_class_key qdisc_rx_lock; 887 888 /* 889 Allocate and initialize new qdisc. 890 891 Parameters are passed via opt. 892 */ 893 894 static struct Qdisc * 895 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, 896 struct Qdisc *p, u32 parent, u32 handle, 897 struct nlattr **tca, int *errp) 898 { 899 int err; 900 struct nlattr *kind = tca[TCA_KIND]; 901 struct Qdisc *sch; 902 struct Qdisc_ops *ops; 903 struct qdisc_size_table *stab; 904 905 ops = qdisc_lookup_ops(kind); 906 #ifdef CONFIG_MODULES 907 if (ops == NULL && kind != NULL) { 908 char name[IFNAMSIZ]; 909 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 910 /* We dropped the RTNL semaphore in order to 911 * perform the module load. So, even if we 912 * succeeded in loading the module we have to 913 * tell the caller to replay the request. We 914 * indicate this using -EAGAIN. 915 * We replay the request because the device may 916 * go away in the mean time. 917 */ 918 rtnl_unlock(); 919 request_module("sch_%s", name); 920 rtnl_lock(); 921 ops = qdisc_lookup_ops(kind); 922 if (ops != NULL) { 923 /* We will try again qdisc_lookup_ops, 924 * so don't keep a reference. 925 */ 926 module_put(ops->owner); 927 err = -EAGAIN; 928 goto err_out; 929 } 930 } 931 } 932 #endif 933 934 err = -ENOENT; 935 if (ops == NULL) 936 goto err_out; 937 938 sch = qdisc_alloc(dev_queue, ops); 939 if (IS_ERR(sch)) { 940 err = PTR_ERR(sch); 941 goto err_out2; 942 } 943 944 sch->parent = parent; 945 946 if (handle == TC_H_INGRESS) { 947 sch->flags |= TCQ_F_INGRESS; 948 handle = TC_H_MAKE(TC_H_INGRESS, 0); 949 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 950 } else { 951 if (handle == 0) { 952 handle = qdisc_alloc_handle(dev); 953 err = -ENOMEM; 954 if (handle == 0) 955 goto err_out3; 956 } 957 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 958 if (!netif_is_multiqueue(dev)) 959 sch->flags |= TCQ_F_ONETXQUEUE; 960 } 961 962 sch->handle = handle; 963 964 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 965 if (qdisc_is_percpu_stats(sch)) { 966 sch->cpu_bstats = 967 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); 968 if (!sch->cpu_bstats) 969 goto err_out4; 970 971 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); 972 if (!sch->cpu_qstats) 973 goto err_out4; 974 } 975 976 if (tca[TCA_STAB]) { 977 stab = qdisc_get_stab(tca[TCA_STAB]); 978 if (IS_ERR(stab)) { 979 err = PTR_ERR(stab); 980 goto err_out4; 981 } 982 rcu_assign_pointer(sch->stab, stab); 983 } 984 if (tca[TCA_RATE]) { 985 seqcount_t *running; 986 987 err = -EOPNOTSUPP; 988 if (sch->flags & TCQ_F_MQROOT) 989 goto err_out4; 990 991 if ((sch->parent != TC_H_ROOT) && 992 !(sch->flags & TCQ_F_INGRESS) && 993 (!p || !(p->flags & TCQ_F_MQROOT))) 994 running = qdisc_root_sleeping_running(sch); 995 else 996 running = &sch->running; 997 998 err = gen_new_estimator(&sch->bstats, 999 sch->cpu_bstats, 1000 &sch->rate_est, 1001 NULL, 1002 running, 1003 tca[TCA_RATE]); 1004 if (err) 1005 goto err_out4; 1006 } 1007 1008 qdisc_list_add(sch); 1009 1010 return sch; 1011 } 1012 err_out3: 1013 dev_put(dev); 1014 kfree((char *) sch - sch->padded); 1015 err_out2: 1016 module_put(ops->owner); 1017 err_out: 1018 *errp = err; 1019 return NULL; 1020 1021 err_out4: 1022 free_percpu(sch->cpu_bstats); 1023 free_percpu(sch->cpu_qstats); 1024 /* 1025 * Any broken qdiscs that would require a ops->reset() here? 1026 * The qdisc was never in action so it shouldn't be necessary. 1027 */ 1028 qdisc_put_stab(rtnl_dereference(sch->stab)); 1029 if (ops->destroy) 1030 ops->destroy(sch); 1031 goto err_out3; 1032 } 1033 1034 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 1035 { 1036 struct qdisc_size_table *ostab, *stab = NULL; 1037 int err = 0; 1038 1039 if (tca[TCA_OPTIONS]) { 1040 if (sch->ops->change == NULL) 1041 return -EINVAL; 1042 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 1043 if (err) 1044 return err; 1045 } 1046 1047 if (tca[TCA_STAB]) { 1048 stab = qdisc_get_stab(tca[TCA_STAB]); 1049 if (IS_ERR(stab)) 1050 return PTR_ERR(stab); 1051 } 1052 1053 ostab = rtnl_dereference(sch->stab); 1054 rcu_assign_pointer(sch->stab, stab); 1055 qdisc_put_stab(ostab); 1056 1057 if (tca[TCA_RATE]) { 1058 /* NB: ignores errors from replace_estimator 1059 because change can't be undone. */ 1060 if (sch->flags & TCQ_F_MQROOT) 1061 goto out; 1062 gen_replace_estimator(&sch->bstats, 1063 sch->cpu_bstats, 1064 &sch->rate_est, 1065 NULL, 1066 qdisc_root_sleeping_running(sch), 1067 tca[TCA_RATE]); 1068 } 1069 out: 1070 return 0; 1071 } 1072 1073 struct check_loop_arg { 1074 struct qdisc_walker w; 1075 struct Qdisc *p; 1076 int depth; 1077 }; 1078 1079 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); 1080 1081 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1082 { 1083 struct check_loop_arg arg; 1084 1085 if (q->ops->cl_ops == NULL) 1086 return 0; 1087 1088 arg.w.stop = arg.w.skip = arg.w.count = 0; 1089 arg.w.fn = check_loop_fn; 1090 arg.depth = depth; 1091 arg.p = p; 1092 q->ops->cl_ops->walk(q, &arg.w); 1093 return arg.w.stop ? -ELOOP : 0; 1094 } 1095 1096 static int 1097 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1098 { 1099 struct Qdisc *leaf; 1100 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1101 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1102 1103 leaf = cops->leaf(q, cl); 1104 if (leaf) { 1105 if (leaf == arg->p || arg->depth > 7) 1106 return -ELOOP; 1107 return check_loop(leaf, arg->p, arg->depth + 1); 1108 } 1109 return 0; 1110 } 1111 1112 /* 1113 * Delete/get qdisc. 1114 */ 1115 1116 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1117 { 1118 struct net *net = sock_net(skb->sk); 1119 struct tcmsg *tcm = nlmsg_data(n); 1120 struct nlattr *tca[TCA_MAX + 1]; 1121 struct net_device *dev; 1122 u32 clid; 1123 struct Qdisc *q = NULL; 1124 struct Qdisc *p = NULL; 1125 int err; 1126 1127 if ((n->nlmsg_type != RTM_GETQDISC) && 1128 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1129 return -EPERM; 1130 1131 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1132 if (err < 0) 1133 return err; 1134 1135 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1136 if (!dev) 1137 return -ENODEV; 1138 1139 clid = tcm->tcm_parent; 1140 if (clid) { 1141 if (clid != TC_H_ROOT) { 1142 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1143 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1144 if (!p) 1145 return -ENOENT; 1146 q = qdisc_leaf(p, clid); 1147 } else if (dev_ingress_queue(dev)) { 1148 q = dev_ingress_queue(dev)->qdisc_sleeping; 1149 } 1150 } else { 1151 q = dev->qdisc; 1152 } 1153 if (!q) 1154 return -ENOENT; 1155 1156 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 1157 return -EINVAL; 1158 } else { 1159 q = qdisc_lookup(dev, tcm->tcm_handle); 1160 if (!q) 1161 return -ENOENT; 1162 } 1163 1164 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1165 return -EINVAL; 1166 1167 if (n->nlmsg_type == RTM_DELQDISC) { 1168 if (!clid) 1169 return -EINVAL; 1170 if (q->handle == 0) 1171 return -ENOENT; 1172 err = qdisc_graft(dev, p, skb, n, clid, NULL, q); 1173 if (err != 0) 1174 return err; 1175 } else { 1176 qdisc_notify(net, skb, n, clid, NULL, q); 1177 } 1178 return 0; 1179 } 1180 1181 /* 1182 * Create/change qdisc. 1183 */ 1184 1185 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n) 1186 { 1187 struct net *net = sock_net(skb->sk); 1188 struct tcmsg *tcm; 1189 struct nlattr *tca[TCA_MAX + 1]; 1190 struct net_device *dev; 1191 u32 clid; 1192 struct Qdisc *q, *p; 1193 int err; 1194 1195 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1196 return -EPERM; 1197 1198 replay: 1199 /* Reinit, just in case something touches this. */ 1200 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1201 if (err < 0) 1202 return err; 1203 1204 tcm = nlmsg_data(n); 1205 clid = tcm->tcm_parent; 1206 q = p = NULL; 1207 1208 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1209 if (!dev) 1210 return -ENODEV; 1211 1212 1213 if (clid) { 1214 if (clid != TC_H_ROOT) { 1215 if (clid != TC_H_INGRESS) { 1216 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1217 if (!p) 1218 return -ENOENT; 1219 q = qdisc_leaf(p, clid); 1220 } else if (dev_ingress_queue_create(dev)) { 1221 q = dev_ingress_queue(dev)->qdisc_sleeping; 1222 } 1223 } else { 1224 q = dev->qdisc; 1225 } 1226 1227 /* It may be default qdisc, ignore it */ 1228 if (q && q->handle == 0) 1229 q = NULL; 1230 1231 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1232 if (tcm->tcm_handle) { 1233 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) 1234 return -EEXIST; 1235 if (TC_H_MIN(tcm->tcm_handle)) 1236 return -EINVAL; 1237 q = qdisc_lookup(dev, tcm->tcm_handle); 1238 if (!q) 1239 goto create_n_graft; 1240 if (n->nlmsg_flags & NLM_F_EXCL) 1241 return -EEXIST; 1242 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1243 return -EINVAL; 1244 if (q == p || 1245 (p && check_loop(q, p, 0))) 1246 return -ELOOP; 1247 atomic_inc(&q->refcnt); 1248 goto graft; 1249 } else { 1250 if (!q) 1251 goto create_n_graft; 1252 1253 /* This magic test requires explanation. 1254 * 1255 * We know, that some child q is already 1256 * attached to this parent and have choice: 1257 * either to change it or to create/graft new one. 1258 * 1259 * 1. We are allowed to create/graft only 1260 * if CREATE and REPLACE flags are set. 1261 * 1262 * 2. If EXCL is set, requestor wanted to say, 1263 * that qdisc tcm_handle is not expected 1264 * to exist, so that we choose create/graft too. 1265 * 1266 * 3. The last case is when no flags are set. 1267 * Alas, it is sort of hole in API, we 1268 * cannot decide what to do unambiguously. 1269 * For now we select create/graft, if 1270 * user gave KIND, which does not match existing. 1271 */ 1272 if ((n->nlmsg_flags & NLM_F_CREATE) && 1273 (n->nlmsg_flags & NLM_F_REPLACE) && 1274 ((n->nlmsg_flags & NLM_F_EXCL) || 1275 (tca[TCA_KIND] && 1276 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1277 goto create_n_graft; 1278 } 1279 } 1280 } else { 1281 if (!tcm->tcm_handle) 1282 return -EINVAL; 1283 q = qdisc_lookup(dev, tcm->tcm_handle); 1284 } 1285 1286 /* Change qdisc parameters */ 1287 if (q == NULL) 1288 return -ENOENT; 1289 if (n->nlmsg_flags & NLM_F_EXCL) 1290 return -EEXIST; 1291 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1292 return -EINVAL; 1293 err = qdisc_change(q, tca); 1294 if (err == 0) 1295 qdisc_notify(net, skb, n, clid, NULL, q); 1296 return err; 1297 1298 create_n_graft: 1299 if (!(n->nlmsg_flags & NLM_F_CREATE)) 1300 return -ENOENT; 1301 if (clid == TC_H_INGRESS) { 1302 if (dev_ingress_queue(dev)) 1303 q = qdisc_create(dev, dev_ingress_queue(dev), p, 1304 tcm->tcm_parent, tcm->tcm_parent, 1305 tca, &err); 1306 else 1307 err = -ENOENT; 1308 } else { 1309 struct netdev_queue *dev_queue; 1310 1311 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1312 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1313 else if (p) 1314 dev_queue = p->dev_queue; 1315 else 1316 dev_queue = netdev_get_tx_queue(dev, 0); 1317 1318 q = qdisc_create(dev, dev_queue, p, 1319 tcm->tcm_parent, tcm->tcm_handle, 1320 tca, &err); 1321 } 1322 if (q == NULL) { 1323 if (err == -EAGAIN) 1324 goto replay; 1325 return err; 1326 } 1327 1328 graft: 1329 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1330 if (err) { 1331 if (q) 1332 qdisc_destroy(q); 1333 return err; 1334 } 1335 1336 return 0; 1337 } 1338 1339 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1340 u32 portid, u32 seq, u16 flags, int event) 1341 { 1342 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL; 1343 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 1344 struct tcmsg *tcm; 1345 struct nlmsghdr *nlh; 1346 unsigned char *b = skb_tail_pointer(skb); 1347 struct gnet_dump d; 1348 struct qdisc_size_table *stab; 1349 __u32 qlen; 1350 1351 cond_resched(); 1352 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1353 if (!nlh) 1354 goto out_nlmsg_trim; 1355 tcm = nlmsg_data(nlh); 1356 tcm->tcm_family = AF_UNSPEC; 1357 tcm->tcm__pad1 = 0; 1358 tcm->tcm__pad2 = 0; 1359 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1360 tcm->tcm_parent = clid; 1361 tcm->tcm_handle = q->handle; 1362 tcm->tcm_info = atomic_read(&q->refcnt); 1363 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1364 goto nla_put_failure; 1365 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1366 goto nla_put_failure; 1367 qlen = q->q.qlen; 1368 1369 stab = rtnl_dereference(q->stab); 1370 if (stab && qdisc_dump_stab(skb, stab) < 0) 1371 goto nla_put_failure; 1372 1373 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1374 NULL, &d, TCA_PAD) < 0) 1375 goto nla_put_failure; 1376 1377 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1378 goto nla_put_failure; 1379 1380 if (qdisc_is_percpu_stats(q)) { 1381 cpu_bstats = q->cpu_bstats; 1382 cpu_qstats = q->cpu_qstats; 1383 } 1384 1385 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q), 1386 &d, cpu_bstats, &q->bstats) < 0 || 1387 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 || 1388 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 1389 goto nla_put_failure; 1390 1391 if (gnet_stats_finish_copy(&d) < 0) 1392 goto nla_put_failure; 1393 1394 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1395 return skb->len; 1396 1397 out_nlmsg_trim: 1398 nla_put_failure: 1399 nlmsg_trim(skb, b); 1400 return -1; 1401 } 1402 1403 static bool tc_qdisc_dump_ignore(struct Qdisc *q) 1404 { 1405 return (q->flags & TCQ_F_BUILTIN) ? true : false; 1406 } 1407 1408 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1409 struct nlmsghdr *n, u32 clid, 1410 struct Qdisc *old, struct Qdisc *new) 1411 { 1412 struct sk_buff *skb; 1413 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1414 1415 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1416 if (!skb) 1417 return -ENOBUFS; 1418 1419 if (old && !tc_qdisc_dump_ignore(old)) { 1420 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1421 0, RTM_DELQDISC) < 0) 1422 goto err_out; 1423 } 1424 if (new && !tc_qdisc_dump_ignore(new)) { 1425 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1426 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1427 goto err_out; 1428 } 1429 1430 if (skb->len) 1431 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1432 n->nlmsg_flags & NLM_F_ECHO); 1433 1434 err_out: 1435 kfree_skb(skb); 1436 return -EINVAL; 1437 } 1438 1439 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1440 struct netlink_callback *cb, 1441 int *q_idx_p, int s_q_idx) 1442 { 1443 int ret = 0, q_idx = *q_idx_p; 1444 struct Qdisc *q; 1445 1446 if (!root) 1447 return 0; 1448 1449 q = root; 1450 if (q_idx < s_q_idx) { 1451 q_idx++; 1452 } else { 1453 if (!tc_qdisc_dump_ignore(q) && 1454 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1455 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1456 goto done; 1457 q_idx++; 1458 } 1459 list_for_each_entry(q, &root->list, list) { 1460 if (q_idx < s_q_idx) { 1461 q_idx++; 1462 continue; 1463 } 1464 if (!tc_qdisc_dump_ignore(q) && 1465 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1466 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1467 goto done; 1468 q_idx++; 1469 } 1470 1471 out: 1472 *q_idx_p = q_idx; 1473 return ret; 1474 done: 1475 ret = -1; 1476 goto out; 1477 } 1478 1479 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1480 { 1481 struct net *net = sock_net(skb->sk); 1482 int idx, q_idx; 1483 int s_idx, s_q_idx; 1484 struct net_device *dev; 1485 1486 s_idx = cb->args[0]; 1487 s_q_idx = q_idx = cb->args[1]; 1488 1489 idx = 0; 1490 ASSERT_RTNL(); 1491 for_each_netdev(net, dev) { 1492 struct netdev_queue *dev_queue; 1493 1494 if (idx < s_idx) 1495 goto cont; 1496 if (idx > s_idx) 1497 s_q_idx = 0; 1498 q_idx = 0; 1499 1500 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0) 1501 goto done; 1502 1503 dev_queue = dev_ingress_queue(dev); 1504 if (dev_queue && 1505 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, 1506 &q_idx, s_q_idx) < 0) 1507 goto done; 1508 1509 cont: 1510 idx++; 1511 } 1512 1513 done: 1514 cb->args[0] = idx; 1515 cb->args[1] = q_idx; 1516 1517 return skb->len; 1518 } 1519 1520 1521 1522 /************************************************ 1523 * Traffic classes manipulation. * 1524 ************************************************/ 1525 1526 1527 1528 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n) 1529 { 1530 struct net *net = sock_net(skb->sk); 1531 struct tcmsg *tcm = nlmsg_data(n); 1532 struct nlattr *tca[TCA_MAX + 1]; 1533 struct net_device *dev; 1534 struct Qdisc *q = NULL; 1535 const struct Qdisc_class_ops *cops; 1536 unsigned long cl = 0; 1537 unsigned long new_cl; 1538 u32 portid; 1539 u32 clid; 1540 u32 qid; 1541 int err; 1542 1543 if ((n->nlmsg_type != RTM_GETTCLASS) && 1544 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) 1545 return -EPERM; 1546 1547 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1548 if (err < 0) 1549 return err; 1550 1551 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1552 if (!dev) 1553 return -ENODEV; 1554 1555 /* 1556 parent == TC_H_UNSPEC - unspecified parent. 1557 parent == TC_H_ROOT - class is root, which has no parent. 1558 parent == X:0 - parent is root class. 1559 parent == X:Y - parent is a node in hierarchy. 1560 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1561 1562 handle == 0:0 - generate handle from kernel pool. 1563 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1564 handle == X:Y - clear. 1565 handle == X:0 - root class. 1566 */ 1567 1568 /* Step 1. Determine qdisc handle X:0 */ 1569 1570 portid = tcm->tcm_parent; 1571 clid = tcm->tcm_handle; 1572 qid = TC_H_MAJ(clid); 1573 1574 if (portid != TC_H_ROOT) { 1575 u32 qid1 = TC_H_MAJ(portid); 1576 1577 if (qid && qid1) { 1578 /* If both majors are known, they must be identical. */ 1579 if (qid != qid1) 1580 return -EINVAL; 1581 } else if (qid1) { 1582 qid = qid1; 1583 } else if (qid == 0) 1584 qid = dev->qdisc->handle; 1585 1586 /* Now qid is genuine qdisc handle consistent 1587 * both with parent and child. 1588 * 1589 * TC_H_MAJ(portid) still may be unspecified, complete it now. 1590 */ 1591 if (portid) 1592 portid = TC_H_MAKE(qid, portid); 1593 } else { 1594 if (qid == 0) 1595 qid = dev->qdisc->handle; 1596 } 1597 1598 /* OK. Locate qdisc */ 1599 q = qdisc_lookup(dev, qid); 1600 if (!q) 1601 return -ENOENT; 1602 1603 /* An check that it supports classes */ 1604 cops = q->ops->cl_ops; 1605 if (cops == NULL) 1606 return -EINVAL; 1607 1608 /* Now try to get class */ 1609 if (clid == 0) { 1610 if (portid == TC_H_ROOT) 1611 clid = qid; 1612 } else 1613 clid = TC_H_MAKE(qid, clid); 1614 1615 if (clid) 1616 cl = cops->get(q, clid); 1617 1618 if (cl == 0) { 1619 err = -ENOENT; 1620 if (n->nlmsg_type != RTM_NEWTCLASS || 1621 !(n->nlmsg_flags & NLM_F_CREATE)) 1622 goto out; 1623 } else { 1624 switch (n->nlmsg_type) { 1625 case RTM_NEWTCLASS: 1626 err = -EEXIST; 1627 if (n->nlmsg_flags & NLM_F_EXCL) 1628 goto out; 1629 break; 1630 case RTM_DELTCLASS: 1631 err = -EOPNOTSUPP; 1632 if (cops->delete) 1633 err = cops->delete(q, cl); 1634 if (err == 0) 1635 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS); 1636 goto out; 1637 case RTM_GETTCLASS: 1638 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1639 goto out; 1640 default: 1641 err = -EINVAL; 1642 goto out; 1643 } 1644 } 1645 1646 new_cl = cl; 1647 err = -EOPNOTSUPP; 1648 if (cops->change) 1649 err = cops->change(q, clid, portid, tca, &new_cl); 1650 if (err == 0) 1651 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1652 1653 out: 1654 if (cl) 1655 cops->put(q, cl); 1656 1657 return err; 1658 } 1659 1660 1661 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1662 unsigned long cl, 1663 u32 portid, u32 seq, u16 flags, int event) 1664 { 1665 struct tcmsg *tcm; 1666 struct nlmsghdr *nlh; 1667 unsigned char *b = skb_tail_pointer(skb); 1668 struct gnet_dump d; 1669 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1670 1671 cond_resched(); 1672 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1673 if (!nlh) 1674 goto out_nlmsg_trim; 1675 tcm = nlmsg_data(nlh); 1676 tcm->tcm_family = AF_UNSPEC; 1677 tcm->tcm__pad1 = 0; 1678 tcm->tcm__pad2 = 0; 1679 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1680 tcm->tcm_parent = q->handle; 1681 tcm->tcm_handle = q->handle; 1682 tcm->tcm_info = 0; 1683 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1684 goto nla_put_failure; 1685 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1686 goto nla_put_failure; 1687 1688 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1689 NULL, &d, TCA_PAD) < 0) 1690 goto nla_put_failure; 1691 1692 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1693 goto nla_put_failure; 1694 1695 if (gnet_stats_finish_copy(&d) < 0) 1696 goto nla_put_failure; 1697 1698 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1699 return skb->len; 1700 1701 out_nlmsg_trim: 1702 nla_put_failure: 1703 nlmsg_trim(skb, b); 1704 return -1; 1705 } 1706 1707 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1708 struct nlmsghdr *n, struct Qdisc *q, 1709 unsigned long cl, int event) 1710 { 1711 struct sk_buff *skb; 1712 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1713 1714 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1715 if (!skb) 1716 return -ENOBUFS; 1717 1718 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) { 1719 kfree_skb(skb); 1720 return -EINVAL; 1721 } 1722 1723 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1724 n->nlmsg_flags & NLM_F_ECHO); 1725 } 1726 1727 struct qdisc_dump_args { 1728 struct qdisc_walker w; 1729 struct sk_buff *skb; 1730 struct netlink_callback *cb; 1731 }; 1732 1733 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) 1734 { 1735 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1736 1737 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 1738 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); 1739 } 1740 1741 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1742 struct tcmsg *tcm, struct netlink_callback *cb, 1743 int *t_p, int s_t) 1744 { 1745 struct qdisc_dump_args arg; 1746 1747 if (tc_qdisc_dump_ignore(q) || 1748 *t_p < s_t || !q->ops->cl_ops || 1749 (tcm->tcm_parent && 1750 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1751 (*t_p)++; 1752 return 0; 1753 } 1754 if (*t_p > s_t) 1755 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1756 arg.w.fn = qdisc_class_dump; 1757 arg.skb = skb; 1758 arg.cb = cb; 1759 arg.w.stop = 0; 1760 arg.w.skip = cb->args[1]; 1761 arg.w.count = 0; 1762 q->ops->cl_ops->walk(q, &arg.w); 1763 cb->args[1] = arg.w.count; 1764 if (arg.w.stop) 1765 return -1; 1766 (*t_p)++; 1767 return 0; 1768 } 1769 1770 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1771 struct tcmsg *tcm, struct netlink_callback *cb, 1772 int *t_p, int s_t) 1773 { 1774 struct Qdisc *q; 1775 1776 if (!root) 1777 return 0; 1778 1779 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1780 return -1; 1781 1782 list_for_each_entry(q, &root->list, list) { 1783 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1784 return -1; 1785 } 1786 1787 return 0; 1788 } 1789 1790 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1791 { 1792 struct tcmsg *tcm = nlmsg_data(cb->nlh); 1793 struct net *net = sock_net(skb->sk); 1794 struct netdev_queue *dev_queue; 1795 struct net_device *dev; 1796 int t, s_t; 1797 1798 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 1799 return 0; 1800 dev = dev_get_by_index(net, tcm->tcm_ifindex); 1801 if (!dev) 1802 return 0; 1803 1804 s_t = cb->args[0]; 1805 t = 0; 1806 1807 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1808 goto done; 1809 1810 dev_queue = dev_ingress_queue(dev); 1811 if (dev_queue && 1812 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, 1813 &t, s_t) < 0) 1814 goto done; 1815 1816 done: 1817 cb->args[0] = t; 1818 1819 dev_put(dev); 1820 return skb->len; 1821 } 1822 1823 /* Main classifier routine: scans classifier chain attached 1824 * to this qdisc, (optionally) tests for protocol and asks 1825 * specific classifiers. 1826 */ 1827 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp, 1828 struct tcf_result *res, bool compat_mode) 1829 { 1830 __be16 protocol = tc_skb_protocol(skb); 1831 #ifdef CONFIG_NET_CLS_ACT 1832 const struct tcf_proto *old_tp = tp; 1833 int limit = 0; 1834 1835 reclassify: 1836 #endif 1837 for (; tp; tp = rcu_dereference_bh(tp->next)) { 1838 int err; 1839 1840 if (tp->protocol != protocol && 1841 tp->protocol != htons(ETH_P_ALL)) 1842 continue; 1843 1844 err = tp->classify(skb, tp, res); 1845 #ifdef CONFIG_NET_CLS_ACT 1846 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) 1847 goto reset; 1848 #endif 1849 if (err >= 0) 1850 return err; 1851 } 1852 1853 return TC_ACT_UNSPEC; /* signal: continue lookup */ 1854 #ifdef CONFIG_NET_CLS_ACT 1855 reset: 1856 if (unlikely(limit++ >= MAX_REC_LOOP)) { 1857 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n", 1858 tp->q->ops->id, tp->prio & 0xffff, 1859 ntohs(tp->protocol)); 1860 return TC_ACT_SHOT; 1861 } 1862 1863 tp = old_tp; 1864 protocol = tc_skb_protocol(skb); 1865 goto reclassify; 1866 #endif 1867 } 1868 EXPORT_SYMBOL(tc_classify); 1869 1870 bool tcf_destroy(struct tcf_proto *tp, bool force) 1871 { 1872 if (tp->ops->destroy(tp, force)) { 1873 module_put(tp->ops->owner); 1874 kfree_rcu(tp, rcu); 1875 return true; 1876 } 1877 1878 return false; 1879 } 1880 1881 void tcf_destroy_chain(struct tcf_proto __rcu **fl) 1882 { 1883 struct tcf_proto *tp; 1884 1885 while ((tp = rtnl_dereference(*fl)) != NULL) { 1886 RCU_INIT_POINTER(*fl, tp->next); 1887 tcf_destroy(tp, true); 1888 } 1889 } 1890 EXPORT_SYMBOL(tcf_destroy_chain); 1891 1892 #ifdef CONFIG_PROC_FS 1893 static int psched_show(struct seq_file *seq, void *v) 1894 { 1895 seq_printf(seq, "%08x %08x %08x %08x\n", 1896 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1897 1000000, 1898 (u32)NSEC_PER_SEC / hrtimer_resolution); 1899 1900 return 0; 1901 } 1902 1903 static int psched_open(struct inode *inode, struct file *file) 1904 { 1905 return single_open(file, psched_show, NULL); 1906 } 1907 1908 static const struct file_operations psched_fops = { 1909 .owner = THIS_MODULE, 1910 .open = psched_open, 1911 .read = seq_read, 1912 .llseek = seq_lseek, 1913 .release = single_release, 1914 }; 1915 1916 static int __net_init psched_net_init(struct net *net) 1917 { 1918 struct proc_dir_entry *e; 1919 1920 e = proc_create("psched", 0, net->proc_net, &psched_fops); 1921 if (e == NULL) 1922 return -ENOMEM; 1923 1924 return 0; 1925 } 1926 1927 static void __net_exit psched_net_exit(struct net *net) 1928 { 1929 remove_proc_entry("psched", net->proc_net); 1930 } 1931 #else 1932 static int __net_init psched_net_init(struct net *net) 1933 { 1934 return 0; 1935 } 1936 1937 static void __net_exit psched_net_exit(struct net *net) 1938 { 1939 } 1940 #endif 1941 1942 static struct pernet_operations psched_net_ops = { 1943 .init = psched_net_init, 1944 .exit = psched_net_exit, 1945 }; 1946 1947 static int __init pktsched_init(void) 1948 { 1949 int err; 1950 1951 err = register_pernet_subsys(&psched_net_ops); 1952 if (err) { 1953 pr_err("pktsched_init: " 1954 "cannot initialize per netns operations\n"); 1955 return err; 1956 } 1957 1958 register_qdisc(&pfifo_fast_ops); 1959 register_qdisc(&pfifo_qdisc_ops); 1960 register_qdisc(&bfifo_qdisc_ops); 1961 register_qdisc(&pfifo_head_drop_qdisc_ops); 1962 register_qdisc(&mq_qdisc_ops); 1963 register_qdisc(&noqueue_qdisc_ops); 1964 1965 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL); 1966 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL); 1967 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL); 1968 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL); 1969 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL); 1970 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL); 1971 1972 return 0; 1973 } 1974 1975 subsys_initcall(pktsched_init); 1976