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 38static int qdisc_notify(struct net *net, struct sk_buff *oskb, 39 struct nlmsghdr *n, u32 clid, 40 struct Qdisc *old, struct Qdisc *new); 41static 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. */ 126static DEFINE_RWLOCK(qdisc_mod_lock); 127 128 129/************************************************ 130 * Queueing disciplines manipulation. * 131 ************************************************/ 132 133 134/* The list of all installed queueing disciplines. */ 135 136static struct Qdisc_ops *qdisc_base; 137 138/* Register/uregister queueing discipline */ 139 140int 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; 174out: 175 write_unlock(&qdisc_mod_lock); 176 return rc; 177 178out_einval: 179 rc = -EINVAL; 180 goto out; 181} 182EXPORT_SYMBOL(register_qdisc); 183 184int 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} 201EXPORT_SYMBOL(unregister_qdisc); 202 203/* We know handle. Find qdisc among all qdisc's attached to device 204 (root qdisc, all its children, children of children etc.) 205 */ 206 207static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) 208{ 209 struct Qdisc *q; 210 211 if (!(root->flags & TCQ_F_BUILTIN) && 212 root->handle == handle) 213 return root; 214 215 list_for_each_entry(q, &root->list, list) { 216 if (q->handle == handle) 217 return q; 218 } 219 return NULL; 220} 221 222static void qdisc_list_add(struct Qdisc *q) 223{ 224 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) 225 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list); 226} 227 228void qdisc_list_del(struct Qdisc *q) 229{ 230 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) 231 list_del(&q->list); 232} 233EXPORT_SYMBOL(qdisc_list_del); 234 235struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 236{ 237 struct Qdisc *q; 238 239 q = qdisc_match_from_root(dev->qdisc, handle); 240 if (q) 241 goto out; 242 243 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle); 244out: 245 return q; 246} 247 248static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 249{ 250 unsigned long cl; 251 struct Qdisc *leaf; 252 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 253 254 if (cops == NULL) 255 return NULL; 256 cl = cops->get(p, classid); 257 258 if (cl == 0) 259 return NULL; 260 leaf = cops->leaf(p, cl); 261 cops->put(p, cl); 262 return leaf; 263} 264 265/* Find queueing discipline by name */ 266 267static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 268{ 269 struct Qdisc_ops *q = NULL; 270 271 if (kind) { 272 read_lock(&qdisc_mod_lock); 273 for (q = qdisc_base; q; q = q->next) { 274 if (nla_strcmp(kind, q->id) == 0) { 275 if (!try_module_get(q->owner)) 276 q = NULL; 277 break; 278 } 279 } 280 read_unlock(&qdisc_mod_lock); 281 } 282 return q; 283} 284 285static struct qdisc_rate_table *qdisc_rtab_list; 286 287struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) 288{ 289 struct qdisc_rate_table *rtab; 290 291 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 292 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { 293 rtab->refcnt++; 294 return rtab; 295 } 296 } 297 298 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 299 nla_len(tab) != TC_RTAB_SIZE) 300 return NULL; 301 302 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 303 if (rtab) { 304 rtab->rate = *r; 305 rtab->refcnt = 1; 306 memcpy(rtab->data, nla_data(tab), 1024); 307 rtab->next = qdisc_rtab_list; 308 qdisc_rtab_list = rtab; 309 } 310 return rtab; 311} 312EXPORT_SYMBOL(qdisc_get_rtab); 313 314void qdisc_put_rtab(struct qdisc_rate_table *tab) 315{ 316 struct qdisc_rate_table *rtab, **rtabp; 317 318 if (!tab || --tab->refcnt) 319 return; 320 321 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { 322 if (rtab == tab) { 323 *rtabp = rtab->next; 324 kfree(rtab); 325 return; 326 } 327 } 328} 329EXPORT_SYMBOL(qdisc_put_rtab); 330 331static LIST_HEAD(qdisc_stab_list); 332static DEFINE_SPINLOCK(qdisc_stab_lock); 333 334static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { 335 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, 336 [TCA_STAB_DATA] = { .type = NLA_BINARY }, 337}; 338 339static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) 340{ 341 struct nlattr *tb[TCA_STAB_MAX + 1]; 342 struct qdisc_size_table *stab; 343 struct tc_sizespec *s; 344 unsigned int tsize = 0; 345 u16 *tab = NULL; 346 int err; 347 348 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); 349 if (err < 0) 350 return ERR_PTR(err); 351 if (!tb[TCA_STAB_BASE]) 352 return ERR_PTR(-EINVAL); 353 354 s = nla_data(tb[TCA_STAB_BASE]); 355 356 if (s->tsize > 0) { 357 if (!tb[TCA_STAB_DATA]) 358 return ERR_PTR(-EINVAL); 359 tab = nla_data(tb[TCA_STAB_DATA]); 360 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); 361 } 362 363 if (!s || tsize != s->tsize || (!tab && tsize > 0)) 364 return ERR_PTR(-EINVAL); 365 366 spin_lock(&qdisc_stab_lock); 367 368 list_for_each_entry(stab, &qdisc_stab_list, list) { 369 if (memcmp(&stab->szopts, s, sizeof(*s))) 370 continue; 371 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) 372 continue; 373 stab->refcnt++; 374 spin_unlock(&qdisc_stab_lock); 375 return stab; 376 } 377 378 spin_unlock(&qdisc_stab_lock); 379 380 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); 381 if (!stab) 382 return ERR_PTR(-ENOMEM); 383 384 stab->refcnt = 1; 385 stab->szopts = *s; 386 if (tsize > 0) 387 memcpy(stab->data, tab, tsize * sizeof(u16)); 388 389 spin_lock(&qdisc_stab_lock); 390 list_add_tail(&stab->list, &qdisc_stab_list); 391 spin_unlock(&qdisc_stab_lock); 392 393 return stab; 394} 395 396void qdisc_put_stab(struct qdisc_size_table *tab) 397{ 398 if (!tab) 399 return; 400 401 spin_lock(&qdisc_stab_lock); 402 403 if (--tab->refcnt == 0) { 404 list_del(&tab->list); 405 kfree(tab); 406 } 407 408 spin_unlock(&qdisc_stab_lock); 409} 410EXPORT_SYMBOL(qdisc_put_stab); 411 412static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) 413{ 414 struct nlattr *nest; 415 416 nest = nla_nest_start(skb, TCA_STAB); 417 if (nest == NULL) 418 goto nla_put_failure; 419 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts); 420 nla_nest_end(skb, nest); 421 422 return skb->len; 423 424nla_put_failure: 425 return -1; 426} 427 428void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab) 429{ 430 int pkt_len, slot; 431 432 pkt_len = skb->len + stab->szopts.overhead; 433 if (unlikely(!stab->szopts.tsize)) 434 goto out; 435 436 slot = pkt_len + stab->szopts.cell_align; 437 if (unlikely(slot < 0)) 438 slot = 0; 439 440 slot >>= stab->szopts.cell_log; 441 if (likely(slot < stab->szopts.tsize)) 442 pkt_len = stab->data[slot]; 443 else 444 pkt_len = stab->data[stab->szopts.tsize - 1] * 445 (slot / stab->szopts.tsize) + 446 stab->data[slot % stab->szopts.tsize]; 447 448 pkt_len <<= stab->szopts.size_log; 449out: 450 if (unlikely(pkt_len < 1)) 451 pkt_len = 1; 452 qdisc_skb_cb(skb)->pkt_len = pkt_len; 453} 454EXPORT_SYMBOL(qdisc_calculate_pkt_len); 455 456void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc) 457{ 458 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { 459 printk(KERN_WARNING 460 "%s: %s qdisc %X: is non-work-conserving?\n", 461 txt, qdisc->ops->id, qdisc->handle >> 16); 462 qdisc->flags |= TCQ_F_WARN_NONWC; 463 } 464} 465EXPORT_SYMBOL(qdisc_warn_nonwc); 466 467static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 468{ 469 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 470 timer); 471 472 wd->qdisc->flags &= ~TCQ_F_THROTTLED; 473 __netif_schedule(qdisc_root(wd->qdisc)); 474 475 return HRTIMER_NORESTART; 476} 477 478void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 479{ 480 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 481 wd->timer.function = qdisc_watchdog; 482 wd->qdisc = qdisc; 483} 484EXPORT_SYMBOL(qdisc_watchdog_init); 485 486void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) 487{ 488 ktime_t time; 489 490 if (test_bit(__QDISC_STATE_DEACTIVATED, 491 &qdisc_root_sleeping(wd->qdisc)->state)) 492 return; 493 494 wd->qdisc->flags |= TCQ_F_THROTTLED; 495 time = ktime_set(0, 0); 496 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires)); 497 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); 498} 499EXPORT_SYMBOL(qdisc_watchdog_schedule); 500 501void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 502{ 503 hrtimer_cancel(&wd->timer); 504 wd->qdisc->flags &= ~TCQ_F_THROTTLED; 505} 506EXPORT_SYMBOL(qdisc_watchdog_cancel); 507 508static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) 509{ 510 unsigned int size = n * sizeof(struct hlist_head), i; 511 struct hlist_head *h; 512 513 if (size <= PAGE_SIZE) 514 h = kmalloc(size, GFP_KERNEL); 515 else 516 h = (struct hlist_head *) 517 __get_free_pages(GFP_KERNEL, get_order(size)); 518 519 if (h != NULL) { 520 for (i = 0; i < n; i++) 521 INIT_HLIST_HEAD(&h[i]); 522 } 523 return h; 524} 525 526static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) 527{ 528 unsigned int size = n * sizeof(struct hlist_head); 529 530 if (size <= PAGE_SIZE) 531 kfree(h); 532 else 533 free_pages((unsigned long)h, get_order(size)); 534} 535 536void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) 537{ 538 struct Qdisc_class_common *cl; 539 struct hlist_node *n, *next; 540 struct hlist_head *nhash, *ohash; 541 unsigned int nsize, nmask, osize; 542 unsigned int i, h; 543 544 /* Rehash when load factor exceeds 0.75 */ 545 if (clhash->hashelems * 4 <= clhash->hashsize * 3) 546 return; 547 nsize = clhash->hashsize * 2; 548 nmask = nsize - 1; 549 nhash = qdisc_class_hash_alloc(nsize); 550 if (nhash == NULL) 551 return; 552 553 ohash = clhash->hash; 554 osize = clhash->hashsize; 555 556 sch_tree_lock(sch); 557 for (i = 0; i < osize; i++) { 558 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { 559 h = qdisc_class_hash(cl->classid, nmask); 560 hlist_add_head(&cl->hnode, &nhash[h]); 561 } 562 } 563 clhash->hash = nhash; 564 clhash->hashsize = nsize; 565 clhash->hashmask = nmask; 566 sch_tree_unlock(sch); 567 568 qdisc_class_hash_free(ohash, osize); 569} 570EXPORT_SYMBOL(qdisc_class_hash_grow); 571 572int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) 573{ 574 unsigned int size = 4; 575 576 clhash->hash = qdisc_class_hash_alloc(size); 577 if (clhash->hash == NULL) 578 return -ENOMEM; 579 clhash->hashsize = size; 580 clhash->hashmask = size - 1; 581 clhash->hashelems = 0; 582 return 0; 583} 584EXPORT_SYMBOL(qdisc_class_hash_init); 585 586void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) 587{ 588 qdisc_class_hash_free(clhash->hash, clhash->hashsize); 589} 590EXPORT_SYMBOL(qdisc_class_hash_destroy); 591 592void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, 593 struct Qdisc_class_common *cl) 594{ 595 unsigned int h; 596 597 INIT_HLIST_NODE(&cl->hnode); 598 h = qdisc_class_hash(cl->classid, clhash->hashmask); 599 hlist_add_head(&cl->hnode, &clhash->hash[h]); 600 clhash->hashelems++; 601} 602EXPORT_SYMBOL(qdisc_class_hash_insert); 603 604void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, 605 struct Qdisc_class_common *cl) 606{ 607 hlist_del(&cl->hnode); 608 clhash->hashelems--; 609} 610EXPORT_SYMBOL(qdisc_class_hash_remove); 611 612/* Allocate an unique handle from space managed by kernel */ 613 614static u32 qdisc_alloc_handle(struct net_device *dev) 615{ 616 int i = 0x10000; 617 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 618 619 do { 620 autohandle += TC_H_MAKE(0x10000U, 0); 621 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 622 autohandle = TC_H_MAKE(0x80000000U, 0); 623 } while (qdisc_lookup(dev, autohandle) && --i > 0); 624 625 return i>0 ? autohandle : 0; 626} 627 628void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) 629{ 630 const struct Qdisc_class_ops *cops; 631 unsigned long cl; 632 u32 parentid; 633 634 if (n == 0) 635 return; 636 while ((parentid = sch->parent)) { 637 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 638 return; 639 640 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); 641 if (sch == NULL) { 642 WARN_ON(parentid != TC_H_ROOT); 643 return; 644 } 645 cops = sch->ops->cl_ops; 646 if (cops->qlen_notify) { 647 cl = cops->get(sch, parentid); 648 cops->qlen_notify(sch, cl); 649 cops->put(sch, cl); 650 } 651 sch->q.qlen -= n; 652 } 653} 654EXPORT_SYMBOL(qdisc_tree_decrease_qlen); 655 656static void notify_and_destroy(struct net *net, struct sk_buff *skb, 657 struct nlmsghdr *n, u32 clid, 658 struct Qdisc *old, struct Qdisc *new) 659{ 660 if (new || old) 661 qdisc_notify(net, skb, n, clid, old, new); 662 663 if (old) 664 qdisc_destroy(old); 665} 666 667/* Graft qdisc "new" to class "classid" of qdisc "parent" or 668 * to device "dev". 669 * 670 * When appropriate send a netlink notification using 'skb' 671 * and "n". 672 * 673 * On success, destroy old qdisc. 674 */ 675 676static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 677 struct sk_buff *skb, struct nlmsghdr *n, u32 classid, 678 struct Qdisc *new, struct Qdisc *old) 679{ 680 struct Qdisc *q = old; 681 struct net *net = dev_net(dev); 682 int err = 0; 683 684 if (parent == NULL) { 685 unsigned int i, num_q, ingress; 686 687 ingress = 0; 688 num_q = dev->num_tx_queues; 689 if ((q && q->flags & TCQ_F_INGRESS) || 690 (new && new->flags & TCQ_F_INGRESS)) { 691 num_q = 1; 692 ingress = 1; 693 } 694 695 if (dev->flags & IFF_UP) 696 dev_deactivate(dev); 697 698 if (new && new->ops->attach) { 699 new->ops->attach(new); 700 num_q = 0; 701 } 702 703 for (i = 0; i < num_q; i++) { 704 struct netdev_queue *dev_queue = &dev->rx_queue; 705 706 if (!ingress) 707 dev_queue = netdev_get_tx_queue(dev, i); 708 709 old = dev_graft_qdisc(dev_queue, new); 710 if (new && i > 0) 711 atomic_inc(&new->refcnt); 712 713 if (!ingress) 714 qdisc_destroy(old); 715 } 716 717 if (!ingress) { 718 notify_and_destroy(net, skb, n, classid, 719 dev->qdisc, new); 720 if (new && !new->ops->attach) 721 atomic_inc(&new->refcnt); 722 dev->qdisc = new ? : &noop_qdisc; 723 } else { 724 notify_and_destroy(net, skb, n, classid, old, new); 725 } 726 727 if (dev->flags & IFF_UP) 728 dev_activate(dev); 729 } else { 730 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 731 732 err = -EOPNOTSUPP; 733 if (cops && cops->graft) { 734 unsigned long cl = cops->get(parent, classid); 735 if (cl) { 736 err = cops->graft(parent, cl, new, &old); 737 cops->put(parent, cl); 738 } else 739 err = -ENOENT; 740 } 741 if (!err) 742 notify_and_destroy(net, skb, n, classid, old, new); 743 } 744 return err; 745} 746 747/* lockdep annotation is needed for ingress; egress gets it only for name */ 748static struct lock_class_key qdisc_tx_lock; 749static struct lock_class_key qdisc_rx_lock; 750 751/* 752 Allocate and initialize new qdisc. 753 754 Parameters are passed via opt. 755 */ 756 757static struct Qdisc * 758qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, 759 struct Qdisc *p, u32 parent, u32 handle, 760 struct nlattr **tca, int *errp) 761{ 762 int err; 763 struct nlattr *kind = tca[TCA_KIND]; 764 struct Qdisc *sch; 765 struct Qdisc_ops *ops; 766 struct qdisc_size_table *stab; 767 768 ops = qdisc_lookup_ops(kind); 769#ifdef CONFIG_MODULES 770 if (ops == NULL && kind != NULL) { 771 char name[IFNAMSIZ]; 772 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 773 /* We dropped the RTNL semaphore in order to 774 * perform the module load. So, even if we 775 * succeeded in loading the module we have to 776 * tell the caller to replay the request. We 777 * indicate this using -EAGAIN. 778 * We replay the request because the device may 779 * go away in the mean time. 780 */ 781 rtnl_unlock(); 782 request_module("sch_%s", name); 783 rtnl_lock(); 784 ops = qdisc_lookup_ops(kind); 785 if (ops != NULL) { 786 /* We will try again qdisc_lookup_ops, 787 * so don't keep a reference. 788 */ 789 module_put(ops->owner); 790 err = -EAGAIN; 791 goto err_out; 792 } 793 } 794 } 795#endif 796 797 err = -ENOENT; 798 if (ops == NULL) 799 goto err_out; 800 801 sch = qdisc_alloc(dev_queue, ops); 802 if (IS_ERR(sch)) { 803 err = PTR_ERR(sch); 804 goto err_out2; 805 } 806 807 sch->parent = parent; 808 809 if (handle == TC_H_INGRESS) { 810 sch->flags |= TCQ_F_INGRESS; 811 handle = TC_H_MAKE(TC_H_INGRESS, 0); 812 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); 813 } else { 814 if (handle == 0) { 815 handle = qdisc_alloc_handle(dev); 816 err = -ENOMEM; 817 if (handle == 0) 818 goto err_out3; 819 } 820 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); 821 } 822 823 sch->handle = handle; 824 825 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 826 if (tca[TCA_STAB]) { 827 stab = qdisc_get_stab(tca[TCA_STAB]); 828 if (IS_ERR(stab)) { 829 err = PTR_ERR(stab); 830 goto err_out4; 831 } 832 sch->stab = stab; 833 } 834 if (tca[TCA_RATE]) { 835 spinlock_t *root_lock; 836 837 err = -EOPNOTSUPP; 838 if (sch->flags & TCQ_F_MQROOT) 839 goto err_out4; 840 841 if ((sch->parent != TC_H_ROOT) && 842 !(sch->flags & TCQ_F_INGRESS) && 843 (!p || !(p->flags & TCQ_F_MQROOT))) 844 root_lock = qdisc_root_sleeping_lock(sch); 845 else 846 root_lock = qdisc_lock(sch); 847 848 err = gen_new_estimator(&sch->bstats, &sch->rate_est, 849 root_lock, tca[TCA_RATE]); 850 if (err) 851 goto err_out4; 852 } 853 854 qdisc_list_add(sch); 855 856 return sch; 857 } 858err_out3: 859 dev_put(dev); 860 kfree((char *) sch - sch->padded); 861err_out2: 862 module_put(ops->owner); 863err_out: 864 *errp = err; 865 return NULL; 866 867err_out4: 868 /* 869 * Any broken qdiscs that would require a ops->reset() here? 870 * The qdisc was never in action so it shouldn't be necessary. 871 */ 872 qdisc_put_stab(sch->stab); 873 if (ops->destroy) 874 ops->destroy(sch); 875 goto err_out3; 876} 877 878static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 879{ 880 struct qdisc_size_table *stab = NULL; 881 int err = 0; 882 883 if (tca[TCA_OPTIONS]) { 884 if (sch->ops->change == NULL) 885 return -EINVAL; 886 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 887 if (err) 888 return err; 889 } 890 891 if (tca[TCA_STAB]) { 892 stab = qdisc_get_stab(tca[TCA_STAB]); 893 if (IS_ERR(stab)) 894 return PTR_ERR(stab); 895 } 896 897 qdisc_put_stab(sch->stab); 898 sch->stab = stab; 899 900 if (tca[TCA_RATE]) { 901 /* NB: ignores errors from replace_estimator 902 because change can't be undone. */ 903 if (sch->flags & TCQ_F_MQROOT) 904 goto out; 905 gen_replace_estimator(&sch->bstats, &sch->rate_est, 906 qdisc_root_sleeping_lock(sch), 907 tca[TCA_RATE]); 908 } 909out: 910 return 0; 911} 912 913struct check_loop_arg 914{ 915 struct qdisc_walker w; 916 struct Qdisc *p; 917 int depth; 918}; 919 920static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); 921 922static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 923{ 924 struct check_loop_arg arg; 925 926 if (q->ops->cl_ops == NULL) 927 return 0; 928 929 arg.w.stop = arg.w.skip = arg.w.count = 0; 930 arg.w.fn = check_loop_fn; 931 arg.depth = depth; 932 arg.p = p; 933 q->ops->cl_ops->walk(q, &arg.w); 934 return arg.w.stop ? -ELOOP : 0; 935} 936 937static int 938check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 939{ 940 struct Qdisc *leaf; 941 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 942 struct check_loop_arg *arg = (struct check_loop_arg *)w; 943 944 leaf = cops->leaf(q, cl); 945 if (leaf) { 946 if (leaf == arg->p || arg->depth > 7) 947 return -ELOOP; 948 return check_loop(leaf, arg->p, arg->depth + 1); 949 } 950 return 0; 951} 952 953/* 954 * Delete/get qdisc. 955 */ 956 957static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 958{ 959 struct net *net = sock_net(skb->sk); 960 struct tcmsg *tcm = NLMSG_DATA(n); 961 struct nlattr *tca[TCA_MAX + 1]; 962 struct net_device *dev; 963 u32 clid = tcm->tcm_parent; 964 struct Qdisc *q = NULL; 965 struct Qdisc *p = NULL; 966 int err; 967 968 if ((dev = __dev_get_by_index(net, tcm->tcm_ifindex)) == NULL) 969 return -ENODEV; 970 971 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 972 if (err < 0) 973 return err; 974 975 if (clid) { 976 if (clid != TC_H_ROOT) { 977 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 978 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) 979 return -ENOENT; 980 q = qdisc_leaf(p, clid); 981 } else { /* ingress */ 982 q = dev->rx_queue.qdisc_sleeping; 983 } 984 } else { 985 q = dev->qdisc; 986 } 987 if (!q) 988 return -ENOENT; 989 990 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 991 return -EINVAL; 992 } else { 993 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) 994 return -ENOENT; 995 } 996 997 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 998 return -EINVAL; 999 1000 if (n->nlmsg_type == RTM_DELQDISC) { 1001 if (!clid) 1002 return -EINVAL; 1003 if (q->handle == 0) 1004 return -ENOENT; 1005 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0) 1006 return err; 1007 } else { 1008 qdisc_notify(net, skb, n, clid, NULL, q); 1009 } 1010 return 0; 1011} 1012 1013/* 1014 Create/change qdisc. 1015 */ 1016 1017static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 1018{ 1019 struct net *net = sock_net(skb->sk); 1020 struct tcmsg *tcm; 1021 struct nlattr *tca[TCA_MAX + 1]; 1022 struct net_device *dev; 1023 u32 clid; 1024 struct Qdisc *q, *p; 1025 int err; 1026 1027replay: 1028 /* Reinit, just in case something touches this. */ 1029 tcm = NLMSG_DATA(n); 1030 clid = tcm->tcm_parent; 1031 q = p = NULL; 1032 1033 if ((dev = __dev_get_by_index(net, tcm->tcm_ifindex)) == NULL) 1034 return -ENODEV; 1035 1036 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1037 if (err < 0) 1038 return err; 1039 1040 if (clid) { 1041 if (clid != TC_H_ROOT) { 1042 if (clid != TC_H_INGRESS) { 1043 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) 1044 return -ENOENT; 1045 q = qdisc_leaf(p, clid); 1046 } else { /*ingress */ 1047 q = dev->rx_queue.qdisc_sleeping; 1048 } 1049 } else { 1050 q = dev->qdisc; 1051 } 1052 1053 /* It may be default qdisc, ignore it */ 1054 if (q && q->handle == 0) 1055 q = NULL; 1056 1057 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 1058 if (tcm->tcm_handle) { 1059 if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) 1060 return -EEXIST; 1061 if (TC_H_MIN(tcm->tcm_handle)) 1062 return -EINVAL; 1063 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) 1064 goto create_n_graft; 1065 if (n->nlmsg_flags&NLM_F_EXCL) 1066 return -EEXIST; 1067 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1068 return -EINVAL; 1069 if (q == p || 1070 (p && check_loop(q, p, 0))) 1071 return -ELOOP; 1072 atomic_inc(&q->refcnt); 1073 goto graft; 1074 } else { 1075 if (q == NULL) 1076 goto create_n_graft; 1077 1078 /* This magic test requires explanation. 1079 * 1080 * We know, that some child q is already 1081 * attached to this parent and have choice: 1082 * either to change it or to create/graft new one. 1083 * 1084 * 1. We are allowed to create/graft only 1085 * if CREATE and REPLACE flags are set. 1086 * 1087 * 2. If EXCL is set, requestor wanted to say, 1088 * that qdisc tcm_handle is not expected 1089 * to exist, so that we choose create/graft too. 1090 * 1091 * 3. The last case is when no flags are set. 1092 * Alas, it is sort of hole in API, we 1093 * cannot decide what to do unambiguously. 1094 * For now we select create/graft, if 1095 * user gave KIND, which does not match existing. 1096 */ 1097 if ((n->nlmsg_flags&NLM_F_CREATE) && 1098 (n->nlmsg_flags&NLM_F_REPLACE) && 1099 ((n->nlmsg_flags&NLM_F_EXCL) || 1100 (tca[TCA_KIND] && 1101 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 1102 goto create_n_graft; 1103 } 1104 } 1105 } else { 1106 if (!tcm->tcm_handle) 1107 return -EINVAL; 1108 q = qdisc_lookup(dev, tcm->tcm_handle); 1109 } 1110 1111 /* Change qdisc parameters */ 1112 if (q == NULL) 1113 return -ENOENT; 1114 if (n->nlmsg_flags&NLM_F_EXCL) 1115 return -EEXIST; 1116 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 1117 return -EINVAL; 1118 err = qdisc_change(q, tca); 1119 if (err == 0) 1120 qdisc_notify(net, skb, n, clid, NULL, q); 1121 return err; 1122 1123create_n_graft: 1124 if (!(n->nlmsg_flags&NLM_F_CREATE)) 1125 return -ENOENT; 1126 if (clid == TC_H_INGRESS) 1127 q = qdisc_create(dev, &dev->rx_queue, p, 1128 tcm->tcm_parent, tcm->tcm_parent, 1129 tca, &err); 1130 else { 1131 struct netdev_queue *dev_queue; 1132 1133 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue) 1134 dev_queue = p->ops->cl_ops->select_queue(p, tcm); 1135 else if (p) 1136 dev_queue = p->dev_queue; 1137 else 1138 dev_queue = netdev_get_tx_queue(dev, 0); 1139 1140 q = qdisc_create(dev, dev_queue, p, 1141 tcm->tcm_parent, tcm->tcm_handle, 1142 tca, &err); 1143 } 1144 if (q == NULL) { 1145 if (err == -EAGAIN) 1146 goto replay; 1147 return err; 1148 } 1149 1150graft: 1151 err = qdisc_graft(dev, p, skb, n, clid, q, NULL); 1152 if (err) { 1153 if (q) 1154 qdisc_destroy(q); 1155 return err; 1156 } 1157 1158 return 0; 1159} 1160 1161static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 1162 u32 pid, u32 seq, u16 flags, int event) 1163{ 1164 struct tcmsg *tcm; 1165 struct nlmsghdr *nlh; 1166 unsigned char *b = skb_tail_pointer(skb); 1167 struct gnet_dump d; 1168 1169 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 1170 tcm = NLMSG_DATA(nlh); 1171 tcm->tcm_family = AF_UNSPEC; 1172 tcm->tcm__pad1 = 0; 1173 tcm->tcm__pad2 = 0; 1174 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1175 tcm->tcm_parent = clid; 1176 tcm->tcm_handle = q->handle; 1177 tcm->tcm_info = atomic_read(&q->refcnt); 1178 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); 1179 if (q->ops->dump && q->ops->dump(q, skb) < 0) 1180 goto nla_put_failure; 1181 q->qstats.qlen = q->q.qlen; 1182 1183 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0) 1184 goto nla_put_failure; 1185 1186 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1187 qdisc_root_sleeping_lock(q), &d) < 0) 1188 goto nla_put_failure; 1189 1190 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 1191 goto nla_put_failure; 1192 1193 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || 1194 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 || 1195 gnet_stats_copy_queue(&d, &q->qstats) < 0) 1196 goto nla_put_failure; 1197 1198 if (gnet_stats_finish_copy(&d) < 0) 1199 goto nla_put_failure; 1200 1201 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1202 return skb->len; 1203 1204nlmsg_failure: 1205nla_put_failure: 1206 nlmsg_trim(skb, b); 1207 return -1; 1208} 1209 1210static bool tc_qdisc_dump_ignore(struct Qdisc *q) 1211{ 1212 return (q->flags & TCQ_F_BUILTIN) ? true : false; 1213} 1214 1215static int qdisc_notify(struct net *net, struct sk_buff *oskb, 1216 struct nlmsghdr *n, u32 clid, 1217 struct Qdisc *old, struct Qdisc *new) 1218{ 1219 struct sk_buff *skb; 1220 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 1221 1222 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1223 if (!skb) 1224 return -ENOBUFS; 1225 1226 if (old && !tc_qdisc_dump_ignore(old)) { 1227 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) 1228 goto err_out; 1229 } 1230 if (new && !tc_qdisc_dump_ignore(new)) { 1231 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 1232 goto err_out; 1233 } 1234 1235 if (skb->len) 1236 return rtnetlink_send(skb, net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 1237 1238err_out: 1239 kfree_skb(skb); 1240 return -EINVAL; 1241} 1242 1243static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, 1244 struct netlink_callback *cb, 1245 int *q_idx_p, int s_q_idx) 1246{ 1247 int ret = 0, q_idx = *q_idx_p; 1248 struct Qdisc *q; 1249 1250 if (!root) 1251 return 0; 1252 1253 q = root; 1254 if (q_idx < s_q_idx) { 1255 q_idx++; 1256 } else { 1257 if (!tc_qdisc_dump_ignore(q) && 1258 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, 1259 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1260 goto done; 1261 q_idx++; 1262 } 1263 list_for_each_entry(q, &root->list, list) { 1264 if (q_idx < s_q_idx) { 1265 q_idx++; 1266 continue; 1267 } 1268 if (!tc_qdisc_dump_ignore(q) && 1269 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, 1270 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 1271 goto done; 1272 q_idx++; 1273 } 1274 1275out: 1276 *q_idx_p = q_idx; 1277 return ret; 1278done: 1279 ret = -1; 1280 goto out; 1281} 1282 1283static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 1284{ 1285 struct net *net = sock_net(skb->sk); 1286 int idx, q_idx; 1287 int s_idx, s_q_idx; 1288 struct net_device *dev; 1289 1290 s_idx = cb->args[0]; 1291 s_q_idx = q_idx = cb->args[1]; 1292 1293 rcu_read_lock(); 1294 idx = 0; 1295 for_each_netdev_rcu(net, dev) { 1296 struct netdev_queue *dev_queue; 1297 1298 if (idx < s_idx) 1299 goto cont; 1300 if (idx > s_idx) 1301 s_q_idx = 0; 1302 q_idx = 0; 1303 1304 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0) 1305 goto done; 1306 1307 dev_queue = &dev->rx_queue; 1308 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0) 1309 goto done; 1310 1311cont: 1312 idx++; 1313 } 1314 1315done: 1316 rcu_read_unlock(); 1317 1318 cb->args[0] = idx; 1319 cb->args[1] = q_idx; 1320 1321 return skb->len; 1322} 1323 1324 1325 1326/************************************************ 1327 * Traffic classes manipulation. * 1328 ************************************************/ 1329 1330 1331 1332static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 1333{ 1334 struct net *net = sock_net(skb->sk); 1335 struct tcmsg *tcm = NLMSG_DATA(n); 1336 struct nlattr *tca[TCA_MAX + 1]; 1337 struct net_device *dev; 1338 struct Qdisc *q = NULL; 1339 const struct Qdisc_class_ops *cops; 1340 unsigned long cl = 0; 1341 unsigned long new_cl; 1342 u32 pid = tcm->tcm_parent; 1343 u32 clid = tcm->tcm_handle; 1344 u32 qid = TC_H_MAJ(clid); 1345 int err; 1346 1347 if ((dev = __dev_get_by_index(net, tcm->tcm_ifindex)) == NULL) 1348 return -ENODEV; 1349 1350 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 1351 if (err < 0) 1352 return err; 1353 1354 /* 1355 parent == TC_H_UNSPEC - unspecified parent. 1356 parent == TC_H_ROOT - class is root, which has no parent. 1357 parent == X:0 - parent is root class. 1358 parent == X:Y - parent is a node in hierarchy. 1359 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 1360 1361 handle == 0:0 - generate handle from kernel pool. 1362 handle == 0:Y - class is X:Y, where X:0 is qdisc. 1363 handle == X:Y - clear. 1364 handle == X:0 - root class. 1365 */ 1366 1367 /* Step 1. Determine qdisc handle X:0 */ 1368 1369 if (pid != TC_H_ROOT) { 1370 u32 qid1 = TC_H_MAJ(pid); 1371 1372 if (qid && qid1) { 1373 /* If both majors are known, they must be identical. */ 1374 if (qid != qid1) 1375 return -EINVAL; 1376 } else if (qid1) { 1377 qid = qid1; 1378 } else if (qid == 0) 1379 qid = dev->qdisc->handle; 1380 1381 /* Now qid is genuine qdisc handle consistent 1382 both with parent and child. 1383 1384 TC_H_MAJ(pid) still may be unspecified, complete it now. 1385 */ 1386 if (pid) 1387 pid = TC_H_MAKE(qid, pid); 1388 } else { 1389 if (qid == 0) 1390 qid = dev->qdisc->handle; 1391 } 1392 1393 /* OK. Locate qdisc */ 1394 if ((q = qdisc_lookup(dev, qid)) == NULL) 1395 return -ENOENT; 1396 1397 /* An check that it supports classes */ 1398 cops = q->ops->cl_ops; 1399 if (cops == NULL) 1400 return -EINVAL; 1401 1402 /* Now try to get class */ 1403 if (clid == 0) { 1404 if (pid == TC_H_ROOT) 1405 clid = qid; 1406 } else 1407 clid = TC_H_MAKE(qid, clid); 1408 1409 if (clid) 1410 cl = cops->get(q, clid); 1411 1412 if (cl == 0) { 1413 err = -ENOENT; 1414 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) 1415 goto out; 1416 } else { 1417 switch (n->nlmsg_type) { 1418 case RTM_NEWTCLASS: 1419 err = -EEXIST; 1420 if (n->nlmsg_flags&NLM_F_EXCL) 1421 goto out; 1422 break; 1423 case RTM_DELTCLASS: 1424 err = -EOPNOTSUPP; 1425 if (cops->delete) 1426 err = cops->delete(q, cl); 1427 if (err == 0) 1428 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS); 1429 goto out; 1430 case RTM_GETTCLASS: 1431 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS); 1432 goto out; 1433 default: 1434 err = -EINVAL; 1435 goto out; 1436 } 1437 } 1438 1439 new_cl = cl; 1440 err = -EOPNOTSUPP; 1441 if (cops->change) 1442 err = cops->change(q, clid, pid, tca, &new_cl); 1443 if (err == 0) 1444 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS); 1445 1446out: 1447 if (cl) 1448 cops->put(q, cl); 1449 1450 return err; 1451} 1452 1453 1454static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1455 unsigned long cl, 1456 u32 pid, u32 seq, u16 flags, int event) 1457{ 1458 struct tcmsg *tcm; 1459 struct nlmsghdr *nlh; 1460 unsigned char *b = skb_tail_pointer(skb); 1461 struct gnet_dump d; 1462 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1463 1464 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 1465 tcm = NLMSG_DATA(nlh); 1466 tcm->tcm_family = AF_UNSPEC; 1467 tcm->tcm__pad1 = 0; 1468 tcm->tcm__pad2 = 0; 1469 tcm->tcm_ifindex = qdisc_dev(q)->ifindex; 1470 tcm->tcm_parent = q->handle; 1471 tcm->tcm_handle = q->handle; 1472 tcm->tcm_info = 0; 1473 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); 1474 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1475 goto nla_put_failure; 1476 1477 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, 1478 qdisc_root_sleeping_lock(q), &d) < 0) 1479 goto nla_put_failure; 1480 1481 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1482 goto nla_put_failure; 1483 1484 if (gnet_stats_finish_copy(&d) < 0) 1485 goto nla_put_failure; 1486 1487 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1488 return skb->len; 1489 1490nlmsg_failure: 1491nla_put_failure: 1492 nlmsg_trim(skb, b); 1493 return -1; 1494} 1495 1496static int tclass_notify(struct net *net, struct sk_buff *oskb, 1497 struct nlmsghdr *n, struct Qdisc *q, 1498 unsigned long cl, int event) 1499{ 1500 struct sk_buff *skb; 1501 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 1502 1503 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1504 if (!skb) 1505 return -ENOBUFS; 1506 1507 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { 1508 kfree_skb(skb); 1509 return -EINVAL; 1510 } 1511 1512 return rtnetlink_send(skb, net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 1513} 1514 1515struct qdisc_dump_args 1516{ 1517 struct qdisc_walker w; 1518 struct sk_buff *skb; 1519 struct netlink_callback *cb; 1520}; 1521 1522static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) 1523{ 1524 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1525 1526 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, 1527 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); 1528} 1529 1530static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, 1531 struct tcmsg *tcm, struct netlink_callback *cb, 1532 int *t_p, int s_t) 1533{ 1534 struct qdisc_dump_args arg; 1535 1536 if (tc_qdisc_dump_ignore(q) || 1537 *t_p < s_t || !q->ops->cl_ops || 1538 (tcm->tcm_parent && 1539 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1540 (*t_p)++; 1541 return 0; 1542 } 1543 if (*t_p > s_t) 1544 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1545 arg.w.fn = qdisc_class_dump; 1546 arg.skb = skb; 1547 arg.cb = cb; 1548 arg.w.stop = 0; 1549 arg.w.skip = cb->args[1]; 1550 arg.w.count = 0; 1551 q->ops->cl_ops->walk(q, &arg.w); 1552 cb->args[1] = arg.w.count; 1553 if (arg.w.stop) 1554 return -1; 1555 (*t_p)++; 1556 return 0; 1557} 1558 1559static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, 1560 struct tcmsg *tcm, struct netlink_callback *cb, 1561 int *t_p, int s_t) 1562{ 1563 struct Qdisc *q; 1564 1565 if (!root) 1566 return 0; 1567 1568 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) 1569 return -1; 1570 1571 list_for_each_entry(q, &root->list, list) { 1572 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) 1573 return -1; 1574 } 1575 1576 return 0; 1577} 1578 1579static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1580{ 1581 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); 1582 struct net *net = sock_net(skb->sk); 1583 struct netdev_queue *dev_queue; 1584 struct net_device *dev; 1585 int t, s_t; 1586 1587 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) 1588 return 0; 1589 if ((dev = dev_get_by_index(net, tcm->tcm_ifindex)) == NULL) 1590 return 0; 1591 1592 s_t = cb->args[0]; 1593 t = 0; 1594 1595 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) 1596 goto done; 1597 1598 dev_queue = &dev->rx_queue; 1599 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0) 1600 goto done; 1601 1602done: 1603 cb->args[0] = t; 1604 1605 dev_put(dev); 1606 return skb->len; 1607} 1608 1609/* Main classifier routine: scans classifier chain attached 1610 to this qdisc, (optionally) tests for protocol and asks 1611 specific classifiers. 1612 */ 1613int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, 1614 struct tcf_result *res) 1615{ 1616 __be16 protocol = skb->protocol; 1617 int err = 0; 1618 1619 for (; tp; tp = tp->next) { 1620 if ((tp->protocol == protocol || 1621 tp->protocol == htons(ETH_P_ALL)) && 1622 (err = tp->classify(skb, tp, res)) >= 0) { 1623#ifdef CONFIG_NET_CLS_ACT 1624 if (err != TC_ACT_RECLASSIFY && skb->tc_verd) 1625 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); 1626#endif 1627 return err; 1628 } 1629 } 1630 return -1; 1631} 1632EXPORT_SYMBOL(tc_classify_compat); 1633 1634int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, 1635 struct tcf_result *res) 1636{ 1637 int err = 0; 1638 __be16 protocol; 1639#ifdef CONFIG_NET_CLS_ACT 1640 struct tcf_proto *otp = tp; 1641reclassify: 1642#endif 1643 protocol = skb->protocol; 1644 1645 err = tc_classify_compat(skb, tp, res); 1646#ifdef CONFIG_NET_CLS_ACT 1647 if (err == TC_ACT_RECLASSIFY) { 1648 u32 verd = G_TC_VERD(skb->tc_verd); 1649 tp = otp; 1650 1651 if (verd++ >= MAX_REC_LOOP) { 1652 if (net_ratelimit()) 1653 printk(KERN_NOTICE 1654 "%s: packet reclassify loop" 1655 " rule prio %u protocol %02x\n", 1656 tp->q->ops->id, 1657 tp->prio & 0xffff, ntohs(tp->protocol)); 1658 return TC_ACT_SHOT; 1659 } 1660 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); 1661 goto reclassify; 1662 } 1663#endif 1664 return err; 1665} 1666EXPORT_SYMBOL(tc_classify); 1667 1668void tcf_destroy(struct tcf_proto *tp) 1669{ 1670 tp->ops->destroy(tp); 1671 module_put(tp->ops->owner); 1672 kfree(tp); 1673} 1674 1675void tcf_destroy_chain(struct tcf_proto **fl) 1676{ 1677 struct tcf_proto *tp; 1678 1679 while ((tp = *fl) != NULL) { 1680 *fl = tp->next; 1681 tcf_destroy(tp); 1682 } 1683} 1684EXPORT_SYMBOL(tcf_destroy_chain); 1685 1686#ifdef CONFIG_PROC_FS 1687static int psched_show(struct seq_file *seq, void *v) 1688{ 1689 struct timespec ts; 1690 1691 hrtimer_get_res(CLOCK_MONOTONIC, &ts); 1692 seq_printf(seq, "%08x %08x %08x %08x\n", 1693 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), 1694 1000000, 1695 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); 1696 1697 return 0; 1698} 1699 1700static int psched_open(struct inode *inode, struct file *file) 1701{ 1702 return single_open(file, psched_show, NULL); 1703} 1704 1705static const struct file_operations psched_fops = { 1706 .owner = THIS_MODULE, 1707 .open = psched_open, 1708 .read = seq_read, 1709 .llseek = seq_lseek, 1710 .release = single_release, 1711}; 1712 1713static int __net_init psched_net_init(struct net *net) 1714{ 1715 struct proc_dir_entry *e; 1716 1717 e = proc_net_fops_create(net, "psched", 0, &psched_fops); 1718 if (e == NULL) 1719 return -ENOMEM; 1720 1721 return 0; 1722} 1723 1724static void __net_exit psched_net_exit(struct net *net) 1725{ 1726 proc_net_remove(net, "psched"); 1727} 1728#else 1729static int __net_init psched_net_init(struct net *net) 1730{ 1731 return 0; 1732} 1733 1734static void __net_exit psched_net_exit(struct net *net) 1735{ 1736} 1737#endif 1738 1739static struct pernet_operations psched_net_ops = { 1740 .init = psched_net_init, 1741 .exit = psched_net_exit, 1742}; 1743 1744static int __init pktsched_init(void) 1745{ 1746 int err; 1747 1748 err = register_pernet_subsys(&psched_net_ops); 1749 if (err) { 1750 printk(KERN_ERR "pktsched_init: " 1751 "cannot initialize per netns operations\n"); 1752 return err; 1753 } 1754 1755 register_qdisc(&pfifo_qdisc_ops); 1756 register_qdisc(&bfifo_qdisc_ops); 1757 register_qdisc(&pfifo_head_drop_qdisc_ops); 1758 register_qdisc(&mq_qdisc_ops); 1759 1760 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); 1761 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); 1762 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); 1763 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); 1764 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); 1765 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); 1766 1767 return 0; 1768} 1769 1770subsys_initcall(pktsched_init); 1771