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
  33 #include <net/net_namespace.h>
  34 #include <net/sock.h>
  35 #include <net/netlink.h>
  36 #include <net/pkt_sched.h>
  37
  38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
  39                         struct nlmsghdr *n, u32 clid,
  40                         struct Qdisc *old, struct Qdisc *new);
  41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
  42                          struct nlmsghdr *n, struct Qdisc *q,
  43                          unsigned long cl, int event);
  44
  45 /*
  46
  47    Short review.
  48    -------------
  49
  50    This file consists of two interrelated parts:
  51
  52    1. queueing disciplines manager frontend.
  53    2. traffic classes manager frontend.
  54
  55    Generally, queueing discipline ("qdisc") is a black box,
  56    which is able to enqueue packets and to dequeue them (when
  57    device is ready to send something) in order and at times
  58    determined by algorithm hidden in it.
  59
  60    qdisc's are divided to two categories:
  61    - "queues", which have no internal structure visible from outside.
  62    - "schedulers", which split all the packets to "traffic classes",
  63      using "packet classifiers" (look at cls_api.c)
  64
  65    In turn, classes may have child qdiscs (as rule, queues)
  66    attached to them etc. etc. etc.
  67
  68    The goal of the routines in this file is to translate
  69    information supplied by user in the form of handles
  70    to more intelligible for kernel form, to make some sanity
  71    checks and part of work, which is common to all qdiscs
  72    and to provide rtnetlink notifications.
  73
  74    All real intelligent work is done inside qdisc modules.
  75
  76
  77
  78    Every discipline has two major routines: enqueue and dequeue.
  79
  80    ---dequeue
  81
  82    dequeue usually returns a skb to send. It is allowed to return NULL,
  83    but it does not mean that queue is empty, it just means that
  84    discipline does not want to send anything this time.
  85    Queue is really empty if q->q.qlen == 0.
  86    For complicated disciplines with multiple queues q->q is not
  87    real packet queue, but however q->q.qlen must be valid.
  88
  89    ---enqueue
  90
  91    enqueue returns 0, if packet was enqueued successfully.
  92    If packet (this one or another one) was dropped, it returns
  93    not zero error code.
  94    NET_XMIT_DROP        - this packet dropped
  95      Expected action: do not backoff, but wait until queue will clear.
  96    NET_XMIT_CN          - probably this packet enqueued, but another one dropped.
  97      Expected action: backoff or ignore
  98
  99    Auxiliary routines:
 100
 101    ---peek
 102
 103    like dequeue but without removing a packet from the queue
 104
 105    ---reset
 106
 107    returns qdisc to initial state: purge all buffers, clear all
 108    timers, counters (except for statistics) etc.
 109
 110    ---init
 111
 112    initializes newly created qdisc.
 113
 114    ---destroy
 115
 116    destroys resources allocated by init and during lifetime of qdisc.
 117
 118    ---change
 119
 120    changes qdisc parameters.
 121  */
 122
 123 /* Protects list of registered TC modules. It is pure SMP lock. */
 124 static DEFINE_RWLOCK(qdisc_mod_lock);
 125
 126
 127 /************************************************
 128  *      Queueing disciplines manipulation.      *
 129  ************************************************/
 130
 131
 132 /* The list of all installed queueing disciplines. */
 133
 134 static struct Qdisc_ops *qdisc_base;
 135
 136 /* Register/unregister queueing discipline */
 137
 138 int register_qdisc(struct Qdisc_ops *qops)
 139 {
 140         struct Qdisc_ops *q, **qp;
 141         int rc = -EEXIST;
 142
 143         write_lock(&qdisc_mod_lock);
 144         for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 145                 if (!strcmp(qops->id, q->id))
 146                         goto out;
 147
 148         if (qops->enqueue == NULL)
 149                 qops->enqueue = noop_qdisc_ops.enqueue;
 150         if (qops->peek == NULL) {
 151                 if (qops->dequeue == NULL)
 152                         qops->peek = noop_qdisc_ops.peek;
 153                 else
 154                         goto out_einval;
 155         }
 156         if (qops->dequeue == NULL)
 157                 qops->dequeue = noop_qdisc_ops.dequeue;
 158
 159         if (qops->cl_ops) {
 160                 const struct Qdisc_class_ops *cops = qops->cl_ops;
 161
 162                 if (!(cops->get && cops->put && cops->walk && cops->leaf))
 163                         goto out_einval;
 164
 165                 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
 166                         goto out_einval;
 167         }
 168
 169         qops->next = NULL;
 170         *qp = qops;
 171         rc = 0;
 172 out:
 173         write_unlock(&qdisc_mod_lock);
 174         return rc;
 175
 176 out_einval:
 177         rc = -EINVAL;
 178         goto out;
 179 }
 180 EXPORT_SYMBOL(register_qdisc);
 181
 182 int unregister_qdisc(struct Qdisc_ops *qops)
 183 {
 184         struct Qdisc_ops *q, **qp;
 185         int err = -ENOENT;
 186
 187         write_lock(&qdisc_mod_lock);
 188         for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 189                 if (q == qops)
 190                         break;
 191         if (q) {
 192                 *qp = q->next;
 193                 q->next = NULL;
 194                 err = 0;
 195         }
 196         write_unlock(&qdisc_mod_lock);
 197         return err;
 198 }
 199 EXPORT_SYMBOL(unregister_qdisc);
 200
 201 /* Get default qdisc if not otherwise specified */
 202 void qdisc_get_default(char *name, size_t len)
 203 {
 204         read_lock(&qdisc_mod_lock);
 205         strlcpy(name, default_qdisc_ops->id, len);
 206         read_unlock(&qdisc_mod_lock);
 207 }
 208
 209 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
 210 {
 211         struct Qdisc_ops *q = NULL;
 212
 213         for (q = qdisc_base; q; q = q->next) {
 214                 if (!strcmp(name, q->id)) {
 215                         if (!try_module_get(q->owner))
 216                                 q = NULL;
 217                         break;
 218                 }
 219         }
 220
 221         return q;
 222 }
 223
 224 /* Set new default qdisc to use */
 225 int qdisc_set_default(const char *name)
 226 {
 227         const struct Qdisc_ops *ops;
 228
 229         if (!capable(CAP_NET_ADMIN))
 230                 return -EPERM;
 231
 232         write_lock(&qdisc_mod_lock);
 233         ops = qdisc_lookup_default(name);
 234         if (!ops) {
 235                 /* Not found, drop lock and try to load module */
 236                 write_unlock(&qdisc_mod_lock);
 237                 request_module("sch_%s", name);
 238                 write_lock(&qdisc_mod_lock);
 239
 240                 ops = qdisc_lookup_default(name);
 241         }
 242
 243         if (ops) {
 244                 /* Set new default */
 245                 module_put(default_qdisc_ops->owner);
 246                 default_qdisc_ops = ops;
 247         }
 248         write_unlock(&qdisc_mod_lock);
 249
 250         return ops ? 0 : -ENOENT;
 251 }
 252
 253 /* We know handle. Find qdisc among all qdisc's attached to device
 254  * (root qdisc, all its children, children of children etc.)
 255  * Note: caller either uses rtnl or rcu_read_lock()
 256  */
 257
 258 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
 259 {
 260         struct Qdisc *q;
 261
 262         if (!(root->flags & TCQ_F_BUILTIN) &&
 263             root->handle == handle)
 264                 return root;
 265
 266         list_for_each_entry_rcu(q, &root->list, list) {
 267                 if (q->handle == handle)
 268                         return q;
 269         }
 270         return NULL;
 271 }
 272
 273 void qdisc_list_add(struct Qdisc *q)
 274 {
 275         if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 276                 struct Qdisc *root = qdisc_dev(q)->qdisc;
 277
 278                 WARN_ON_ONCE(root == &noop_qdisc);
 279                 ASSERT_RTNL();
 280                 list_add_tail_rcu(&q->list, &root->list);
 281         }
 282 }
 283 EXPORT_SYMBOL(qdisc_list_add);
 284
 285 void qdisc_list_del(struct Qdisc *q)
 286 {
 287         if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 288                 ASSERT_RTNL();
 289                 list_del_rcu(&q->list);
 290         }
 291 }
 292 EXPORT_SYMBOL(qdisc_list_del);
 293
 294 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
 295 {
 296         struct Qdisc *q;
 297
 298         q = qdisc_match_from_root(dev->qdisc, handle);
 299         if (q)
 300                 goto out;
 301
 302         if (dev_ingress_queue(dev))
 303                 q = qdisc_match_from_root(
 304                         dev_ingress_queue(dev)->qdisc_sleeping,
 305                         handle);
 306 out:
 307         return q;
 308 }
 309
 310 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
 311 {
 312         unsigned long cl;
 313         struct Qdisc *leaf;
 314         const struct Qdisc_class_ops *cops = p->ops->cl_ops;
 315
 316         if (cops == NULL)
 317                 return NULL;
 318         cl = cops->get(p, classid);
 319
 320         if (cl == 0)
 321                 return NULL;
 322         leaf = cops->leaf(p, cl);
 323         cops->put(p, cl);
 324         return leaf;
 325 }
 326
 327 /* Find queueing discipline by name */
 328
 329 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
 330 {
 331         struct Qdisc_ops *q = NULL;
 332
 333         if (kind) {
 334                 read_lock(&qdisc_mod_lock);
 335                 for (q = qdisc_base; q; q = q->next) {
 336                         if (nla_strcmp(kind, q->id) == 0) {
 337                                 if (!try_module_get(q->owner))
 338                                         q = NULL;
 339                                 break;
 340                         }
 341                 }
 342                 read_unlock(&qdisc_mod_lock);
 343         }
 344         return q;
 345 }
 346
 347 /* The linklayer setting were not transferred from iproute2, in older
 348  * versions, and the rate tables lookup systems have been dropped in
 349  * the kernel. To keep backward compatible with older iproute2 tc
 350  * utils, we detect the linklayer setting by detecting if the rate
 351  * table were modified.
 352  *
 353  * For linklayer ATM table entries, the rate table will be aligned to
 354  * 48 bytes, thus some table entries will contain the same value.  The
 355  * mpu (min packet unit) is also encoded into the old rate table, thus
 356  * starting from the mpu, we find low and high table entries for
 357  * mapping this cell.  If these entries contain the same value, when
 358  * the rate tables have been modified for linklayer ATM.
 359  *
 360  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
 361  * and then roundup to the next cell, calc the table entry one below,
 362  * and compare.
 363  */
 364 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
 365 {
 366         int low       = roundup(r->mpu, 48);
 367         int high      = roundup(low+1, 48);
 368         int cell_low  = low >> r->cell_log;
 369         int cell_high = (high >> r->cell_log) - 1;
 370
 371         /* rtab is too inaccurate at rates > 100Mbit/s */
 372         if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
 373                 pr_debug("TC linklayer: Giving up ATM detection\n");
 374                 return TC_LINKLAYER_ETHERNET;
 375         }
 376
 377         if ((cell_high > cell_low) && (cell_high < 256)
 378             && (rtab[cell_low] == rtab[cell_high])) {
 379                 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
 380                          cell_low, cell_high, rtab[cell_high]);
 381                 return TC_LINKLAYER_ATM;
 382         }
 383         return TC_LINKLAYER_ETHERNET;
 384 }
 385
 386 static struct qdisc_rate_table *qdisc_rtab_list;
 387
 388 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
 389 {
 390         struct qdisc_rate_table *rtab;
 391
 392         if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
 393             nla_len(tab) != TC_RTAB_SIZE)
 394                 return NULL;
 395
 396         for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
 397                 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
 398                     !memcmp(&rtab->data, nla_data(tab), 1024)) {
 399                         rtab->refcnt++;
 400                         return rtab;
 401                 }
 402         }
 403
 404         rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
 405         if (rtab) {
 406                 rtab->rate = *r;
 407                 rtab->refcnt = 1;
 408                 memcpy(rtab->data, nla_data(tab), 1024);
 409                 if (r->linklayer == TC_LINKLAYER_UNAWARE)
 410                         r->linklayer = __detect_linklayer(r, rtab->data);
 411                 rtab->next = qdisc_rtab_list;
 412                 qdisc_rtab_list = rtab;
 413         }
 414         return rtab;
 415 }
 416 EXPORT_SYMBOL(qdisc_get_rtab);
 417
 418 void qdisc_put_rtab(struct qdisc_rate_table *tab)
 419 {
 420         struct qdisc_rate_table *rtab, **rtabp;
 421
 422         if (!tab || --tab->refcnt)
 423                 return;
 424
 425         for (rtabp = &qdisc_rtab_list;
 426              (rtab = *rtabp) != NULL;
 427              rtabp = &rtab->next) {
 428                 if (rtab == tab) {
 429                         *rtabp = rtab->next;
 430                         kfree(rtab);
 431                         return;
 432                 }
 433         }
 434 }
 435 EXPORT_SYMBOL(qdisc_put_rtab);
 436
 437 static LIST_HEAD(qdisc_stab_list);
 438 static DEFINE_SPINLOCK(qdisc_stab_lock);
 439
 440 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
 441         [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
 442         [TCA_STAB_DATA] = { .type = NLA_BINARY },
 443 };
 444
 445 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
 446 {
 447         struct nlattr *tb[TCA_STAB_MAX + 1];
 448         struct qdisc_size_table *stab;
 449         struct tc_sizespec *s;
 450         unsigned int tsize = 0;
 451         u16 *tab = NULL;
 452         int err;
 453
 454         err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
 455         if (err < 0)
 456                 return ERR_PTR(err);
 457         if (!tb[TCA_STAB_BASE])
 458                 return ERR_PTR(-EINVAL);
 459
 460         s = nla_data(tb[TCA_STAB_BASE]);
 461
 462         if (s->tsize > 0) {
 463                 if (!tb[TCA_STAB_DATA])
 464                         return ERR_PTR(-EINVAL);
 465                 tab = nla_data(tb[TCA_STAB_DATA]);
 466                 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
 467         }
 468
 469         if (tsize != s->tsize || (!tab && tsize > 0))
 470                 return ERR_PTR(-EINVAL);
 471
 472         spin_lock(&qdisc_stab_lock);
 473
 474         list_for_each_entry(stab, &qdisc_stab_list, list) {
 475                 if (memcmp(&stab->szopts, s, sizeof(*s)))
 476                         continue;
 477                 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
 478                         continue;
 479                 stab->refcnt++;
 480                 spin_unlock(&qdisc_stab_lock);
 481                 return stab;
 482         }
 483
 484         spin_unlock(&qdisc_stab_lock);
 485
 486         stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
 487         if (!stab)
 488                 return ERR_PTR(-ENOMEM);
 489
 490         stab->refcnt = 1;
 491         stab->szopts = *s;
 492         if (tsize > 0)
 493                 memcpy(stab->data, tab, tsize * sizeof(u16));
 494
 495         spin_lock(&qdisc_stab_lock);
 496         list_add_tail(&stab->list, &qdisc_stab_list);
 497         spin_unlock(&qdisc_stab_lock);
 498
 499         return stab;
 500 }
 501
 502 static void stab_kfree_rcu(struct rcu_head *head)
 503 {
 504         kfree(container_of(head, struct qdisc_size_table, rcu));
 505 }
 506
 507 void qdisc_put_stab(struct qdisc_size_table *tab)
 508 {
 509         if (!tab)
 510                 return;
 511
 512         spin_lock(&qdisc_stab_lock);
 513
 514         if (--tab->refcnt == 0) {
 515                 list_del(&tab->list);
 516                 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
 517         }
 518
 519         spin_unlock(&qdisc_stab_lock);
 520 }
 521 EXPORT_SYMBOL(qdisc_put_stab);
 522
 523 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
 524 {
 525         struct nlattr *nest;
 526
 527         nest = nla_nest_start(skb, TCA_STAB);
 528         if (nest == NULL)
 529                 goto nla_put_failure;
 530         if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
 531                 goto nla_put_failure;
 532         nla_nest_end(skb, nest);
 533
 534         return skb->len;
 535
 536 nla_put_failure:
 537         return -1;
 538 }
 539
 540 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
 541 {
 542         int pkt_len, slot;
 543
 544         pkt_len = skb->len + stab->szopts.overhead;
 545         if (unlikely(!stab->szopts.tsize))
 546                 goto out;
 547
 548         slot = pkt_len + stab->szopts.cell_align;
 549         if (unlikely(slot < 0))
 550                 slot = 0;
 551
 552         slot >>= stab->szopts.cell_log;
 553         if (likely(slot < stab->szopts.tsize))
 554                 pkt_len = stab->data[slot];
 555         else
 556                 pkt_len = stab->data[stab->szopts.tsize - 1] *
 557                                 (slot / stab->szopts.tsize) +
 558                                 stab->data[slot % stab->szopts.tsize];
 559
 560         pkt_len <<= stab->szopts.size_log;
 561 out:
 562         if (unlikely(pkt_len < 1))
 563                 pkt_len = 1;
 564         qdisc_skb_cb(skb)->pkt_len = pkt_len;
 565 }
 566 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
 567
 568 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
 569 {
 570         if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
 571                 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
 572                         txt, qdisc->ops->id, qdisc->handle >> 16);
 573                 qdisc->flags |= TCQ_F_WARN_NONWC;
 574         }
 575 }
 576 EXPORT_SYMBOL(qdisc_warn_nonwc);
 577
 578 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
 579 {
 580         struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
 581                                                  timer);
 582
 583         rcu_read_lock();
 584         __netif_schedule(qdisc_root(wd->qdisc));
 585         rcu_read_unlock();
 586
 587         return HRTIMER_NORESTART;
 588 }
 589
 590 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
 591 {
 592         hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
 593         wd->timer.function = qdisc_watchdog;
 594         wd->qdisc = qdisc;
 595 }
 596 EXPORT_SYMBOL(qdisc_watchdog_init);
 597
 598 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
 599 {
 600         if (test_bit(__QDISC_STATE_DEACTIVATED,
 601                      &qdisc_root_sleeping(wd->qdisc)->state))
 602                 return;
 603
 604         if (wd->last_expires == expires)
 605                 return;
 606
 607         wd->last_expires = expires;
 608         hrtimer_start(&wd->timer,
 609                       ns_to_ktime(expires),
 610                       HRTIMER_MODE_ABS_PINNED);
 611 }
 612 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
 613
 614 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
 615 {
 616         hrtimer_cancel(&wd->timer);
 617 }
 618 EXPORT_SYMBOL(qdisc_watchdog_cancel);
 619
 620 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
 621 {
 622         unsigned int size = n * sizeof(struct hlist_head), i;
 623         struct hlist_head *h;
 624
 625         if (size <= PAGE_SIZE)
 626                 h = kmalloc(size, GFP_KERNEL);
 627         else
 628                 h = (struct hlist_head *)
 629                         __get_free_pages(GFP_KERNEL, get_order(size));
 630
 631         if (h != NULL) {
 632                 for (i = 0; i < n; i++)
 633                         INIT_HLIST_HEAD(&h[i]);
 634         }
 635         return h;
 636 }
 637
 638 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
 639 {
 640         unsigned int size = n * sizeof(struct hlist_head);
 641
 642         if (size <= PAGE_SIZE)
 643                 kfree(h);
 644         else
 645                 free_pages((unsigned long)h, get_order(size));
 646 }
 647
 648 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
 649 {
 650         struct Qdisc_class_common *cl;
 651         struct hlist_node *next;
 652         struct hlist_head *nhash, *ohash;
 653         unsigned int nsize, nmask, osize;
 654         unsigned int i, h;
 655
 656         /* Rehash when load factor exceeds 0.75 */
 657         if (clhash->hashelems * 4 <= clhash->hashsize * 3)
 658                 return;
 659         nsize = clhash->hashsize * 2;
 660         nmask = nsize - 1;
 661         nhash = qdisc_class_hash_alloc(nsize);
 662         if (nhash == NULL)
 663                 return;
 664
 665         ohash = clhash->hash;
 666         osize = clhash->hashsize;
 667
 668         sch_tree_lock(sch);
 669         for (i = 0; i < osize; i++) {
 670                 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
 671                         h = qdisc_class_hash(cl->classid, nmask);
 672                         hlist_add_head(&cl->hnode, &nhash[h]);
 673                 }
 674         }
 675         clhash->hash     = nhash;
 676         clhash->hashsize = nsize;
 677         clhash->hashmask = nmask;
 678         sch_tree_unlock(sch);
 679
 680         qdisc_class_hash_free(ohash, osize);
 681 }
 682 EXPORT_SYMBOL(qdisc_class_hash_grow);
 683
 684 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
 685 {
 686         unsigned int size = 4;
 687
 688         clhash->hash = qdisc_class_hash_alloc(size);
 689         if (clhash->hash == NULL)
 690                 return -ENOMEM;
 691         clhash->hashsize  = size;
 692         clhash->hashmask  = size - 1;
 693         clhash->hashelems = 0;
 694         return 0;
 695 }
 696 EXPORT_SYMBOL(qdisc_class_hash_init);
 697
 698 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
 699 {
 700         qdisc_class_hash_free(clhash->hash, clhash->hashsize);
 701 }
 702 EXPORT_SYMBOL(qdisc_class_hash_destroy);
 703
 704 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
 705                              struct Qdisc_class_common *cl)
 706 {
 707         unsigned int h;
 708
 709         INIT_HLIST_NODE(&cl->hnode);
 710         h = qdisc_class_hash(cl->classid, clhash->hashmask);
 711         hlist_add_head(&cl->hnode, &clhash->hash[h]);
 712         clhash->hashelems++;
 713 }
 714 EXPORT_SYMBOL(qdisc_class_hash_insert);
 715
 716 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
 717                              struct Qdisc_class_common *cl)
 718 {
 719         hlist_del(&cl->hnode);
 720         clhash->hashelems--;
 721 }
 722 EXPORT_SYMBOL(qdisc_class_hash_remove);
 723
 724 /* Allocate an unique handle from space managed by kernel
 725  * Possible range is [8000-FFFF]:0000 (0x8000 values)
 726  */
 727 static u32 qdisc_alloc_handle(struct net_device *dev)
 728 {
 729         int i = 0x8000;
 730         static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
 731
 732         do {
 733                 autohandle += TC_H_MAKE(0x10000U, 0);
 734                 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
 735                         autohandle = TC_H_MAKE(0x80000000U, 0);
 736                 if (!qdisc_lookup(dev, autohandle))
 737                         return autohandle;
 738                 cond_resched();
 739         } while (--i > 0);
 740
 741         return 0;
 742 }
 743
 744 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
 745                                unsigned int len)
 746 {
 747         const struct Qdisc_class_ops *cops;
 748         unsigned long cl;
 749         u32 parentid;
 750         int drops;
 751
 752         if (n == 0 && len == 0)
 753                 return;
 754         drops = max_t(int, n, 0);
 755         rcu_read_lock();
 756         while ((parentid = sch->parent)) {
 757                 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
 758                         break;
 759
 760                 if (sch->flags & TCQ_F_NOPARENT)
 761                         break;
 762                 /* TODO: perform the search on a per txq basis */
 763                 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
 764                 if (sch == NULL) {
 765                         WARN_ON_ONCE(parentid != TC_H_ROOT);
 766                         break;
 767                 }
 768                 cops = sch->ops->cl_ops;
 769                 if (cops->qlen_notify) {
 770                         cl = cops->get(sch, parentid);
 771                         cops->qlen_notify(sch, cl);
 772                         cops->put(sch, cl);
 773                 }
 774                 sch->q.qlen -= n;
 775                 sch->qstats.backlog -= len;
 776                 __qdisc_qstats_drop(sch, drops);
 777         }
 778         rcu_read_unlock();
 779 }
 780 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
 781
 782 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
 783                                struct nlmsghdr *n, u32 clid,
 784                                struct Qdisc *old, struct Qdisc *new)
 785 {
 786         if (new || old)
 787                 qdisc_notify(net, skb, n, clid, old, new);
 788
 789         if (old)
 790                 qdisc_destroy(old);
 791 }
 792
 793 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
 794  * to device "dev".
 795  *
 796  * When appropriate send a netlink notification using 'skb'
 797  * and "n".
 798  *
 799  * On success, destroy old qdisc.
 800  */
 801
 802 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
 803                        struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
 804                        struct Qdisc *new, struct Qdisc *old)
 805 {
 806         struct Qdisc *q = old;
 807         struct net *net = dev_net(dev);
 808         int err = 0;
 809
 810         if (parent == NULL) {
 811                 unsigned int i, num_q, ingress;
 812
 813                 ingress = 0;
 814                 num_q = dev->num_tx_queues;
 815                 if ((q && q->flags & TCQ_F_INGRESS) ||
 816                     (new && new->flags & TCQ_F_INGRESS)) {
 817                         num_q = 1;
 818                         ingress = 1;
 819                         if (!dev_ingress_queue(dev))
 820                                 return -ENOENT;
 821                 }
 822
 823                 if (dev->flags & IFF_UP)
 824                         dev_deactivate(dev);
 825
 826                 if (new && new->ops->attach)
 827                         goto skip;
 828
 829                 for (i = 0; i < num_q; i++) {
 830                         struct netdev_queue *dev_queue = dev_ingress_queue(dev);
 831
 832                         if (!ingress)
 833                                 dev_queue = netdev_get_tx_queue(dev, i);
 834
 835                         old = dev_graft_qdisc(dev_queue, new);
 836                         if (new && i > 0)
 837                                 atomic_inc(&new->refcnt);
 838
 839                         if (!ingress)
 840                                 qdisc_destroy(old);
 841                 }
 842
 843 skip:
 844                 if (!ingress) {
 845                         notify_and_destroy(net, skb, n, classid,
 846                                            dev->qdisc, new);
 847                         if (new && !new->ops->attach)
 848                                 atomic_inc(&new->refcnt);
 849                         dev->qdisc = new ? : &noop_qdisc;
 850
 851                         if (new && new->ops->attach)
 852                                 new->ops->attach(new);
 853                 } else {
 854                         notify_and_destroy(net, skb, n, classid, old, new);
 855                 }
 856
 857                 if (dev->flags & IFF_UP)
 858                         dev_activate(dev);
 859         } else {
 860                 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
 861
 862                 err = -EOPNOTSUPP;
 863                 if (cops && cops->graft) {
 864                         unsigned long cl = cops->get(parent, classid);
 865                         if (cl) {
 866                                 err = cops->graft(parent, cl, new, &old);
 867                                 cops->put(parent, cl);
 868                         } else
 869                                 err = -ENOENT;
 870                 }
 871                 if (!err)
 872                         notify_and_destroy(net, skb, n, classid, old, new);
 873         }
 874         return err;
 875 }
 876
 877 /* lockdep annotation is needed for ingress; egress gets it only for name */
 878 static struct lock_class_key qdisc_tx_lock;
 879 static struct lock_class_key qdisc_rx_lock;
 880
 881 /*
 882    Allocate and initialize new qdisc.
 883
 884    Parameters are passed via opt.
 885  */
 886
 887 static struct Qdisc *
 888 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
 889              struct Qdisc *p, u32 parent, u32 handle,
 890              struct nlattr **tca, int *errp)
 891 {
 892         int err;
 893         struct nlattr *kind = tca[TCA_KIND];
 894         struct Qdisc *sch;
 895         struct Qdisc_ops *ops;
 896         struct qdisc_size_table *stab;
 897
 898         ops = qdisc_lookup_ops(kind);
 899 #ifdef CONFIG_MODULES
 900         if (ops == NULL && kind != NULL) {
 901                 char name[IFNAMSIZ];
 902                 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 903                         /* We dropped the RTNL semaphore in order to
 904                          * perform the module load.  So, even if we
 905                          * succeeded in loading the module we have to
 906                          * tell the caller to replay the request.  We
 907                          * indicate this using -EAGAIN.
 908                          * We replay the request because the device may
 909                          * go away in the mean time.
 910                          */
 911                         rtnl_unlock();
 912                         request_module("sch_%s", name);
 913                         rtnl_lock();
 914                         ops = qdisc_lookup_ops(kind);
 915                         if (ops != NULL) {
 916                                 /* We will try again qdisc_lookup_ops,
 917                                  * so don't keep a reference.
 918                                  */
 919                                 module_put(ops->owner);
 920                                 err = -EAGAIN;
 921                                 goto err_out;
 922                         }
 923                 }
 924         }
 925 #endif
 926
 927         err = -ENOENT;
 928         if (ops == NULL)
 929                 goto err_out;
 930
 931         sch = qdisc_alloc(dev_queue, ops);
 932         if (IS_ERR(sch)) {
 933                 err = PTR_ERR(sch);
 934                 goto err_out2;
 935         }
 936
 937         sch->parent = parent;
 938
 939         if (handle == TC_H_INGRESS) {
 940                 sch->flags |= TCQ_F_INGRESS;
 941                 handle = TC_H_MAKE(TC_H_INGRESS, 0);
 942                 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
 943         } else {
 944                 if (handle == 0) {
 945                         handle = qdisc_alloc_handle(dev);
 946                         err = -ENOMEM;
 947                         if (handle == 0)
 948                                 goto err_out3;
 949                 }
 950                 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
 951                 if (!netif_is_multiqueue(dev))
 952                         sch->flags |= TCQ_F_ONETXQUEUE;
 953         }
 954
 955         sch->handle = handle;
 956
 957         if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
 958                 if (qdisc_is_percpu_stats(sch)) {
 959                         sch->cpu_bstats =
 960                                 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 961                         if (!sch->cpu_bstats)
 962                                 goto err_out4;
 963
 964                         sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 965                         if (!sch->cpu_qstats)
 966                                 goto err_out4;
 967                 }
 968
 969                 if (tca[TCA_STAB]) {
 970                         stab = qdisc_get_stab(tca[TCA_STAB]);
 971                         if (IS_ERR(stab)) {
 972                                 err = PTR_ERR(stab);
 973                                 goto err_out4;
 974                         }
 975                         rcu_assign_pointer(sch->stab, stab);
 976                 }
 977                 if (tca[TCA_RATE]) {
 978                         seqcount_t *running;
 979
 980                         err = -EOPNOTSUPP;
 981                         if (sch->flags & TCQ_F_MQROOT)
 982                                 goto err_out4;
 983
 984                         if ((sch->parent != TC_H_ROOT) &&
 985                             !(sch->flags & TCQ_F_INGRESS) &&
 986                             (!p || !(p->flags & TCQ_F_MQROOT)))
 987                                 running = qdisc_root_sleeping_running(sch);
 988                         else
 989                                 running = &sch->running;
 990
 991                         err = gen_new_estimator(&sch->bstats,
 992                                                 sch->cpu_bstats,
 993                                                 &sch->rate_est,
 994                                                 NULL,
 995                                                 running,
 996                                                 tca[TCA_RATE]);
 997                         if (err)
 998                                 goto err_out4;
 999                 }
1000
1001                 qdisc_list_add(sch);
1002
1003                 return sch;
1004         }
1005 err_out3:
1006         dev_put(dev);
1007         kfree((char *) sch - sch->padded);
1008 err_out2:
1009         module_put(ops->owner);
1010 err_out:
1011         *errp = err;
1012         return NULL;
1013
1014 err_out4:
1015         free_percpu(sch->cpu_bstats);
1016         free_percpu(sch->cpu_qstats);
1017         /*
1018          * Any broken qdiscs that would require a ops->reset() here?
1019          * The qdisc was never in action so it shouldn't be necessary.
1020          */
1021         qdisc_put_stab(rtnl_dereference(sch->stab));
1022         if (ops->destroy)
1023                 ops->destroy(sch);
1024         goto err_out3;
1025 }
1026
1027 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1028 {
1029         struct qdisc_size_table *ostab, *stab = NULL;
1030         int err = 0;
1031
1032         if (tca[TCA_OPTIONS]) {
1033                 if (sch->ops->change == NULL)
1034                         return -EINVAL;
1035                 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1036                 if (err)
1037                         return err;
1038         }
1039
1040         if (tca[TCA_STAB]) {
1041                 stab = qdisc_get_stab(tca[TCA_STAB]);
1042                 if (IS_ERR(stab))
1043                         return PTR_ERR(stab);
1044         }
1045
1046         ostab = rtnl_dereference(sch->stab);
1047         rcu_assign_pointer(sch->stab, stab);
1048         qdisc_put_stab(ostab);
1049
1050         if (tca[TCA_RATE]) {
1051                 /* NB: ignores errors from replace_estimator
1052                    because change can't be undone. */
1053                 if (sch->flags & TCQ_F_MQROOT)
1054                         goto out;
1055                 gen_replace_estimator(&sch->bstats,
1056                                       sch->cpu_bstats,
1057                                       &sch->rate_est,
1058                                       NULL,
1059                                       qdisc_root_sleeping_running(sch),
1060                                       tca[TCA_RATE]);
1061         }
1062 out:
1063         return 0;
1064 }
1065
1066 struct check_loop_arg {
1067         struct qdisc_walker     w;
1068         struct Qdisc            *p;
1069         int                     depth;
1070 };
1071
1072 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1073
1074 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1075 {
1076         struct check_loop_arg   arg;
1077
1078         if (q->ops->cl_ops == NULL)
1079                 return 0;
1080
1081         arg.w.stop = arg.w.skip = arg.w.count = 0;
1082         arg.w.fn = check_loop_fn;
1083         arg.depth = depth;
1084         arg.p = p;
1085         q->ops->cl_ops->walk(q, &arg.w);
1086         return arg.w.stop ? -ELOOP : 0;
1087 }
1088
1089 static int
1090 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1091 {
1092         struct Qdisc *leaf;
1093         const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1094         struct check_loop_arg *arg = (struct check_loop_arg *)w;
1095
1096         leaf = cops->leaf(q, cl);
1097         if (leaf) {
1098                 if (leaf == arg->p || arg->depth > 7)
1099                         return -ELOOP;
1100                 return check_loop(leaf, arg->p, arg->depth + 1);
1101         }
1102         return 0;
1103 }
1104
1105 /*
1106  * Delete/get qdisc.
1107  */
1108
1109 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1110 {
1111         struct net *net = sock_net(skb->sk);
1112         struct tcmsg *tcm = nlmsg_data(n);
1113         struct nlattr *tca[TCA_MAX + 1];
1114         struct net_device *dev;
1115         u32 clid;
1116         struct Qdisc *q = NULL;
1117         struct Qdisc *p = NULL;
1118         int err;
1119
1120         if ((n->nlmsg_type != RTM_GETQDISC) &&
1121             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1122                 return -EPERM;
1123
1124         err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1125         if (err < 0)
1126                 return err;
1127
1128         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1129         if (!dev)
1130                 return -ENODEV;
1131
1132         clid = tcm->tcm_parent;
1133         if (clid) {
1134                 if (clid != TC_H_ROOT) {
1135                         if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1136                                 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1137                                 if (!p)
1138                                         return -ENOENT;
1139                                 q = qdisc_leaf(p, clid);
1140                         } else if (dev_ingress_queue(dev)) {
1141                                 q = dev_ingress_queue(dev)->qdisc_sleeping;
1142                         }
1143                 } else {
1144                         q = dev->qdisc;
1145                 }
1146                 if (!q)
1147                         return -ENOENT;
1148
1149                 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1150                         return -EINVAL;
1151         } else {
1152                 q = qdisc_lookup(dev, tcm->tcm_handle);
1153                 if (!q)
1154                         return -ENOENT;
1155         }
1156
1157         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1158                 return -EINVAL;
1159
1160         if (n->nlmsg_type == RTM_DELQDISC) {
1161                 if (!clid)
1162                         return -EINVAL;
1163                 if (q->handle == 0)
1164                         return -ENOENT;
1165                 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1166                 if (err != 0)
1167                         return err;
1168         } else {
1169                 qdisc_notify(net, skb, n, clid, NULL, q);
1170         }
1171         return 0;
1172 }
1173
1174 /*
1175  * Create/change qdisc.
1176  */
1177
1178 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1179 {
1180         struct net *net = sock_net(skb->sk);
1181         struct tcmsg *tcm;
1182         struct nlattr *tca[TCA_MAX + 1];
1183         struct net_device *dev;
1184         u32 clid;
1185         struct Qdisc *q, *p;
1186         int err;
1187
1188         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1189                 return -EPERM;
1190
1191 replay:
1192         /* Reinit, just in case something touches this. */
1193         err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1194         if (err < 0)
1195                 return err;
1196
1197         tcm = nlmsg_data(n);
1198         clid = tcm->tcm_parent;
1199         q = p = NULL;
1200
1201         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1202         if (!dev)
1203                 return -ENODEV;
1204
1205
1206         if (clid) {
1207                 if (clid != TC_H_ROOT) {
1208                         if (clid != TC_H_INGRESS) {
1209                                 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1210                                 if (!p)
1211                                         return -ENOENT;
1212                                 q = qdisc_leaf(p, clid);
1213                         } else if (dev_ingress_queue_create(dev)) {
1214                                 q = dev_ingress_queue(dev)->qdisc_sleeping;
1215                         }
1216                 } else {
1217                         q = dev->qdisc;
1218                 }
1219
1220                 /* It may be default qdisc, ignore it */
1221                 if (q && q->handle == 0)
1222                         q = NULL;
1223
1224                 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1225                         if (tcm->tcm_handle) {
1226                                 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1227                                         return -EEXIST;
1228                                 if (TC_H_MIN(tcm->tcm_handle))
1229                                         return -EINVAL;
1230                                 q = qdisc_lookup(dev, tcm->tcm_handle);
1231                                 if (!q)
1232                                         goto create_n_graft;
1233                                 if (n->nlmsg_flags & NLM_F_EXCL)
1234                                         return -EEXIST;
1235                                 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1236                                         return -EINVAL;
1237                                 if (q == p ||
1238                                     (p && check_loop(q, p, 0)))
1239                                         return -ELOOP;
1240                                 atomic_inc(&q->refcnt);
1241                                 goto graft;
1242                         } else {
1243                                 if (!q)
1244                                         goto create_n_graft;
1245
1246                                 /* This magic test requires explanation.
1247                                  *
1248                                  *   We know, that some child q is already
1249                                  *   attached to this parent and have choice:
1250                                  *   either to change it or to create/graft new one.
1251                                  *
1252                                  *   1. We are allowed to create/graft only
1253                                  *   if CREATE and REPLACE flags are set.
1254                                  *
1255                                  *   2. If EXCL is set, requestor wanted to say,
1256                                  *   that qdisc tcm_handle is not expected
1257                                  *   to exist, so that we choose create/graft too.
1258                                  *
1259                                  *   3. The last case is when no flags are set.
1260                                  *   Alas, it is sort of hole in API, we
1261                                  *   cannot decide what to do unambiguously.
1262                                  *   For now we select create/graft, if
1263                                  *   user gave KIND, which does not match existing.
1264                                  */
1265                                 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1266                                     (n->nlmsg_flags & NLM_F_REPLACE) &&
1267                                     ((n->nlmsg_flags & NLM_F_EXCL) ||
1268                                      (tca[TCA_KIND] &&
1269                                       nla_strcmp(tca[TCA_KIND], q->ops->id))))
1270                                         goto create_n_graft;
1271                         }
1272                 }
1273         } else {
1274                 if (!tcm->tcm_handle)
1275                         return -EINVAL;
1276                 q = qdisc_lookup(dev, tcm->tcm_handle);
1277         }
1278
1279         /* Change qdisc parameters */
1280         if (q == NULL)
1281                 return -ENOENT;
1282         if (n->nlmsg_flags & NLM_F_EXCL)
1283                 return -EEXIST;
1284         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1285                 return -EINVAL;
1286         err = qdisc_change(q, tca);
1287         if (err == 0)
1288                 qdisc_notify(net, skb, n, clid, NULL, q);
1289         return err;
1290
1291 create_n_graft:
1292         if (!(n->nlmsg_flags & NLM_F_CREATE))
1293                 return -ENOENT;
1294         if (clid == TC_H_INGRESS) {
1295                 if (dev_ingress_queue(dev))
1296                         q = qdisc_create(dev, dev_ingress_queue(dev), p,
1297                                          tcm->tcm_parent, tcm->tcm_parent,
1298                                          tca, &err);
1299                 else
1300                         err = -ENOENT;
1301         } else {
1302                 struct netdev_queue *dev_queue;
1303
1304                 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1305                         dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1306                 else if (p)
1307                         dev_queue = p->dev_queue;
1308                 else
1309                         dev_queue = netdev_get_tx_queue(dev, 0);
1310
1311                 q = qdisc_create(dev, dev_queue, p,
1312                                  tcm->tcm_parent, tcm->tcm_handle,
1313                                  tca, &err);
1314         }
1315         if (q == NULL) {
1316                 if (err == -EAGAIN)
1317                         goto replay;
1318                 return err;
1319         }
1320
1321 graft:
1322         err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1323         if (err) {
1324                 if (q)
1325                         qdisc_destroy(q);
1326                 return err;
1327         }
1328
1329         return 0;
1330 }
1331
1332 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1333                          u32 portid, u32 seq, u16 flags, int event)
1334 {
1335         struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1336         struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1337         struct tcmsg *tcm;
1338         struct nlmsghdr  *nlh;
1339         unsigned char *b = skb_tail_pointer(skb);
1340         struct gnet_dump d;
1341         struct qdisc_size_table *stab;
1342         __u32 qlen;
1343
1344         cond_resched();
1345         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1346         if (!nlh)
1347                 goto out_nlmsg_trim;
1348         tcm = nlmsg_data(nlh);
1349         tcm->tcm_family = AF_UNSPEC;
1350         tcm->tcm__pad1 = 0;
1351         tcm->tcm__pad2 = 0;
1352         tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1353         tcm->tcm_parent = clid;
1354         tcm->tcm_handle = q->handle;
1355         tcm->tcm_info = atomic_read(&q->refcnt);
1356         if (nla_put_string(skb, TCA_KIND, q->ops->id))
1357                 goto nla_put_failure;
1358         if (q->ops->dump && q->ops->dump(q, skb) < 0)
1359                 goto nla_put_failure;
1360         qlen = q->q.qlen;
1361
1362         stab = rtnl_dereference(q->stab);
1363         if (stab && qdisc_dump_stab(skb, stab) < 0)
1364                 goto nla_put_failure;
1365
1366         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1367                                          NULL, &d, TCA_PAD) < 0)
1368                 goto nla_put_failure;
1369
1370         if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1371                 goto nla_put_failure;
1372
1373         if (qdisc_is_percpu_stats(q)) {
1374                 cpu_bstats = q->cpu_bstats;
1375                 cpu_qstats = q->cpu_qstats;
1376         }
1377
1378         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
1379                                   &d, cpu_bstats, &q->bstats) < 0 ||
1380             gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1381             gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1382                 goto nla_put_failure;
1383
1384         if (gnet_stats_finish_copy(&d) < 0)
1385                 goto nla_put_failure;
1386
1387         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1388         return skb->len;
1389
1390 out_nlmsg_trim:
1391 nla_put_failure:
1392         nlmsg_trim(skb, b);
1393         return -1;
1394 }
1395
1396 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1397 {
1398         return (q->flags & TCQ_F_BUILTIN) ? true : false;
1399 }
1400
1401 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1402                         struct nlmsghdr *n, u32 clid,
1403                         struct Qdisc *old, struct Qdisc *new)
1404 {
1405         struct sk_buff *skb;
1406         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1407
1408         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1409         if (!skb)
1410                 return -ENOBUFS;
1411
1412         if (old && !tc_qdisc_dump_ignore(old)) {
1413                 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1414                                   0, RTM_DELQDISC) < 0)
1415                         goto err_out;
1416         }
1417         if (new && !tc_qdisc_dump_ignore(new)) {
1418                 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1419                                   old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1420                         goto err_out;
1421         }
1422
1423         if (skb->len)
1424                 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1425                                       n->nlmsg_flags & NLM_F_ECHO);
1426
1427 err_out:
1428         kfree_skb(skb);
1429         return -EINVAL;
1430 }
1431
1432 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1433                               struct netlink_callback *cb,
1434                               int *q_idx_p, int s_q_idx)
1435 {
1436         int ret = 0, q_idx = *q_idx_p;
1437         struct Qdisc *q;
1438
1439         if (!root)
1440                 return 0;
1441
1442         q = root;
1443         if (q_idx < s_q_idx) {
1444                 q_idx++;
1445         } else {
1446                 if (!tc_qdisc_dump_ignore(q) &&
1447                     tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1448                                   cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1449                         goto done;
1450                 q_idx++;
1451         }
1452         list_for_each_entry(q, &root->list, list) {
1453                 if (q_idx < s_q_idx) {
1454                         q_idx++;
1455                         continue;
1456                 }
1457                 if (!tc_qdisc_dump_ignore(q) &&
1458                     tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1459                                   cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1460                         goto done;
1461                 q_idx++;
1462         }
1463
1464 out:
1465         *q_idx_p = q_idx;
1466         return ret;
1467 done:
1468         ret = -1;
1469         goto out;
1470 }
1471
1472 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1473 {
1474         struct net *net = sock_net(skb->sk);
1475         int idx, q_idx;
1476         int s_idx, s_q_idx;
1477         struct net_device *dev;
1478
1479         s_idx = cb->args[0];
1480         s_q_idx = q_idx = cb->args[1];
1481
1482         idx = 0;
1483         ASSERT_RTNL();
1484         for_each_netdev(net, dev) {
1485                 struct netdev_queue *dev_queue;
1486
1487                 if (idx < s_idx)
1488                         goto cont;
1489                 if (idx > s_idx)
1490                         s_q_idx = 0;
1491                 q_idx = 0;
1492
1493                 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1494                         goto done;
1495
1496                 dev_queue = dev_ingress_queue(dev);
1497                 if (dev_queue &&
1498                     tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1499                                        &q_idx, s_q_idx) < 0)
1500                         goto done;
1501
1502 cont:
1503                 idx++;
1504         }
1505
1506 done:
1507         cb->args[0] = idx;
1508         cb->args[1] = q_idx;
1509
1510         return skb->len;
1511 }
1512
1513
1514
1515 /************************************************
1516  *      Traffic classes manipulation.           *
1517  ************************************************/
1518
1519
1520
1521 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1522 {
1523         struct net *net = sock_net(skb->sk);
1524         struct tcmsg *tcm = nlmsg_data(n);
1525         struct nlattr *tca[TCA_MAX + 1];
1526         struct net_device *dev;
1527         struct Qdisc *q = NULL;
1528         const struct Qdisc_class_ops *cops;
1529         unsigned long cl = 0;
1530         unsigned long new_cl;
1531         u32 portid;
1532         u32 clid;
1533         u32 qid;
1534         int err;
1535
1536         if ((n->nlmsg_type != RTM_GETTCLASS) &&
1537             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1538                 return -EPERM;
1539
1540         err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1541         if (err < 0)
1542                 return err;
1543
1544         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1545         if (!dev)
1546                 return -ENODEV;
1547
1548         /*
1549            parent == TC_H_UNSPEC - unspecified parent.
1550            parent == TC_H_ROOT   - class is root, which has no parent.
1551            parent == X:0         - parent is root class.
1552            parent == X:Y         - parent is a node in hierarchy.
1553            parent == 0:Y         - parent is X:Y, where X:0 is qdisc.
1554
1555            handle == 0:0         - generate handle from kernel pool.
1556            handle == 0:Y         - class is X:Y, where X:0 is qdisc.
1557            handle == X:Y         - clear.
1558            handle == X:0         - root class.
1559          */
1560
1561         /* Step 1. Determine qdisc handle X:0 */
1562
1563         portid = tcm->tcm_parent;
1564         clid = tcm->tcm_handle;
1565         qid = TC_H_MAJ(clid);
1566
1567         if (portid != TC_H_ROOT) {
1568                 u32 qid1 = TC_H_MAJ(portid);
1569
1570                 if (qid && qid1) {
1571                         /* If both majors are known, they must be identical. */
1572                         if (qid != qid1)
1573                                 return -EINVAL;
1574                 } else if (qid1) {
1575                         qid = qid1;
1576                 } else if (qid == 0)
1577                         qid = dev->qdisc->handle;
1578
1579                 /* Now qid is genuine qdisc handle consistent
1580                  * both with parent and child.
1581                  *
1582                  * TC_H_MAJ(portid) still may be unspecified, complete it now.
1583                  */
1584                 if (portid)
1585                         portid = TC_H_MAKE(qid, portid);
1586         } else {
1587                 if (qid == 0)
1588                         qid = dev->qdisc->handle;
1589         }
1590
1591         /* OK. Locate qdisc */
1592         q = qdisc_lookup(dev, qid);
1593         if (!q)
1594                 return -ENOENT;
1595
1596         /* An check that it supports classes */
1597         cops = q->ops->cl_ops;
1598         if (cops == NULL)
1599                 return -EINVAL;
1600
1601         /* Now try to get class */
1602         if (clid == 0) {
1603                 if (portid == TC_H_ROOT)
1604                         clid = qid;
1605         } else
1606                 clid = TC_H_MAKE(qid, clid);
1607
1608         if (clid)
1609                 cl = cops->get(q, clid);
1610
1611         if (cl == 0) {
1612                 err = -ENOENT;
1613                 if (n->nlmsg_type != RTM_NEWTCLASS ||
1614                     !(n->nlmsg_flags & NLM_F_CREATE))
1615                         goto out;
1616         } else {
1617                 switch (n->nlmsg_type) {
1618                 case RTM_NEWTCLASS:
1619                         err = -EEXIST;
1620                         if (n->nlmsg_flags & NLM_F_EXCL)
1621                                 goto out;
1622                         break;
1623                 case RTM_DELTCLASS:
1624                         err = -EOPNOTSUPP;
1625                         if (cops->delete)
1626                                 err = cops->delete(q, cl);
1627                         if (err == 0)
1628                                 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1629                         goto out;
1630                 case RTM_GETTCLASS:
1631                         err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1632                         goto out;
1633                 default:
1634                         err = -EINVAL;
1635                         goto out;
1636                 }
1637         }
1638
1639         new_cl = cl;
1640         err = -EOPNOTSUPP;
1641         if (cops->change)
1642                 err = cops->change(q, clid, portid, tca, &new_cl);
1643         if (err == 0)
1644                 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1645
1646 out:
1647         if (cl)
1648                 cops->put(q, cl);
1649
1650         return err;
1651 }
1652
1653
1654 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1655                           unsigned long cl,
1656                           u32 portid, u32 seq, u16 flags, int event)
1657 {
1658         struct tcmsg *tcm;
1659         struct nlmsghdr  *nlh;
1660         unsigned char *b = skb_tail_pointer(skb);
1661         struct gnet_dump d;
1662         const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1663
1664         cond_resched();
1665         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1666         if (!nlh)
1667                 goto out_nlmsg_trim;
1668         tcm = nlmsg_data(nlh);
1669         tcm->tcm_family = AF_UNSPEC;
1670         tcm->tcm__pad1 = 0;
1671         tcm->tcm__pad2 = 0;
1672         tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1673         tcm->tcm_parent = q->handle;
1674         tcm->tcm_handle = q->handle;
1675         tcm->tcm_info = 0;
1676         if (nla_put_string(skb, TCA_KIND, q->ops->id))
1677                 goto nla_put_failure;
1678         if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1679                 goto nla_put_failure;
1680
1681         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1682                                          NULL, &d, TCA_PAD) < 0)
1683                 goto nla_put_failure;
1684
1685         if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1686                 goto nla_put_failure;
1687
1688         if (gnet_stats_finish_copy(&d) < 0)
1689                 goto nla_put_failure;
1690
1691         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1692         return skb->len;
1693
1694 out_nlmsg_trim:
1695 nla_put_failure:
1696         nlmsg_trim(skb, b);
1697         return -1;
1698 }
1699
1700 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1701                          struct nlmsghdr *n, struct Qdisc *q,
1702                          unsigned long cl, int event)
1703 {
1704         struct sk_buff *skb;
1705         u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1706
1707         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1708         if (!skb)
1709                 return -ENOBUFS;
1710
1711         if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1712                 kfree_skb(skb);
1713                 return -EINVAL;
1714         }
1715
1716         return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1717                               n->nlmsg_flags & NLM_F_ECHO);
1718 }
1719
1720 struct qdisc_dump_args {
1721         struct qdisc_walker     w;
1722         struct sk_buff          *skb;
1723         struct netlink_callback *cb;
1724 };
1725
1726 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1727 {
1728         struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1729
1730         return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1731                               a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1732 }
1733
1734 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1735                                 struct tcmsg *tcm, struct netlink_callback *cb,
1736                                 int *t_p, int s_t)
1737 {
1738         struct qdisc_dump_args arg;
1739
1740         if (tc_qdisc_dump_ignore(q) ||
1741             *t_p < s_t || !q->ops->cl_ops ||
1742             (tcm->tcm_parent &&
1743              TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1744                 (*t_p)++;
1745                 return 0;
1746         }
1747         if (*t_p > s_t)
1748                 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1749         arg.w.fn = qdisc_class_dump;
1750         arg.skb = skb;
1751         arg.cb = cb;
1752         arg.w.stop  = 0;
1753         arg.w.skip = cb->args[1];
1754         arg.w.count = 0;
1755         q->ops->cl_ops->walk(q, &arg.w);
1756         cb->args[1] = arg.w.count;
1757         if (arg.w.stop)
1758                 return -1;
1759         (*t_p)++;
1760         return 0;
1761 }
1762
1763 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1764                                struct tcmsg *tcm, struct netlink_callback *cb,
1765                                int *t_p, int s_t)
1766 {
1767         struct Qdisc *q;
1768
1769         if (!root)
1770                 return 0;
1771
1772         if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1773                 return -1;
1774
1775         list_for_each_entry(q, &root->list, list) {
1776                 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1777                         return -1;
1778         }
1779
1780         return 0;
1781 }
1782
1783 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1784 {
1785         struct tcmsg *tcm = nlmsg_data(cb->nlh);
1786         struct net *net = sock_net(skb->sk);
1787         struct netdev_queue *dev_queue;
1788         struct net_device *dev;
1789         int t, s_t;
1790
1791         if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1792                 return 0;
1793         dev = dev_get_by_index(net, tcm->tcm_ifindex);
1794         if (!dev)
1795                 return 0;
1796
1797         s_t = cb->args[0];
1798         t = 0;
1799
1800         if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1801                 goto done;
1802
1803         dev_queue = dev_ingress_queue(dev);
1804         if (dev_queue &&
1805             tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1806                                 &t, s_t) < 0)
1807                 goto done;
1808
1809 done:
1810         cb->args[0] = t;
1811
1812         dev_put(dev);
1813         return skb->len;
1814 }
1815
1816 /* Main classifier routine: scans classifier chain attached
1817  * to this qdisc, (optionally) tests for protocol and asks
1818  * specific classifiers.
1819  */
1820 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1821                 struct tcf_result *res, bool compat_mode)
1822 {
1823         __be16 protocol = tc_skb_protocol(skb);
1824 #ifdef CONFIG_NET_CLS_ACT
1825         const struct tcf_proto *old_tp = tp;
1826         int limit = 0;
1827
1828 reclassify:
1829 #endif
1830         for (; tp; tp = rcu_dereference_bh(tp->next)) {
1831                 int err;
1832
1833                 if (tp->protocol != protocol &&
1834                     tp->protocol != htons(ETH_P_ALL))
1835                         continue;
1836
1837                 err = tp->classify(skb, tp, res);
1838 #ifdef CONFIG_NET_CLS_ACT
1839                 if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1840                         goto reset;
1841 #endif
1842                 if (err >= 0)
1843                         return err;
1844         }
1845
1846         return TC_ACT_UNSPEC; /* signal: continue lookup */
1847 #ifdef CONFIG_NET_CLS_ACT
1848 reset:
1849         if (unlikely(limit++ >= MAX_REC_LOOP)) {
1850                 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1851                                        tp->q->ops->id, tp->prio & 0xffff,
1852                                        ntohs(tp->protocol));
1853                 return TC_ACT_SHOT;
1854         }
1855
1856         tp = old_tp;
1857         protocol = tc_skb_protocol(skb);
1858         goto reclassify;
1859 #endif
1860 }
1861 EXPORT_SYMBOL(tc_classify);
1862
1863 bool tcf_destroy(struct tcf_proto *tp, bool force)
1864 {
1865         if (tp->ops->destroy(tp, force)) {
1866                 module_put(tp->ops->owner);
1867                 kfree_rcu(tp, rcu);
1868                 return true;
1869         }
1870
1871         return false;
1872 }
1873
1874 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1875 {
1876         struct tcf_proto *tp;
1877
1878         while ((tp = rtnl_dereference(*fl)) != NULL) {
1879                 RCU_INIT_POINTER(*fl, tp->next);
1880                 tcf_destroy(tp, true);
1881         }
1882 }
1883 EXPORT_SYMBOL(tcf_destroy_chain);
1884
1885 #ifdef CONFIG_PROC_FS
1886 static int psched_show(struct seq_file *seq, void *v)
1887 {
1888         seq_printf(seq, "%08x %08x %08x %08x\n",
1889                    (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1890                    1000000,
1891                    (u32)NSEC_PER_SEC / hrtimer_resolution);
1892
1893         return 0;
1894 }
1895
1896 static int psched_open(struct inode *inode, struct file *file)
1897 {
1898         return single_open(file, psched_show, NULL);
1899 }
1900
1901 static const struct file_operations psched_fops = {
1902         .owner = THIS_MODULE,
1903         .open = psched_open,
1904         .read  = seq_read,
1905         .llseek = seq_lseek,
1906         .release = single_release,
1907 };
1908
1909 static int __net_init psched_net_init(struct net *net)
1910 {
1911         struct proc_dir_entry *e;
1912
1913         e = proc_create("psched", 0, net->proc_net, &psched_fops);
1914         if (e == NULL)
1915                 return -ENOMEM;
1916
1917         return 0;
1918 }
1919
1920 static void __net_exit psched_net_exit(struct net *net)
1921 {
1922         remove_proc_entry("psched", net->proc_net);
1923 }
1924 #else
1925 static int __net_init psched_net_init(struct net *net)
1926 {
1927         return 0;
1928 }
1929
1930 static void __net_exit psched_net_exit(struct net *net)
1931 {
1932 }
1933 #endif
1934
1935 static struct pernet_operations psched_net_ops = {
1936         .init = psched_net_init,
1937         .exit = psched_net_exit,
1938 };
1939
1940 static int __init pktsched_init(void)
1941 {
1942         int err;
1943
1944         err = register_pernet_subsys(&psched_net_ops);
1945         if (err) {
1946                 pr_err("pktsched_init: "
1947                        "cannot initialize per netns operations\n");
1948                 return err;
1949         }
1950
1951         register_qdisc(&pfifo_fast_ops);
1952         register_qdisc(&pfifo_qdisc_ops);
1953         register_qdisc(&bfifo_qdisc_ops);
1954         register_qdisc(&pfifo_head_drop_qdisc_ops);
1955         register_qdisc(&mq_qdisc_ops);
1956         register_qdisc(&noqueue_qdisc_ops);
1957
1958         rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1959         rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1960         rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1961         rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1962         rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1963         rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1964
1965         return 0;
1966 }
1967
1968 subsys_initcall(pktsched_init);