net/sched/sch_api.c

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