1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/sch_api.c Packet scheduler API. 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * 7 * Fixes: 8 * 9 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. 10 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 11 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support 12 */ 13 14 #include <linux/module.h> 15 #include <linux/types.h> 16 #include <linux/kernel.h> 17 #include <linux/string.h> 18 #include <linux/errno.h> 19 #include <linux/skbuff.h> 20 #include <linux/init.h> 21 #include <linux/proc_fs.h> 22 #include <linux/seq_file.h> 23 #include <linux/kmod.h> 24 #include <linux/list.h> 25 #include <linux/hrtimer.h> 26 #include <linux/slab.h> 27 #include <linux/hashtable.h> 28 29 #include <net/net_namespace.h> 30 #include <net/sock.h> 31 #include <net/netlink.h> 32 #include <net/pkt_sched.h> 33 #include <net/pkt_cls.h> 34 #include <net/tc_wrapper.h> 35 36 #include <trace/events/qdisc.h> 37 38 /* 39 40 Short review. 41 ------------- 42 43 This file consists of two interrelated parts: 44 45 1. queueing disciplines manager frontend. 46 2. traffic classes manager frontend. 47 48 Generally, queueing discipline ("qdisc") is a black box, 49 which is able to enqueue packets and to dequeue them (when 50 device is ready to send something) in order and at times 51 determined by algorithm hidden in it. 52 53 qdisc's are divided to two categories: 54 - "queues", which have no internal structure visible from outside. 55 - "schedulers", which split all the packets to "traffic classes", 56 using "packet classifiers" (look at cls_api.c) 57 58 In turn, classes may have child qdiscs (as rule, queues) 59 attached to them etc. etc. etc. 60 61 The goal of the routines in this file is to translate 62 information supplied by user in the form of handles 63 to more intelligible for kernel form, to make some sanity 64 checks and part of work, which is common to all qdiscs 65 and to provide rtnetlink notifications. 66 67 All real intelligent work is done inside qdisc modules. 68 69 70 71 Every discipline has two major routines: enqueue and dequeue. 72 73 ---dequeue 74 75 dequeue usually returns a skb to send. It is allowed to return NULL, 76 but it does not mean that queue is empty, it just means that 77 discipline does not want to send anything this time. 78 Queue is really empty if q->q.qlen == 0. 79 For complicated disciplines with multiple queues q->q is not 80 real packet queue, but however q->q.qlen must be valid. 81 82 ---enqueue 83 84 enqueue returns 0, if packet was enqueued successfully. 85 If packet (this one or another one) was dropped, it returns 86 not zero error code. 87 NET_XMIT_DROP - this packet dropped 88 Expected action: do not backoff, but wait until queue will clear. 89 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 90 Expected action: backoff or ignore 91 92 Auxiliary routines: 93 94 ---peek 95 96 like dequeue but without removing a packet from the queue 97 98 ---reset 99 100 returns qdisc to initial state: purge all buffers, clear all 101 timers, counters (except for statistics) etc. 102 103 ---init 104 105 initializes newly created qdisc. 106 107 ---destroy 108 109 destroys resources allocated by init and during lifetime of qdisc. 110 111 ---change 112 113 changes qdisc parameters. 114 */ 115 116 /* Protects list of registered TC modules. It is pure SMP lock. */ 117 static DEFINE_RWLOCK(qdisc_mod_lock); 118 119 120 /************************************************ 121 * Queueing disciplines manipulation. * 122 ************************************************/ 123 124 125 /* The list of all installed queueing disciplines. */ 126 127 static struct Qdisc_ops *qdisc_base; 128 129 /* Register/unregister queueing discipline */ 130 131 int register_qdisc(struct Qdisc_ops *qops) 132 { 133 struct Qdisc_ops *q, **qp; 134 int rc = -EEXIST; 135 136 write_lock(&qdisc_mod_lock); 137 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 138 if (!strcmp(qops->id, q->id)) 139 goto out; 140 141 if (qops->enqueue == NULL) 142 qops->enqueue = noop_qdisc_ops.enqueue; 143 if (qops->peek == NULL) { 144 if (qops->dequeue == NULL) 145 qops->peek = noop_qdisc_ops.peek; 146 else 147 goto out_einval; 148 } 149 if (qops->dequeue == NULL) 150 qops->dequeue = noop_qdisc_ops.dequeue; 151 152 if (qops->cl_ops) { 153 const struct Qdisc_class_ops *cops = qops->cl_ops; 154 155 if (!(cops->find && cops->walk && cops->leaf)) 156 goto out_einval; 157 158 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf)) 159 goto out_einval; 160 } 161 162 qops->next = NULL; 163 *qp = qops; 164 rc = 0; 165 out: 166 write_unlock(&qdisc_mod_lock); 167 return rc; 168 169 out_einval: 170 rc = -EINVAL; 171 goto out; 172 } 173 EXPORT_SYMBOL(register_qdisc); 174 175 void unregister_qdisc(struct Qdisc_ops *qops) 176 { 177 struct Qdisc_ops *q, **qp; 178 int err = -ENOENT; 179 180 write_lock(&qdisc_mod_lock); 181 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 182 if (q == qops) 183 break; 184 if (q) { 185 *qp = q->next; 186 q->next = NULL; 187 err = 0; 188 } 189 write_unlock(&qdisc_mod_lock); 190 191 WARN(err, "unregister qdisc(%s) failed\n", qops->id); 192 } 193 EXPORT_SYMBOL(unregister_qdisc); 194 195 /* Get default qdisc if not otherwise specified */ 196 void qdisc_get_default(char *name, size_t len) 197 { 198 read_lock(&qdisc_mod_lock); 199 strscpy(name, default_qdisc_ops->id, len); 200 read_unlock(&qdisc_mod_lock); 201 } 202 203 static struct Qdisc_ops *qdisc_lookup_default(const char *name) 204 { 205 struct Qdisc_ops *q = NULL; 206 207 for (q = qdisc_base; q; q = q->next) { 208 if (!strcmp(name, q->id)) { 209 if (!try_module_get(q->owner)) 210 q = NULL; 211 break; 212 } 213 } 214 215 return q; 216 } 217 218 /* Set new default qdisc to use */ 219 int qdisc_set_default(const char *name) 220 { 221 const struct Qdisc_ops *ops; 222 223 if (!capable(CAP_NET_ADMIN)) 224 return -EPERM; 225 226 write_lock(&qdisc_mod_lock); 227 ops = qdisc_lookup_default(name); 228 if (!ops) { 229 /* Not found, drop lock and try to load module */ 230 write_unlock(&qdisc_mod_lock); 231 request_module(NET_SCH_ALIAS_PREFIX "%s", name); 232 write_lock(&qdisc_mod_lock); 233 234 ops = qdisc_lookup_default(name); 235 } 236 237 if (ops) { 238 /* Set new default */ 239 module_put(default_qdisc_ops->owner); 240 default_qdisc_ops = ops; 241 } 242 write_unlock(&qdisc_mod_lock); 243 244 return ops ? 0 : -ENOENT; 245 } 246 247 #ifdef CONFIG_NET_SCH_DEFAULT 248 /* Set default value from kernel config */ 249 static int __init sch_default_qdisc(void) 250 { 251 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH); 252 } 253 late_initcall(sch_default_qdisc); 254 #endif 255 256 /* We know handle. Find qdisc among all qdisc's attached to device 257 * (root qdisc, all its children, children of children etc.) 258 * Note: caller either uses rtnl or rcu_read_lock() 259 */ 260 261 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 262 { 263 struct Qdisc *q; 264 265 if (!qdisc_dev(root)) 266 return (root->handle == handle ? root : NULL); 267 268 if (!(root->flags & TCQ_F_BUILTIN) && 269 root->handle == handle) 270 return root; 271 272 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle, 273 lockdep_rtnl_is_held()) { 274 if (q->handle == handle) 275 return q; 276 } 277 return NULL; 278 } 279 280 void qdisc_hash_add(struct Qdisc *q, bool invisible) 281 { 282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 283 ASSERT_RTNL(); 284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle); 285 if (invisible) 286 q->flags |= TCQ_F_INVISIBLE; 287 } 288 } 289 EXPORT_SYMBOL(qdisc_hash_add); 290 291 void qdisc_hash_del(struct Qdisc *q) 292 { 293 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) { 294 ASSERT_RTNL(); 295 hash_del_rcu(&q->hash); 296 } 297 } 298 EXPORT_SYMBOL(qdisc_hash_del); 299 300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 301 { 302 struct Qdisc *q; 303 304 if (!handle) 305 return NULL; 306 q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle); 307 if (q) 308 goto out; 309 310 if (dev_ingress_queue(dev)) 311 q = qdisc_match_from_root( 312 rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping), 313 handle); 314 out: 315 return q; 316 } 317 318 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle) 319 { 320 struct netdev_queue *nq; 321 struct Qdisc *q; 322 323 if (!handle) 324 return NULL; 325 q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle); 326 if (q) 327 goto out; 328 329 nq = dev_ingress_queue_rcu(dev); 330 if (nq) 331 q = qdisc_match_from_root(rcu_dereference(nq->qdisc_sleeping), 332 handle); 333 out: 334 return q; 335 } 336 337 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 338 { 339 unsigned long cl; 340 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 341 342 if (cops == NULL) 343 return NULL; 344 cl = cops->find(p, classid); 345 346 if (cl == 0) 347 return NULL; 348 return cops->leaf(p, cl); 349 } 350 351 /* Find queueing discipline by name */ 352 353 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 354 { 355 struct Qdisc_ops *q = NULL; 356 357 if (kind) { 358 read_lock(&qdisc_mod_lock); 359 for (q = qdisc_base; q; q = q->next) { 360 if (nla_strcmp(kind, q->id) == 0) { 361 if (!try_module_get(q->owner)) 362 q = NULL; 363 break; 364 } 365 } 366 read_unlock(&qdisc_mod_lock); 367 } 368 return q; 369 } 370 371 /* The linklayer setting were not transferred from iproute2, in older 372 * versions, and the rate tables lookup systems have been dropped in 373 * the kernel. To keep backward compatible with older iproute2 tc 374 * utils, we detect the linklayer setting by detecting if the rate 375 * table were modified. 376 * 377 * For linklayer ATM table entries, the rate table will be aligned to 378 * 48 bytes, thus some table entries will contain the same value. The 379 * mpu (min packet unit) is also encoded into the old rate table, thus 380 * starting from the mpu, we find low and high table entries for 381 * mapping this cell. If these entries contain the same value, when 382 * the rate tables have been modified for linklayer ATM. 383 * 384 * This is done by rounding mpu to the nearest 48 bytes cell/entry, 385 * and then roundup to the next cell, calc the table entry one below, 386 * and compare. 387 */ 388 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab) 389 { 390 int low = roundup(r->mpu, 48); 391 int high = roundup(low+1, 48); 392 int cell_low = low >> r->cell_log; 393 int cell_high = (high >> r->cell_log) - 1; 394 395 /* rtab is too inaccurate at rates > 100Mbit/s */ 396 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) { 397 pr_debug("TC linklayer: Giving up ATM detection\n"); 398 return TC_LINKLAYER_ETHERNET; 399 } 400 401 if ((cell_high > cell_low) && (cell_high < 256) 402 && (rtab[cell_low] == rtab[cell_high])) { 403 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n", 404 cell_low, cell_high, rtab[cell_high]); 405 return TC_LINKLAYER_ATM; 406 } 407 return TC_LINKLAYER_ETHERNET; 408 } 409 410 static struct qdisc_rate_table *qdisc_rtab_list; 411 412 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, 413 struct nlattr *tab, 414 struct netlink_ext_ack *extack) 415 { 416 struct qdisc_rate_table *rtab; 417 418 if (tab == NULL || r->rate == 0 || 419 r->cell_log == 0 || r->cell_log >= 32 || 420 nla_len(tab) != TC_RTAB_SIZE) { 421 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching"); 422 return NULL; 423 } 424 425 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 426 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) && 427 !memcmp(&rtab->data, nla_data(tab), 1024)) { 428 rtab->refcnt++; 429 return rtab; 430 } 431 } 432 433 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 434 if (rtab) { 435 rtab->rate = *r; 436 rtab->refcnt = 1; 437 memcpy(rtab->data, nla_data(tab), 1024); 438 if (r->linklayer == TC_LINKLAYER_UNAWARE) 439 r->linklayer = __detect_linklayer(r, rtab->data); 440 rtab->next = qdisc_rtab_list; 441 qdisc_rtab_list = rtab; 442 } else { 443 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table"); 444 } 445 return rtab; 446 } 447 EXPORT_SYMBOL(qdisc_get_rtab); 448 449 void qdisc_put_rtab(struct qdisc_rate_table *tab) 450 { 451 struct qdisc_rate_table *rtab, **rtabp; 452 453 if (!tab || --tab->refcnt) 454 return; 455 456 for (rtabp = &qdisc_rtab_list; 457 (rtab = *rtabp) != NULL; 458 rtabp = &rtab->next) { 459 if (rtab == tab) { 460 *rtabp = rtab->next; 461 kfree(rtab); 462 return; 463 } 464 } 465 } 466 EXPORT_SYMBOL(qdisc_put_rtab); 467 468 static LIST_HEAD(qdisc_stab_list); 469 470 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 471 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 472 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 473 }; 474 475 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt, 476 struct netlink_ext_ack *extack) 477 { 478 struct nlattr *tb[TCA_STAB_MAX + 1]; 479 struct qdisc_size_table *stab; 480 struct tc_sizespec *s; 481 unsigned int tsize = 0; 482 u16 *tab = NULL; 483 int err; 484 485 err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy, 486 extack); 487 if (err < 0) 488 return ERR_PTR(err); 489 if (!tb[TCA_STAB_BASE]) { 490 NL_SET_ERR_MSG(extack, "Size table base attribute is missing"); 491 return ERR_PTR(-EINVAL); 492 } 493 494 s = nla_data(tb[TCA_STAB_BASE]); 495 496 if (s->tsize > 0) { 497 if (!tb[TCA_STAB_DATA]) { 498 NL_SET_ERR_MSG(extack, "Size table data attribute is missing"); 499 return ERR_PTR(-EINVAL); 500 } 501 tab = nla_data(tb[TCA_STAB_DATA]); 502 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 503 } 504 505 if (tsize != s->tsize || (!tab && tsize > 0)) { 506 NL_SET_ERR_MSG(extack, "Invalid size of size table"); 507 return ERR_PTR(-EINVAL); 508 } 509 510 list_for_each_entry(stab, &qdisc_stab_list, list) { 511 if (memcmp(&stab->szopts, s, sizeof(*s))) 512 continue; 513 if (tsize > 0 && 514 memcmp(stab->data, tab, flex_array_size(stab, data, tsize))) 515 continue; 516 stab->refcnt++; 517 return stab; 518 } 519 520 if (s->size_log > STAB_SIZE_LOG_MAX || 521 s->cell_log > STAB_SIZE_LOG_MAX) { 522 NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table"); 523 return ERR_PTR(-EINVAL); 524 } 525 526 stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL); 527 if (!stab) 528 return ERR_PTR(-ENOMEM); 529 530 stab->refcnt = 1; 531 stab->szopts = *s; 532 if (tsize > 0) 533 memcpy(stab->data, tab, flex_array_size(stab, data, tsize)); 534 535 list_add_tail(&stab->list, &qdisc_stab_list); 536 537 return stab; 538 } 539 540 void qdisc_put_stab(struct qdisc_size_table *tab) 541 { 542 if (!tab) 543 return; 544 545 if (--tab->refcnt == 0) { 546 list_del(&tab->list); 547 kfree_rcu(tab, rcu); 548 } 549 } 550 EXPORT_SYMBOL(qdisc_put_stab); 551 552 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 553 { 554 struct nlattr *nest; 555 556 nest = nla_nest_start_noflag(skb, TCA_STAB); 557 if (nest == NULL) 558 goto nla_put_failure; 559 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts)) 560 goto nla_put_failure; 561 nla_nest_end(skb, nest); 562 563 return skb->len; 564 565 nla_put_failure: 566 return -1; 567 } 568 569 void __qdisc_calculate_pkt_len(struct sk_buff *skb, 570 const struct qdisc_size_table *stab) 571 { 572 int pkt_len, slot; 573 574 pkt_len = skb->len + stab->szopts.overhead; 575 if (unlikely(!stab->szopts.tsize)) 576 goto out; 577 578 slot = pkt_len + stab->szopts.cell_align; 579 if (unlikely(slot < 0)) 580 slot = 0; 581 582 slot >>= stab->szopts.cell_log; 583 if (likely(slot < stab->szopts.tsize)) 584 pkt_len = stab->data[slot]; 585 else 586 pkt_len = stab->data[stab->szopts.tsize - 1] * 587 (slot / stab->szopts.tsize) + 588 stab->data[slot % stab->szopts.tsize]; 589 590 pkt_len <<= stab->szopts.size_log; 591 out: 592 if (unlikely(pkt_len < 1)) 593 pkt_len = 1; 594 qdisc_skb_cb(skb)->pkt_len = pkt_len; 595 } 596 597 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc) 598 { 599 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 600 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n", 601 txt, qdisc->ops->id, qdisc->handle >> 16); 602 qdisc->flags |= TCQ_F_WARN_NONWC; 603 } 604 } 605 EXPORT_SYMBOL(qdisc_warn_nonwc); 606 607 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 608 { 609 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 610 timer); 611 612 rcu_read_lock(); 613 __netif_schedule(qdisc_root(wd->qdisc)); 614 rcu_read_unlock(); 615 616 return HRTIMER_NORESTART; 617 } 618 619 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc, 620 clockid_t clockid) 621 { 622 hrtimer_setup(&wd->timer, qdisc_watchdog, clockid, HRTIMER_MODE_ABS_PINNED); 623 wd->qdisc = qdisc; 624 } 625 EXPORT_SYMBOL(qdisc_watchdog_init_clockid); 626 627 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 628 { 629 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC); 630 } 631 EXPORT_SYMBOL(qdisc_watchdog_init); 632 633 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires, 634 u64 delta_ns) 635 { 636 bool deactivated; 637 638 rcu_read_lock(); 639 deactivated = test_bit(__QDISC_STATE_DEACTIVATED, 640 &qdisc_root_sleeping(wd->qdisc)->state); 641 rcu_read_unlock(); 642 if (deactivated) 643 return; 644 645 if (hrtimer_is_queued(&wd->timer)) { 646 u64 softexpires; 647 648 softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer)); 649 /* If timer is already set in [expires, expires + delta_ns], 650 * do not reprogram it. 651 */ 652 if (softexpires - expires <= delta_ns) 653 return; 654 } 655 656 hrtimer_start_range_ns(&wd->timer, 657 ns_to_ktime(expires), 658 delta_ns, 659 HRTIMER_MODE_ABS_PINNED); 660 } 661 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns); 662 663 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 664 { 665 hrtimer_cancel(&wd->timer); 666 } 667 EXPORT_SYMBOL(qdisc_watchdog_cancel); 668 669 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 670 { 671 struct hlist_head *h; 672 unsigned int i; 673 674 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL); 675 676 if (h != NULL) { 677 for (i = 0; i < n; i++) 678 INIT_HLIST_HEAD(&h[i]); 679 } 680 return h; 681 } 682 683 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 684 { 685 struct Qdisc_class_common *cl; 686 struct hlist_node *next; 687 struct hlist_head *nhash, *ohash; 688 unsigned int nsize, nmask, osize; 689 unsigned int i, h; 690 691 /* Rehash when load factor exceeds 0.75 */ 692 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 693 return; 694 nsize = clhash->hashsize * 2; 695 nmask = nsize - 1; 696 nhash = qdisc_class_hash_alloc(nsize); 697 if (nhash == NULL) 698 return; 699 700 ohash = clhash->hash; 701 osize = clhash->hashsize; 702 703 sch_tree_lock(sch); 704 for (i = 0; i < osize; i++) { 705 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) { 706 h = qdisc_class_hash(cl->classid, nmask); 707 hlist_add_head(&cl->hnode, &nhash[h]); 708 } 709 } 710 clhash->hash = nhash; 711 clhash->hashsize = nsize; 712 clhash->hashmask = nmask; 713 sch_tree_unlock(sch); 714 715 kvfree(ohash); 716 } 717 EXPORT_SYMBOL(qdisc_class_hash_grow); 718 719 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 720 { 721 unsigned int size = 4; 722 723 clhash->hash = qdisc_class_hash_alloc(size); 724 if (!clhash->hash) 725 return -ENOMEM; 726 clhash->hashsize = size; 727 clhash->hashmask = size - 1; 728 clhash->hashelems = 0; 729 return 0; 730 } 731 EXPORT_SYMBOL(qdisc_class_hash_init); 732 733 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 734 { 735 kvfree(clhash->hash); 736 } 737 EXPORT_SYMBOL(qdisc_class_hash_destroy); 738 739 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 740 struct Qdisc_class_common *cl) 741 { 742 unsigned int h; 743 744 INIT_HLIST_NODE(&cl->hnode); 745 h = qdisc_class_hash(cl->classid, clhash->hashmask); 746 hlist_add_head(&cl->hnode, &clhash->hash[h]); 747 clhash->hashelems++; 748 } 749 EXPORT_SYMBOL(qdisc_class_hash_insert); 750 751 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 752 struct Qdisc_class_common *cl) 753 { 754 hlist_del(&cl->hnode); 755 clhash->hashelems--; 756 } 757 EXPORT_SYMBOL(qdisc_class_hash_remove); 758 759 /* Allocate an unique handle from space managed by kernel 760 * Possible range is [8000-FFFF]:0000 (0x8000 values) 761 */ 762 static u32 qdisc_alloc_handle(struct net_device *dev) 763 { 764 int i = 0x8000; 765 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 766 767 do { 768 autohandle += TC_H_MAKE(0x10000U, 0); 769 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 770 autohandle = TC_H_MAKE(0x80000000U, 0); 771 if (!qdisc_lookup(dev, autohandle)) 772 return autohandle; 773 cond_resched(); 774 } while (--i > 0); 775 776 return 0; 777 } 778 779 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len) 780 { 781 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED; 782 const struct Qdisc_class_ops *cops; 783 unsigned long cl; 784 u32 parentid; 785 bool notify; 786 int drops; 787 788 if (n == 0 && len == 0) 789 return; 790 drops = max_t(int, n, 0); 791 rcu_read_lock(); 792 while ((parentid = sch->parent)) { 793 if (parentid == TC_H_ROOT) 794 break; 795 796 if (sch->flags & TCQ_F_NOPARENT) 797 break; 798 /* Notify parent qdisc only if child qdisc becomes empty. 799 * 800 * If child was empty even before update then backlog 801 * counter is screwed and we skip notification because 802 * parent class is already passive. 803 * 804 * If the original child was offloaded then it is allowed 805 * to be seem as empty, so the parent is notified anyway. 806 */ 807 notify = !sch->q.qlen && !WARN_ON_ONCE(!n && 808 !qdisc_is_offloaded); 809 /* TODO: perform the search on a per txq basis */ 810 sch = qdisc_lookup_rcu(qdisc_dev(sch), TC_H_MAJ(parentid)); 811 if (sch == NULL) { 812 WARN_ON_ONCE(parentid != TC_H_ROOT); 813 break; 814 } 815 cops = sch->ops->cl_ops; 816 if (notify && cops->qlen_notify) { 817 cl = cops->find(sch, parentid); 818 cops->qlen_notify(sch, cl); 819 } 820 sch->q.qlen -= n; 821 sch->qstats.backlog -= len; 822 __qdisc_qstats_drop(sch, drops); 823 } 824 rcu_read_unlock(); 825 } 826 EXPORT_SYMBOL(qdisc_tree_reduce_backlog); 827 828 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type, 829 void *type_data) 830 { 831 struct net_device *dev = qdisc_dev(sch); 832 int err; 833 834 sch->flags &= ~TCQ_F_OFFLOADED; 835 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 836 return 0; 837 838 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data); 839 if (err == -EOPNOTSUPP) 840 return 0; 841 842 if (!err) 843 sch->flags |= TCQ_F_OFFLOADED; 844 845 return err; 846 } 847 EXPORT_SYMBOL(qdisc_offload_dump_helper); 848 849 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch, 850 struct Qdisc *new, struct Qdisc *old, 851 enum tc_setup_type type, void *type_data, 852 struct netlink_ext_ack *extack) 853 { 854 bool any_qdisc_is_offloaded; 855 int err; 856 857 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 858 return; 859 860 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data); 861 862 /* Don't report error if the graft is part of destroy operation. */ 863 if (!err || !new || new == &noop_qdisc) 864 return; 865 866 /* Don't report error if the parent, the old child and the new 867 * one are not offloaded. 868 */ 869 any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED; 870 any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED; 871 any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED; 872 873 if (any_qdisc_is_offloaded) 874 NL_SET_ERR_MSG(extack, "Offloading graft operation failed."); 875 } 876 EXPORT_SYMBOL(qdisc_offload_graft_helper); 877 878 void qdisc_offload_query_caps(struct net_device *dev, 879 enum tc_setup_type type, 880 void *caps, size_t caps_len) 881 { 882 const struct net_device_ops *ops = dev->netdev_ops; 883 struct tc_query_caps_base base = { 884 .type = type, 885 .caps = caps, 886 }; 887 888 memset(caps, 0, caps_len); 889 890 if (ops->ndo_setup_tc) 891 ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base); 892 } 893 EXPORT_SYMBOL(qdisc_offload_query_caps); 894 895 static void qdisc_offload_graft_root(struct net_device *dev, 896 struct Qdisc *new, struct Qdisc *old, 897 struct netlink_ext_ack *extack) 898 { 899 struct tc_root_qopt_offload graft_offload = { 900 .command = TC_ROOT_GRAFT, 901 .handle = new ? new->handle : 0, 902 .ingress = (new && new->flags & TCQ_F_INGRESS) || 903 (old && old->flags & TCQ_F_INGRESS), 904 }; 905 906 qdisc_offload_graft_helper(dev, NULL, new, old, 907 TC_SETUP_ROOT_QDISC, &graft_offload, extack); 908 } 909 910 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 911 u32 portid, u32 seq, u16 flags, int event, 912 struct netlink_ext_ack *extack) 913 { 914 struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL; 915 struct gnet_stats_queue __percpu *cpu_qstats = NULL; 916 struct tcmsg *tcm; 917 struct nlmsghdr *nlh; 918 unsigned char *b = skb_tail_pointer(skb); 919 struct gnet_dump d; 920 struct qdisc_size_table *stab; 921 u32 block_index; 922 __u32 qlen; 923 924 cond_resched(); 925 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 926 if (!nlh) 927 goto out_nlmsg_trim; 928 tcm = nlmsg_data(nlh); 929 tcm->tcm_family = AF_UNSPEC; 930 tcm->tcm__pad1 = 0; 931 tcm->tcm__pad2 = 0; 932 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 933 tcm->tcm_parent = clid; 934 tcm->tcm_handle = q->handle; 935 tcm->tcm_info = refcount_read(&q->refcnt); 936 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 937 goto nla_put_failure; 938 if (q->ops->ingress_block_get) { 939 block_index = q->ops->ingress_block_get(q); 940 if (block_index && 941 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index)) 942 goto nla_put_failure; 943 } 944 if (q->ops->egress_block_get) { 945 block_index = q->ops->egress_block_get(q); 946 if (block_index && 947 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index)) 948 goto nla_put_failure; 949 } 950 if (q->ops->dump && q->ops->dump(q, skb) < 0) 951 goto nla_put_failure; 952 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED))) 953 goto nla_put_failure; 954 qlen = qdisc_qlen_sum(q); 955 956 stab = rtnl_dereference(q->stab); 957 if (stab && qdisc_dump_stab(skb, stab) < 0) 958 goto nla_put_failure; 959 960 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 961 NULL, &d, TCA_PAD) < 0) 962 goto nla_put_failure; 963 964 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 965 goto nla_put_failure; 966 967 if (qdisc_is_percpu_stats(q)) { 968 cpu_bstats = q->cpu_bstats; 969 cpu_qstats = q->cpu_qstats; 970 } 971 972 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 || 973 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 974 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0) 975 goto nla_put_failure; 976 977 if (gnet_stats_finish_copy(&d) < 0) 978 goto nla_put_failure; 979 980 if (extack && extack->_msg && 981 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg)) 982 goto out_nlmsg_trim; 983 984 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 985 986 return skb->len; 987 988 out_nlmsg_trim: 989 nla_put_failure: 990 nlmsg_trim(skb, b); 991 return -1; 992 } 993 994 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible) 995 { 996 if (q->flags & TCQ_F_BUILTIN) 997 return true; 998 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible) 999 return true; 1000 1001 return false; 1002 } 1003 1004 static int qdisc_get_notify(struct net *net, struct sk_buff *oskb, 1005 struct nlmsghdr *n, u32 clid, struct Qdisc *q, 1006 struct netlink_ext_ack *extack) 1007 { 1008 struct sk_buff *skb; 1009 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1010 1011 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1012 if (!skb) 1013 return -ENOBUFS; 1014 1015 if (!tc_qdisc_dump_ignore(q, false)) { 1016 if (tc_fill_qdisc(skb, q, clid, portid, n->nlmsg_seq, 0, 1017 RTM_NEWQDISC, extack) < 0) 1018 goto err_out; 1019 } 1020 1021 if (skb->len) 1022 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1023 n->nlmsg_flags & NLM_F_ECHO); 1024 1025 err_out: 1026 kfree_skb(skb); 1027 return -EINVAL; 1028 } 1029 1030 static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1031 struct nlmsghdr *n, u32 clid, 1032 struct Qdisc *old, struct Qdisc *new, 1033 struct netlink_ext_ack *extack) 1034 { 1035 struct sk_buff *skb; 1036 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1037 1038 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) 1039 return 0; 1040 1041 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1042 if (!skb) 1043 return -ENOBUFS; 1044 1045 if (old && !tc_qdisc_dump_ignore(old, false)) { 1046 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq, 1047 0, RTM_DELQDISC, extack) < 0) 1048 goto err_out; 1049 } 1050 if (new && !tc_qdisc_dump_ignore(new, false)) { 1051 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq, 1052 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0) 1053 goto err_out; 1054 } 1055 1056 if (skb->len) 1057 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1058 n->nlmsg_flags & NLM_F_ECHO); 1059 1060 err_out: 1061 kfree_skb(skb); 1062 return -EINVAL; 1063 } 1064 1065 static void notify_and_destroy(struct net *net, struct sk_buff *skb, 1066 struct nlmsghdr *n, u32 clid, 1067 struct Qdisc *old, struct Qdisc *new, 1068 struct netlink_ext_ack *extack) 1069 { 1070 if (new || old) 1071 qdisc_notify(net, skb, n, clid, old, new, extack); 1072 1073 if (old) 1074 qdisc_put(old); 1075 } 1076 1077 static void qdisc_clear_nolock(struct Qdisc *sch) 1078 { 1079 sch->flags &= ~TCQ_F_NOLOCK; 1080 if (!(sch->flags & TCQ_F_CPUSTATS)) 1081 return; 1082 1083 free_percpu(sch->cpu_bstats); 1084 free_percpu(sch->cpu_qstats); 1085 sch->cpu_bstats = NULL; 1086 sch->cpu_qstats = NULL; 1087 sch->flags &= ~TCQ_F_CPUSTATS; 1088 } 1089 1090 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 1091 * to device "dev". 1092 * 1093 * When appropriate send a netlink notification using 'skb' 1094 * and "n". 1095 * 1096 * On success, destroy old qdisc. 1097 */ 1098 1099 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 1100 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 1101 struct Qdisc *new, struct Qdisc *old, 1102 struct netlink_ext_ack *extack) 1103 { 1104 struct Qdisc *q = old; 1105 struct net *net = dev_net(dev); 1106 1107 if (parent == NULL) { 1108 unsigned int i, num_q, ingress; 1109 struct netdev_queue *dev_queue; 1110 1111 ingress = 0; 1112 num_q = dev->num_tx_queues; 1113 if ((q && q->flags & TCQ_F_INGRESS) || 1114 (new && new->flags & TCQ_F_INGRESS)) { 1115 ingress = 1; 1116 dev_queue = dev_ingress_queue(dev); 1117 if (!dev_queue) { 1118 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue"); 1119 return -ENOENT; 1120 } 1121 1122 q = rtnl_dereference(dev_queue->qdisc_sleeping); 1123 1124 /* This is the counterpart of that qdisc_refcount_inc_nz() call in 1125 * __tcf_qdisc_find() for filter requests. 1126 */ 1127 if (!qdisc_refcount_dec_if_one(q)) { 1128 NL_SET_ERR_MSG(extack, 1129 "Current ingress or clsact Qdisc has ongoing filter requests"); 1130 return -EBUSY; 1131 } 1132 } 1133 1134 if (dev->flags & IFF_UP) 1135 dev_deactivate(dev); 1136 1137 qdisc_offload_graft_root(dev, new, old, extack); 1138 1139 if (new && new->ops->attach && !ingress) 1140 goto skip; 1141 1142 if (!ingress) { 1143 for (i = 0; i < num_q; i++) { 1144 dev_queue = netdev_get_tx_queue(dev, i); 1145 old = dev_graft_qdisc(dev_queue, new); 1146 1147 if (new && i > 0) 1148 qdisc_refcount_inc(new); 1149 qdisc_put(old); 1150 } 1151 } else { 1152 old = dev_graft_qdisc(dev_queue, NULL); 1153 1154 /* {ingress,clsact}_destroy() @old before grafting @new to avoid 1155 * unprotected concurrent accesses to net_device::miniq_{in,e}gress 1156 * pointer(s) in mini_qdisc_pair_swap(). 1157 */ 1158 qdisc_notify(net, skb, n, classid, old, new, extack); 1159 qdisc_destroy(old); 1160 1161 dev_graft_qdisc(dev_queue, new); 1162 } 1163 1164 skip: 1165 if (!ingress) { 1166 old = rtnl_dereference(dev->qdisc); 1167 if (new && !new->ops->attach) 1168 qdisc_refcount_inc(new); 1169 rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc); 1170 1171 notify_and_destroy(net, skb, n, classid, old, new, extack); 1172 1173 if (new && new->ops->attach) 1174 new->ops->attach(new); 1175 } 1176 1177 if (dev->flags & IFF_UP) 1178 dev_activate(dev); 1179 } else { 1180 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 1181 unsigned long cl; 1182 int err; 1183 1184 /* Only support running class lockless if parent is lockless */ 1185 if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK)) 1186 qdisc_clear_nolock(new); 1187 1188 if (!cops || !cops->graft) 1189 return -EOPNOTSUPP; 1190 1191 cl = cops->find(parent, classid); 1192 if (!cl) { 1193 NL_SET_ERR_MSG(extack, "Specified class not found"); 1194 return -ENOENT; 1195 } 1196 1197 if (new && new->ops == &noqueue_qdisc_ops) { 1198 NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class"); 1199 return -EINVAL; 1200 } 1201 1202 if (new && 1203 !(parent->flags & TCQ_F_MQROOT) && 1204 rcu_access_pointer(new->stab)) { 1205 NL_SET_ERR_MSG(extack, "STAB not supported on a non root"); 1206 return -EINVAL; 1207 } 1208 err = cops->graft(parent, cl, new, &old, extack); 1209 if (err) 1210 return err; 1211 notify_and_destroy(net, skb, n, classid, old, new, extack); 1212 } 1213 return 0; 1214 } 1215 1216 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca, 1217 struct netlink_ext_ack *extack) 1218 { 1219 u32 block_index; 1220 1221 if (tca[TCA_INGRESS_BLOCK]) { 1222 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]); 1223 1224 if (!block_index) { 1225 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0"); 1226 return -EINVAL; 1227 } 1228 if (!sch->ops->ingress_block_set) { 1229 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported"); 1230 return -EOPNOTSUPP; 1231 } 1232 sch->ops->ingress_block_set(sch, block_index); 1233 } 1234 if (tca[TCA_EGRESS_BLOCK]) { 1235 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]); 1236 1237 if (!block_index) { 1238 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0"); 1239 return -EINVAL; 1240 } 1241 if (!sch->ops->egress_block_set) { 1242 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported"); 1243 return -EOPNOTSUPP; 1244 } 1245 sch->ops->egress_block_set(sch, block_index); 1246 } 1247 return 0; 1248 } 1249 1250 /* 1251 Allocate and initialize new qdisc. 1252 1253 Parameters are passed via opt. 1254 */ 1255 1256 static struct Qdisc *qdisc_create(struct net_device *dev, 1257 struct netdev_queue *dev_queue, 1258 u32 parent, u32 handle, 1259 struct nlattr **tca, int *errp, 1260 struct netlink_ext_ack *extack) 1261 { 1262 int err; 1263 struct nlattr *kind = tca[TCA_KIND]; 1264 struct Qdisc *sch; 1265 struct Qdisc_ops *ops; 1266 struct qdisc_size_table *stab; 1267 1268 ops = qdisc_lookup_ops(kind); 1269 #ifdef CONFIG_MODULES 1270 if (ops == NULL && kind != NULL) { 1271 char name[IFNAMSIZ]; 1272 if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) { 1273 /* We dropped the RTNL semaphore in order to 1274 * perform the module load. So, even if we 1275 * succeeded in loading the module we have to 1276 * tell the caller to replay the request. We 1277 * indicate this using -EAGAIN. 1278 * We replay the request because the device may 1279 * go away in the mean time. 1280 */ 1281 rtnl_unlock(); 1282 request_module(NET_SCH_ALIAS_PREFIX "%s", name); 1283 rtnl_lock(); 1284 ops = qdisc_lookup_ops(kind); 1285 if (ops != NULL) { 1286 /* We will try again qdisc_lookup_ops, 1287 * so don't keep a reference. 1288 */ 1289 module_put(ops->owner); 1290 err = -EAGAIN; 1291 goto err_out; 1292 } 1293 } 1294 } 1295 #endif 1296 1297 err = -ENOENT; 1298 if (!ops) { 1299 NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown"); 1300 goto err_out; 1301 } 1302 1303 sch = qdisc_alloc(dev_queue, ops, extack); 1304 if (IS_ERR(sch)) { 1305 err = PTR_ERR(sch); 1306 goto err_out2; 1307 } 1308 1309 sch->parent = parent; 1310 1311 if (handle == TC_H_INGRESS) { 1312 if (!(sch->flags & TCQ_F_INGRESS)) { 1313 NL_SET_ERR_MSG(extack, 1314 "Specified parent ID is reserved for ingress and clsact Qdiscs"); 1315 err = -EINVAL; 1316 goto err_out3; 1317 } 1318 handle = TC_H_MAKE(TC_H_INGRESS, 0); 1319 } else { 1320 if (handle == 0) { 1321 handle = qdisc_alloc_handle(dev); 1322 if (handle == 0) { 1323 NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded"); 1324 err = -ENOSPC; 1325 goto err_out3; 1326 } 1327 } 1328 if (!netif_is_multiqueue(dev)) 1329 sch->flags |= TCQ_F_ONETXQUEUE; 1330 } 1331 1332 sch->handle = handle; 1333 1334 /* This exist to keep backward compatible with a userspace 1335 * loophole, what allowed userspace to get IFF_NO_QUEUE 1336 * facility on older kernels by setting tx_queue_len=0 (prior 1337 * to qdisc init), and then forgot to reinit tx_queue_len 1338 * before again attaching a qdisc. 1339 */ 1340 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) { 1341 WRITE_ONCE(dev->tx_queue_len, DEFAULT_TX_QUEUE_LEN); 1342 netdev_info(dev, "Caught tx_queue_len zero misconfig\n"); 1343 } 1344 1345 err = qdisc_block_indexes_set(sch, tca, extack); 1346 if (err) 1347 goto err_out3; 1348 1349 if (tca[TCA_STAB]) { 1350 stab = qdisc_get_stab(tca[TCA_STAB], extack); 1351 if (IS_ERR(stab)) { 1352 err = PTR_ERR(stab); 1353 goto err_out3; 1354 } 1355 rcu_assign_pointer(sch->stab, stab); 1356 } 1357 1358 if (ops->init) { 1359 err = ops->init(sch, tca[TCA_OPTIONS], extack); 1360 if (err != 0) 1361 goto err_out4; 1362 } 1363 1364 if (tca[TCA_RATE]) { 1365 err = -EOPNOTSUPP; 1366 if (sch->flags & TCQ_F_MQROOT) { 1367 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc"); 1368 goto err_out4; 1369 } 1370 1371 err = gen_new_estimator(&sch->bstats, 1372 sch->cpu_bstats, 1373 &sch->rate_est, 1374 NULL, 1375 true, 1376 tca[TCA_RATE]); 1377 if (err) { 1378 NL_SET_ERR_MSG(extack, "Failed to generate new estimator"); 1379 goto err_out4; 1380 } 1381 } 1382 1383 qdisc_hash_add(sch, false); 1384 trace_qdisc_create(ops, dev, parent); 1385 1386 return sch; 1387 1388 err_out4: 1389 /* Even if ops->init() failed, we call ops->destroy() 1390 * like qdisc_create_dflt(). 1391 */ 1392 if (ops->destroy) 1393 ops->destroy(sch); 1394 qdisc_put_stab(rtnl_dereference(sch->stab)); 1395 err_out3: 1396 lockdep_unregister_key(&sch->root_lock_key); 1397 netdev_put(dev, &sch->dev_tracker); 1398 qdisc_free(sch); 1399 err_out2: 1400 module_put(ops->owner); 1401 err_out: 1402 *errp = err; 1403 return NULL; 1404 } 1405 1406 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca, 1407 struct netlink_ext_ack *extack) 1408 { 1409 struct qdisc_size_table *ostab, *stab = NULL; 1410 int err = 0; 1411 1412 if (tca[TCA_OPTIONS]) { 1413 if (!sch->ops->change) { 1414 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc"); 1415 return -EINVAL; 1416 } 1417 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) { 1418 NL_SET_ERR_MSG(extack, "Change of blocks is not supported"); 1419 return -EOPNOTSUPP; 1420 } 1421 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack); 1422 if (err) 1423 return err; 1424 } 1425 1426 if (tca[TCA_STAB]) { 1427 stab = qdisc_get_stab(tca[TCA_STAB], extack); 1428 if (IS_ERR(stab)) 1429 return PTR_ERR(stab); 1430 } 1431 1432 ostab = rtnl_dereference(sch->stab); 1433 rcu_assign_pointer(sch->stab, stab); 1434 qdisc_put_stab(ostab); 1435 1436 if (tca[TCA_RATE]) { 1437 /* NB: ignores errors from replace_estimator 1438 because change can't be undone. */ 1439 if (sch->flags & TCQ_F_MQROOT) 1440 goto out; 1441 gen_replace_estimator(&sch->bstats, 1442 sch->cpu_bstats, 1443 &sch->rate_est, 1444 NULL, 1445 true, 1446 tca[TCA_RATE]); 1447 } 1448 out: 1449 return 0; 1450 } 1451 1452 struct check_loop_arg { 1453 struct qdisc_walker w; 1454 struct Qdisc *p; 1455 int depth; 1456 }; 1457 1458 static int check_loop_fn(struct Qdisc *q, unsigned long cl, 1459 struct qdisc_walker *w); 1460 1461 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 1462 { 1463 struct check_loop_arg arg; 1464 1465 if (q->ops->cl_ops == NULL) 1466 return 0; 1467 1468 arg.w.stop = arg.w.skip = arg.w.count = 0; 1469 arg.w.fn = check_loop_fn; 1470 arg.depth = depth; 1471 arg.p = p; 1472 q->ops->cl_ops->walk(q, &arg.w); 1473 return arg.w.stop ? -ELOOP : 0; 1474 } 1475 1476 static int 1477 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 1478 { 1479 struct Qdisc *leaf; 1480 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 1481 struct check_loop_arg *arg = (struct check_loop_arg *)w; 1482 1483 leaf = cops->leaf(q, cl); 1484 if (leaf) { 1485 if (leaf == arg->p || arg->depth > 7) 1486 return -ELOOP; 1487 return check_loop(leaf, arg->p, arg->depth + 1); 1488 } 1489 return 0; 1490 } 1491 1492 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = { 1493 [TCA_KIND] = { .type = NLA_STRING }, 1494 [TCA_RATE] = { .type = NLA_BINARY, 1495 .len = sizeof(struct tc_estimator) }, 1496 [TCA_STAB] = { .type = NLA_NESTED }, 1497 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG }, 1498 [TCA_CHAIN] = { .type = NLA_U32 }, 1499 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 }, 1500 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 }, 1501 }; 1502 1503 /* 1504 * Delete/get qdisc. 1505 */ 1506 1507 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1508 struct netlink_ext_ack *extack) 1509 { 1510 struct net *net = sock_net(skb->sk); 1511 struct tcmsg *tcm = nlmsg_data(n); 1512 struct nlattr *tca[TCA_MAX + 1]; 1513 struct net_device *dev; 1514 u32 clid; 1515 struct Qdisc *q = NULL; 1516 struct Qdisc *p = NULL; 1517 int err; 1518 1519 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX, 1520 rtm_tca_policy, extack); 1521 if (err < 0) 1522 return err; 1523 1524 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1525 if (!dev) 1526 return -ENODEV; 1527 1528 clid = tcm->tcm_parent; 1529 if (clid) { 1530 if (clid != TC_H_ROOT) { 1531 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 1532 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1533 if (!p) { 1534 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid"); 1535 return -ENOENT; 1536 } 1537 q = qdisc_leaf(p, clid); 1538 } else if (dev_ingress_queue(dev)) { 1539 q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping); 1540 } 1541 } else { 1542 q = rtnl_dereference(dev->qdisc); 1543 } 1544 if (!q) { 1545 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device"); 1546 return -ENOENT; 1547 } 1548 1549 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) { 1550 NL_SET_ERR_MSG(extack, "Invalid handle"); 1551 return -EINVAL; 1552 } 1553 } else { 1554 q = qdisc_lookup(dev, tcm->tcm_handle); 1555 if (!q) { 1556 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle"); 1557 return -ENOENT; 1558 } 1559 } 1560 1561 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) { 1562 NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc"); 1563 return -EINVAL; 1564 } 1565 1566 if (n->nlmsg_type == RTM_DELQDISC) { 1567 if (!clid) { 1568 NL_SET_ERR_MSG(extack, "Classid cannot be zero"); 1569 return -EINVAL; 1570 } 1571 if (q->handle == 0) { 1572 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero"); 1573 return -ENOENT; 1574 } 1575 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack); 1576 if (err != 0) 1577 return err; 1578 } else { 1579 qdisc_get_notify(net, skb, n, clid, q, NULL); 1580 } 1581 return 0; 1582 } 1583 1584 static bool req_create_or_replace(struct nlmsghdr *n) 1585 { 1586 return (n->nlmsg_flags & NLM_F_CREATE && 1587 n->nlmsg_flags & NLM_F_REPLACE); 1588 } 1589 1590 static bool req_create_exclusive(struct nlmsghdr *n) 1591 { 1592 return (n->nlmsg_flags & NLM_F_CREATE && 1593 n->nlmsg_flags & NLM_F_EXCL); 1594 } 1595 1596 static bool req_change(struct nlmsghdr *n) 1597 { 1598 return (!(n->nlmsg_flags & NLM_F_CREATE) && 1599 !(n->nlmsg_flags & NLM_F_REPLACE) && 1600 !(n->nlmsg_flags & NLM_F_EXCL)); 1601 } 1602 1603 /* 1604 * Create/change qdisc. 1605 */ 1606 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, 1607 struct netlink_ext_ack *extack) 1608 { 1609 struct net *net = sock_net(skb->sk); 1610 struct tcmsg *tcm; 1611 struct nlattr *tca[TCA_MAX + 1]; 1612 struct net_device *dev; 1613 u32 clid; 1614 struct Qdisc *q, *p; 1615 int err; 1616 1617 replay: 1618 /* Reinit, just in case something touches this. */ 1619 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX, 1620 rtm_tca_policy, extack); 1621 if (err < 0) 1622 return err; 1623 1624 tcm = nlmsg_data(n); 1625 clid = tcm->tcm_parent; 1626 q = p = NULL; 1627 1628 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 1629 if (!dev) 1630 return -ENODEV; 1631 1632 1633 if (clid) { 1634 if (clid != TC_H_ROOT) { 1635 if (clid != TC_H_INGRESS) { 1636 p = qdisc_lookup(dev, TC_H_MAJ(clid)); 1637 if (!p) { 1638 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc"); 1639 return -ENOENT; 1640 } 1641 q = qdisc_leaf(p, clid); 1642 } else if (dev_ingress_queue_create(dev)) { 1643 q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping); 1644 } 1645 } else { 1646 q = rtnl_dereference(dev->qdisc); 1647 } 1648 1649 /* It may be default qdisc, ignore it */ 1650 if (q && q->handle == 0) 1651 q = NULL; 1652 1653 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1654 if (tcm->tcm_handle) { 1655 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) { 1656 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override"); 1657 return -EEXIST; 1658 } 1659 if (TC_H_MIN(tcm->tcm_handle)) { 1660 NL_SET_ERR_MSG(extack, "Invalid minor handle"); 1661 return -EINVAL; 1662 } 1663 q = qdisc_lookup(dev, tcm->tcm_handle); 1664 if (!q) 1665 goto create_n_graft; 1666 if (q->parent != tcm->tcm_parent) { 1667 NL_SET_ERR_MSG(extack, "Cannot move an existing qdisc to a different parent"); 1668 return -EINVAL; 1669 } 1670 if (n->nlmsg_flags & NLM_F_EXCL) { 1671 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override"); 1672 return -EEXIST; 1673 } 1674 if (tca[TCA_KIND] && 1675 nla_strcmp(tca[TCA_KIND], q->ops->id)) { 1676 NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc"); 1677 return -EINVAL; 1678 } 1679 if (q->flags & TCQ_F_INGRESS) { 1680 NL_SET_ERR_MSG(extack, 1681 "Cannot regraft ingress or clsact Qdiscs"); 1682 return -EINVAL; 1683 } 1684 if (q == p || 1685 (p && check_loop(q, p, 0))) { 1686 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected"); 1687 return -ELOOP; 1688 } 1689 if (clid == TC_H_INGRESS) { 1690 NL_SET_ERR_MSG(extack, "Ingress cannot graft directly"); 1691 return -EINVAL; 1692 } 1693 qdisc_refcount_inc(q); 1694 goto graft; 1695 } else { 1696 if (!q) 1697 goto create_n_graft; 1698 1699 /* This magic test requires explanation. 1700 * 1701 * We know, that some child q is already 1702 * attached to this parent and have choice: 1703 * 1) change it or 2) create/graft new one. 1704 * If the requested qdisc kind is different 1705 * than the existing one, then we choose graft. 1706 * If they are the same then this is "change" 1707 * operation - just let it fallthrough.. 1708 * 1709 * 1. We are allowed to create/graft only 1710 * if the request is explicitly stating 1711 * "please create if it doesn't exist". 1712 * 1713 * 2. If the request is to exclusive create 1714 * then the qdisc tcm_handle is not expected 1715 * to exist, so that we choose create/graft too. 1716 * 1717 * 3. The last case is when no flags are set. 1718 * This will happen when for example tc 1719 * utility issues a "change" command. 1720 * Alas, it is sort of hole in API, we 1721 * cannot decide what to do unambiguously. 1722 * For now we select create/graft. 1723 */ 1724 if (tca[TCA_KIND] && 1725 nla_strcmp(tca[TCA_KIND], q->ops->id)) { 1726 if (req_create_or_replace(n) || 1727 req_create_exclusive(n)) 1728 goto create_n_graft; 1729 else if (req_change(n)) 1730 goto create_n_graft2; 1731 } 1732 } 1733 } 1734 } else { 1735 if (!tcm->tcm_handle) { 1736 NL_SET_ERR_MSG(extack, "Handle cannot be zero"); 1737 return -EINVAL; 1738 } 1739 q = qdisc_lookup(dev, tcm->tcm_handle); 1740 } 1741 1742 /* Change qdisc parameters */ 1743 if (!q) { 1744 NL_SET_ERR_MSG(extack, "Specified qdisc not found"); 1745 return -ENOENT; 1746 } 1747 if (n->nlmsg_flags & NLM_F_EXCL) { 1748 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify"); 1749 return -EEXIST; 1750 } 1751 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) { 1752 NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc"); 1753 return -EINVAL; 1754 } 1755 err = qdisc_change(q, tca, extack); 1756 if (err == 0) 1757 qdisc_notify(net, skb, n, clid, NULL, q, extack); 1758 return err; 1759 1760 create_n_graft: 1761 if (!(n->nlmsg_flags & NLM_F_CREATE)) { 1762 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag"); 1763 return -ENOENT; 1764 } 1765 create_n_graft2: 1766 if (clid == TC_H_INGRESS) { 1767 if (dev_ingress_queue(dev)) { 1768 q = qdisc_create(dev, dev_ingress_queue(dev), 1769 tcm->tcm_parent, tcm->tcm_parent, 1770 tca, &err, extack); 1771 } else { 1772 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device"); 1773 err = -ENOENT; 1774 } 1775 } else { 1776 struct netdev_queue *dev_queue; 1777 1778 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1779 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1780 else if (p) 1781 dev_queue = p->dev_queue; 1782 else 1783 dev_queue = netdev_get_tx_queue(dev, 0); 1784 1785 q = qdisc_create(dev, dev_queue, 1786 tcm->tcm_parent, tcm->tcm_handle, 1787 tca, &err, extack); 1788 } 1789 if (q == NULL) { 1790 if (err == -EAGAIN) 1791 goto replay; 1792 return err; 1793 } 1794 1795 graft: 1796 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack); 1797 if (err) { 1798 if (q) 1799 qdisc_put(q); 1800 return err; 1801 } 1802 1803 return 0; 1804 } 1805 1806 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1807 struct netlink_callback *cb, 1808 int *q_idx_p, int s_q_idx, bool recur, 1809 bool dump_invisible) 1810 { 1811 int ret = 0, q_idx = *q_idx_p; 1812 struct Qdisc *q; 1813 int b; 1814 1815 if (!root) 1816 return 0; 1817 1818 q = root; 1819 if (q_idx < s_q_idx) { 1820 q_idx++; 1821 } else { 1822 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1823 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1824 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1825 RTM_NEWQDISC, NULL) <= 0) 1826 goto done; 1827 q_idx++; 1828 } 1829 1830 /* If dumping singletons, there is no qdisc_dev(root) and the singleton 1831 * itself has already been dumped. 1832 * 1833 * If we've already dumped the top-level (ingress) qdisc above and the global 1834 * qdisc hashtable, we don't want to hit it again 1835 */ 1836 if (!qdisc_dev(root) || !recur) 1837 goto out; 1838 1839 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 1840 if (q_idx < s_q_idx) { 1841 q_idx++; 1842 continue; 1843 } 1844 if (!tc_qdisc_dump_ignore(q, dump_invisible) && 1845 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid, 1846 cb->nlh->nlmsg_seq, NLM_F_MULTI, 1847 RTM_NEWQDISC, NULL) <= 0) 1848 goto done; 1849 q_idx++; 1850 } 1851 1852 out: 1853 *q_idx_p = q_idx; 1854 return ret; 1855 done: 1856 ret = -1; 1857 goto out; 1858 } 1859 1860 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1861 { 1862 struct net *net = sock_net(skb->sk); 1863 int idx, q_idx; 1864 int s_idx, s_q_idx; 1865 struct net_device *dev; 1866 const struct nlmsghdr *nlh = cb->nlh; 1867 struct nlattr *tca[TCA_MAX + 1]; 1868 int err; 1869 1870 s_idx = cb->args[0]; 1871 s_q_idx = q_idx = cb->args[1]; 1872 1873 idx = 0; 1874 ASSERT_RTNL(); 1875 1876 err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX, 1877 rtm_tca_policy, cb->extack); 1878 if (err < 0) 1879 return err; 1880 1881 for_each_netdev(net, dev) { 1882 struct netdev_queue *dev_queue; 1883 1884 if (idx < s_idx) 1885 goto cont; 1886 if (idx > s_idx) 1887 s_q_idx = 0; 1888 q_idx = 0; 1889 1890 if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc), 1891 skb, cb, &q_idx, s_q_idx, 1892 true, tca[TCA_DUMP_INVISIBLE]) < 0) 1893 goto done; 1894 1895 dev_queue = dev_ingress_queue(dev); 1896 if (dev_queue && 1897 tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping), 1898 skb, cb, &q_idx, s_q_idx, false, 1899 tca[TCA_DUMP_INVISIBLE]) < 0) 1900 goto done; 1901 1902 cont: 1903 idx++; 1904 } 1905 1906 done: 1907 cb->args[0] = idx; 1908 cb->args[1] = q_idx; 1909 1910 return skb->len; 1911 } 1912 1913 1914 1915 /************************************************ 1916 * Traffic classes manipulation. * 1917 ************************************************/ 1918 1919 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1920 unsigned long cl, u32 portid, u32 seq, u16 flags, 1921 int event, struct netlink_ext_ack *extack) 1922 { 1923 struct tcmsg *tcm; 1924 struct nlmsghdr *nlh; 1925 unsigned char *b = skb_tail_pointer(skb); 1926 struct gnet_dump d; 1927 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1928 1929 cond_resched(); 1930 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags); 1931 if (!nlh) 1932 goto out_nlmsg_trim; 1933 tcm = nlmsg_data(nlh); 1934 tcm->tcm_family = AF_UNSPEC; 1935 tcm->tcm__pad1 = 0; 1936 tcm->tcm__pad2 = 0; 1937 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1938 tcm->tcm_parent = q->handle; 1939 tcm->tcm_handle = q->handle; 1940 tcm->tcm_info = 0; 1941 if (nla_put_string(skb, TCA_KIND, q->ops->id)) 1942 goto nla_put_failure; 1943 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1944 goto nla_put_failure; 1945 1946 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1947 NULL, &d, TCA_PAD) < 0) 1948 goto nla_put_failure; 1949 1950 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1951 goto nla_put_failure; 1952 1953 if (gnet_stats_finish_copy(&d) < 0) 1954 goto nla_put_failure; 1955 1956 if (extack && extack->_msg && 1957 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg)) 1958 goto out_nlmsg_trim; 1959 1960 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1961 1962 return skb->len; 1963 1964 out_nlmsg_trim: 1965 nla_put_failure: 1966 nlmsg_trim(skb, b); 1967 return -1; 1968 } 1969 1970 static int tclass_notify(struct net *net, struct sk_buff *oskb, 1971 struct nlmsghdr *n, struct Qdisc *q, 1972 unsigned long cl, int event, struct netlink_ext_ack *extack) 1973 { 1974 struct sk_buff *skb; 1975 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1976 1977 if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) 1978 return 0; 1979 1980 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1981 if (!skb) 1982 return -ENOBUFS; 1983 1984 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) { 1985 kfree_skb(skb); 1986 return -EINVAL; 1987 } 1988 1989 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 1990 n->nlmsg_flags & NLM_F_ECHO); 1991 } 1992 1993 static int tclass_get_notify(struct net *net, struct sk_buff *oskb, 1994 struct nlmsghdr *n, struct Qdisc *q, 1995 unsigned long cl, struct netlink_ext_ack *extack) 1996 { 1997 struct sk_buff *skb; 1998 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 1999 2000 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2001 if (!skb) 2002 return -ENOBUFS; 2003 2004 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, RTM_NEWTCLASS, 2005 extack) < 0) { 2006 kfree_skb(skb); 2007 return -EINVAL; 2008 } 2009 2010 return rtnetlink_send(skb, net, portid, RTNLGRP_TC, 2011 n->nlmsg_flags & NLM_F_ECHO); 2012 } 2013 2014 static int tclass_del_notify(struct net *net, 2015 const struct Qdisc_class_ops *cops, 2016 struct sk_buff *oskb, struct nlmsghdr *n, 2017 struct Qdisc *q, unsigned long cl, 2018 struct netlink_ext_ack *extack) 2019 { 2020 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0; 2021 struct sk_buff *skb; 2022 int err = 0; 2023 2024 if (!cops->delete) 2025 return -EOPNOTSUPP; 2026 2027 if (rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) { 2028 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2029 if (!skb) 2030 return -ENOBUFS; 2031 2032 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, 2033 RTM_DELTCLASS, extack) < 0) { 2034 kfree_skb(skb); 2035 return -EINVAL; 2036 } 2037 } else { 2038 skb = NULL; 2039 } 2040 2041 err = cops->delete(q, cl, extack); 2042 if (err) { 2043 kfree_skb(skb); 2044 return err; 2045 } 2046 2047 err = rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC, 2048 n->nlmsg_flags & NLM_F_ECHO); 2049 return err; 2050 } 2051 2052 #ifdef CONFIG_NET_CLS 2053 2054 struct tcf_bind_args { 2055 struct tcf_walker w; 2056 unsigned long base; 2057 unsigned long cl; 2058 u32 classid; 2059 }; 2060 2061 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg) 2062 { 2063 struct tcf_bind_args *a = (void *)arg; 2064 2065 if (n && tp->ops->bind_class) { 2066 struct Qdisc *q = tcf_block_q(tp->chain->block); 2067 2068 sch_tree_lock(q); 2069 tp->ops->bind_class(n, a->classid, a->cl, q, a->base); 2070 sch_tree_unlock(q); 2071 } 2072 return 0; 2073 } 2074 2075 struct tc_bind_class_args { 2076 struct qdisc_walker w; 2077 unsigned long new_cl; 2078 u32 portid; 2079 u32 clid; 2080 }; 2081 2082 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl, 2083 struct qdisc_walker *w) 2084 { 2085 struct tc_bind_class_args *a = (struct tc_bind_class_args *)w; 2086 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 2087 struct tcf_block *block; 2088 struct tcf_chain *chain; 2089 2090 block = cops->tcf_block(q, cl, NULL); 2091 if (!block) 2092 return 0; 2093 for (chain = tcf_get_next_chain(block, NULL); 2094 chain; 2095 chain = tcf_get_next_chain(block, chain)) { 2096 struct tcf_proto *tp; 2097 2098 for (tp = tcf_get_next_proto(chain, NULL); 2099 tp; tp = tcf_get_next_proto(chain, tp)) { 2100 struct tcf_bind_args arg = {}; 2101 2102 arg.w.fn = tcf_node_bind; 2103 arg.classid = a->clid; 2104 arg.base = cl; 2105 arg.cl = a->new_cl; 2106 tp->ops->walk(tp, &arg.w, true); 2107 } 2108 } 2109 2110 return 0; 2111 } 2112 2113 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 2114 unsigned long new_cl) 2115 { 2116 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 2117 struct tc_bind_class_args args = {}; 2118 2119 if (!cops->tcf_block) 2120 return; 2121 args.portid = portid; 2122 args.clid = clid; 2123 args.new_cl = new_cl; 2124 args.w.fn = tc_bind_class_walker; 2125 q->ops->cl_ops->walk(q, &args.w); 2126 } 2127 2128 #else 2129 2130 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid, 2131 unsigned long new_cl) 2132 { 2133 } 2134 2135 #endif 2136 2137 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, 2138 struct netlink_ext_ack *extack) 2139 { 2140 struct net *net = sock_net(skb->sk); 2141 struct tcmsg *tcm = nlmsg_data(n); 2142 struct nlattr *tca[TCA_MAX + 1]; 2143 struct net_device *dev; 2144 struct Qdisc *q = NULL; 2145 const struct Qdisc_class_ops *cops; 2146 unsigned long cl = 0; 2147 unsigned long new_cl; 2148 u32 portid; 2149 u32 clid; 2150 u32 qid; 2151 int err; 2152 2153 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX, 2154 rtm_tca_policy, extack); 2155 if (err < 0) 2156 return err; 2157 2158 dev = __dev_get_by_index(net, tcm->tcm_ifindex); 2159 if (!dev) 2160 return -ENODEV; 2161 2162 /* 2163 parent == TC_H_UNSPEC - unspecified parent. 2164 parent == TC_H_ROOT - class is root, which has no parent. 2165 parent == X:0 - parent is root class. 2166 parent == X:Y - parent is a node in hierarchy. 2167 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 2168 2169 handle == 0:0 - generate handle from kernel pool. 2170 handle == 0:Y - class is X:Y, where X:0 is qdisc. 2171 handle == X:Y - clear. 2172 handle == X:0 - root class. 2173 */ 2174 2175 /* Step 1. Determine qdisc handle X:0 */ 2176 2177 portid = tcm->tcm_parent; 2178 clid = tcm->tcm_handle; 2179 qid = TC_H_MAJ(clid); 2180 2181 if (portid != TC_H_ROOT) { 2182 u32 qid1 = TC_H_MAJ(portid); 2183 2184 if (qid && qid1) { 2185 /* If both majors are known, they must be identical. */ 2186 if (qid != qid1) 2187 return -EINVAL; 2188 } else if (qid1) { 2189 qid = qid1; 2190 } else if (qid == 0) 2191 qid = rtnl_dereference(dev->qdisc)->handle; 2192 2193 /* Now qid is genuine qdisc handle consistent 2194 * both with parent and child. 2195 * 2196 * TC_H_MAJ(portid) still may be unspecified, complete it now. 2197 */ 2198 if (portid) 2199 portid = TC_H_MAKE(qid, portid); 2200 } else { 2201 if (qid == 0) 2202 qid = rtnl_dereference(dev->qdisc)->handle; 2203 } 2204 2205 /* OK. Locate qdisc */ 2206 q = qdisc_lookup(dev, qid); 2207 if (!q) 2208 return -ENOENT; 2209 2210 /* An check that it supports classes */ 2211 cops = q->ops->cl_ops; 2212 if (cops == NULL) 2213 return -EINVAL; 2214 2215 /* Now try to get class */ 2216 if (clid == 0) { 2217 if (portid == TC_H_ROOT) 2218 clid = qid; 2219 } else 2220 clid = TC_H_MAKE(qid, clid); 2221 2222 if (clid) 2223 cl = cops->find(q, clid); 2224 2225 if (cl == 0) { 2226 err = -ENOENT; 2227 if (n->nlmsg_type != RTM_NEWTCLASS || 2228 !(n->nlmsg_flags & NLM_F_CREATE)) 2229 goto out; 2230 } else { 2231 switch (n->nlmsg_type) { 2232 case RTM_NEWTCLASS: 2233 err = -EEXIST; 2234 if (n->nlmsg_flags & NLM_F_EXCL) 2235 goto out; 2236 break; 2237 case RTM_DELTCLASS: 2238 err = tclass_del_notify(net, cops, skb, n, q, cl, extack); 2239 /* Unbind the class with flilters with 0 */ 2240 tc_bind_tclass(q, portid, clid, 0); 2241 goto out; 2242 case RTM_GETTCLASS: 2243 err = tclass_get_notify(net, skb, n, q, cl, extack); 2244 goto out; 2245 default: 2246 err = -EINVAL; 2247 goto out; 2248 } 2249 } 2250 2251 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) { 2252 NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes"); 2253 return -EOPNOTSUPP; 2254 } 2255 2256 /* Prevent creation of traffic classes with classid TC_H_ROOT */ 2257 if (clid == TC_H_ROOT) { 2258 NL_SET_ERR_MSG(extack, "Cannot create traffic class with classid TC_H_ROOT"); 2259 return -EINVAL; 2260 } 2261 2262 new_cl = cl; 2263 err = -EOPNOTSUPP; 2264 if (cops->change) 2265 err = cops->change(q, clid, portid, tca, &new_cl, extack); 2266 if (err == 0) { 2267 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack); 2268 /* We just create a new class, need to do reverse binding. */ 2269 if (cl != new_cl) 2270 tc_bind_tclass(q, portid, clid, new_cl); 2271 } 2272 out: 2273 return err; 2274 } 2275 2276 struct qdisc_dump_args { 2277 struct qdisc_walker w; 2278 struct sk_buff *skb; 2279 struct netlink_callback *cb; 2280 }; 2281 2282 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, 2283 struct qdisc_walker *arg) 2284 { 2285 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 2286 2287 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid, 2288 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, 2289 RTM_NEWTCLASS, NULL); 2290 } 2291 2292 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 2293 struct tcmsg *tcm, struct netlink_callback *cb, 2294 int *t_p, int s_t) 2295 { 2296 struct qdisc_dump_args arg; 2297 2298 if (tc_qdisc_dump_ignore(q, false) || 2299 *t_p < s_t || !q->ops->cl_ops || 2300 (tcm->tcm_parent && 2301 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 2302 (*t_p)++; 2303 return 0; 2304 } 2305 if (*t_p > s_t) 2306 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 2307 arg.w.fn = qdisc_class_dump; 2308 arg.skb = skb; 2309 arg.cb = cb; 2310 arg.w.stop = 0; 2311 arg.w.skip = cb->args[1]; 2312 arg.w.count = 0; 2313 q->ops->cl_ops->walk(q, &arg.w); 2314 cb->args[1] = arg.w.count; 2315 if (arg.w.stop) 2316 return -1; 2317 (*t_p)++; 2318 return 0; 2319 } 2320 2321 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 2322 struct tcmsg *tcm, struct netlink_callback *cb, 2323 int *t_p, int s_t, bool recur) 2324 { 2325 struct Qdisc *q; 2326 int b; 2327 2328 if (!root) 2329 return 0; 2330 2331 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 2332 return -1; 2333 2334 if (!qdisc_dev(root) || !recur) 2335 return 0; 2336 2337 if (tcm->tcm_parent) { 2338 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent)); 2339 if (q && q != root && 2340 tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 2341 return -1; 2342 return 0; 2343 } 2344 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) { 2345 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 2346 return -1; 2347 } 2348 2349 return 0; 2350 } 2351 2352 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 2353 { 2354 struct tcmsg *tcm = nlmsg_data(cb->nlh); 2355 struct net *net = sock_net(skb->sk); 2356 struct netdev_queue *dev_queue; 2357 struct net_device *dev; 2358 int t, s_t; 2359 2360 if (nlmsg_len(cb->nlh) < sizeof(*tcm)) 2361 return 0; 2362 dev = dev_get_by_index(net, tcm->tcm_ifindex); 2363 if (!dev) 2364 return 0; 2365 2366 s_t = cb->args[0]; 2367 t = 0; 2368 2369 if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc), 2370 skb, tcm, cb, &t, s_t, true) < 0) 2371 goto done; 2372 2373 dev_queue = dev_ingress_queue(dev); 2374 if (dev_queue && 2375 tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping), 2376 skb, tcm, cb, &t, s_t, false) < 0) 2377 goto done; 2378 2379 done: 2380 cb->args[0] = t; 2381 2382 dev_put(dev); 2383 return skb->len; 2384 } 2385 2386 #ifdef CONFIG_PROC_FS 2387 static int psched_show(struct seq_file *seq, void *v) 2388 { 2389 seq_printf(seq, "%08x %08x %08x %08x\n", 2390 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 2391 1000000, 2392 (u32)NSEC_PER_SEC / hrtimer_resolution); 2393 2394 return 0; 2395 } 2396 2397 static int __net_init psched_net_init(struct net *net) 2398 { 2399 struct proc_dir_entry *e; 2400 2401 e = proc_create_single("psched", 0, net->proc_net, psched_show); 2402 if (e == NULL) 2403 return -ENOMEM; 2404 2405 return 0; 2406 } 2407 2408 static void __net_exit psched_net_exit(struct net *net) 2409 { 2410 remove_proc_entry("psched", net->proc_net); 2411 } 2412 #else 2413 static int __net_init psched_net_init(struct net *net) 2414 { 2415 return 0; 2416 } 2417 2418 static void __net_exit psched_net_exit(struct net *net) 2419 { 2420 } 2421 #endif 2422 2423 static struct pernet_operations psched_net_ops = { 2424 .init = psched_net_init, 2425 .exit = psched_net_exit, 2426 }; 2427 2428 #if IS_ENABLED(CONFIG_MITIGATION_RETPOLINE) 2429 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper); 2430 #endif 2431 2432 static const struct rtnl_msg_handler psched_rtnl_msg_handlers[] __initconst = { 2433 {.msgtype = RTM_NEWQDISC, .doit = tc_modify_qdisc}, 2434 {.msgtype = RTM_DELQDISC, .doit = tc_get_qdisc}, 2435 {.msgtype = RTM_GETQDISC, .doit = tc_get_qdisc, 2436 .dumpit = tc_dump_qdisc}, 2437 {.msgtype = RTM_NEWTCLASS, .doit = tc_ctl_tclass}, 2438 {.msgtype = RTM_DELTCLASS, .doit = tc_ctl_tclass}, 2439 {.msgtype = RTM_GETTCLASS, .doit = tc_ctl_tclass, 2440 .dumpit = tc_dump_tclass}, 2441 }; 2442 2443 static int __init pktsched_init(void) 2444 { 2445 int err; 2446 2447 err = register_pernet_subsys(&psched_net_ops); 2448 if (err) { 2449 pr_err("pktsched_init: " 2450 "cannot initialize per netns operations\n"); 2451 return err; 2452 } 2453 2454 register_qdisc(&pfifo_fast_ops); 2455 register_qdisc(&pfifo_qdisc_ops); 2456 register_qdisc(&bfifo_qdisc_ops); 2457 register_qdisc(&pfifo_head_drop_qdisc_ops); 2458 register_qdisc(&mq_qdisc_ops); 2459 register_qdisc(&noqueue_qdisc_ops); 2460 2461 rtnl_register_many(psched_rtnl_msg_handlers); 2462 2463 tc_wrapper_init(); 2464 2465 return 0; 2466 } 2467 2468 subsys_initcall(pktsched_init); 2469