kernel/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    NET_XMIT_POLICED     - dropped by police.
  99      Expected action: backoff or error to real-time apps.
 100
 101    Auxiliary routines:
 102
 103    ---peek
 104
 105    like dequeue but without removing a packet from the queue
 106
 107    ---reset
 108
 109    returns qdisc to initial state: purge all buffers, clear all
 110    timers, counters (except for statistics) etc.
 111
 112    ---init
 113
 114    initializes newly created qdisc.
 115
 116    ---destroy
 117
 118    destroys resources allocated by init and during lifetime of qdisc.
 119
 120    ---change
 121
 122    changes qdisc parameters.
 123  */
 124
 125 /* Protects list of registered TC modules. It is pure SMP lock. */
 126 static DEFINE_RWLOCK(qdisc_mod_lock);
 127
 128
 129 /************************************************
 130  *      Queueing disciplines manipulation.      *
 131  ************************************************/
 132
 133
 134 /* The list of all installed queueing disciplines. */
 135
 136 static struct Qdisc_ops *qdisc_base;
 137
 138 /* Register/unregister queueing discipline */
 139
 140 int register_qdisc(struct Qdisc_ops *qops)
 141 {
 142         struct Qdisc_ops *q, **qp;
 143         int rc = -EEXIST;
 144
 145         write_lock(&qdisc_mod_lock);
 146         for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 147                 if (!strcmp(qops->id, q->id))
 148                         goto out;
 149
 150         if (qops->enqueue == NULL)
 151                 qops->enqueue = noop_qdisc_ops.enqueue;
 152         if (qops->peek == NULL) {
 153                 if (qops->dequeue == NULL)
 154                         qops->peek = noop_qdisc_ops.peek;
 155                 else
 156                         goto out_einval;
 157         }
 158         if (qops->dequeue == NULL)
 159                 qops->dequeue = noop_qdisc_ops.dequeue;
 160
 161         if (qops->cl_ops) {
 162                 const struct Qdisc_class_ops *cops = qops->cl_ops;
 163
 164                 if (!(cops->get && cops->put && cops->walk && cops->leaf))
 165                         goto out_einval;
 166
 167                 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
 168                         goto out_einval;
 169         }
 170
 171         qops->next = NULL;
 172         *qp = qops;
 173         rc = 0;
 174 out:
 175         write_unlock(&qdisc_mod_lock);
 176         return rc;
 177
 178 out_einval:
 179         rc = -EINVAL;
 180         goto out;
 181 }
 182 EXPORT_SYMBOL(register_qdisc);
 183
 184 int unregister_qdisc(struct Qdisc_ops *qops)
 185 {
 186         struct Qdisc_ops *q, **qp;
 187         int err = -ENOENT;
 188
 189         write_lock(&qdisc_mod_lock);
 190         for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
 191                 if (q == qops)
 192                         break;
 193         if (q) {
 194                 *qp = q->next;
 195                 q->next = NULL;
 196                 err = 0;
 197         }
 198         write_unlock(&qdisc_mod_lock);
 199         return err;
 200 }
 201 EXPORT_SYMBOL(unregister_qdisc);
 202
 203 /* Get default qdisc if not otherwise specified */
 204 void qdisc_get_default(char *name, size_t len)
 205 {
 206         read_lock(&qdisc_mod_lock);
 207         strlcpy(name, default_qdisc_ops->id, len);
 208         read_unlock(&qdisc_mod_lock);
 209 }
 210
 211 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
 212 {
 213         struct Qdisc_ops *q = NULL;
 214
 215         for (q = qdisc_base; q; q = q->next) {
 216                 if (!strcmp(name, q->id)) {
 217                         if (!try_module_get(q->owner))
 218                                 q = NULL;
 219                         break;
 220                 }
 221         }
 222
 223         return q;
 224 }
 225
 226 /* Set new default qdisc to use */
 227 int qdisc_set_default(const char *name)
 228 {
 229         const struct Qdisc_ops *ops;
 230
 231         if (!capable(CAP_NET_ADMIN))
 232                 return -EPERM;
 233
 234         write_lock(&qdisc_mod_lock);
 235         ops = qdisc_lookup_default(name);
 236         if (!ops) {
 237                 /* Not found, drop lock and try to load module */
 238                 write_unlock(&qdisc_mod_lock);
 239                 request_module("sch_%s", name);
 240                 write_lock(&qdisc_mod_lock);
 241
 242                 ops = qdisc_lookup_default(name);
 243         }
 244
 245         if (ops) {
 246                 /* Set new default */
 247                 module_put(default_qdisc_ops->owner);
 248                 default_qdisc_ops = ops;
 249         }
 250         write_unlock(&qdisc_mod_lock);
 251
 252         return ops ? 0 : -ENOENT;
 253 }
 254
 255 /* We know handle. Find qdisc among all qdisc's attached to device
 256  * (root qdisc, all its children, children of children etc.)
 257  * Note: caller either uses rtnl or rcu_read_lock()
 258  */
 259
 260 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
 261 {
 262         struct Qdisc *q;
 263
 264         if (!(root->flags & TCQ_F_BUILTIN) &&
 265             root->handle == handle)
 266                 return root;
 267
 268         list_for_each_entry_rcu(q, &root->list, list) {
 269                 if (q->handle == handle)
 270                         return q;
 271         }
 272         return NULL;
 273 }
 274
 275 void qdisc_list_add(struct Qdisc *q)
 276 {
 277         if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 278                 struct Qdisc *root = qdisc_dev(q)->qdisc;
 279
 280                 WARN_ON_ONCE(root == &noop_qdisc);
 281                 ASSERT_RTNL();
 282                 list_add_tail_rcu(&q->list, &root->list);
 283         }
 284 }
 285 EXPORT_SYMBOL(qdisc_list_add);
 286
 287 void qdisc_list_del(struct Qdisc *q)
 288 {
 289         if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
 290                 ASSERT_RTNL();
 291                 list_del_rcu(&q->list);
 292         }
 293 }
 294 EXPORT_SYMBOL(qdisc_list_del);
 295
 296 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
 297 {
 298         struct Qdisc *q;
 299
 300         q = qdisc_match_from_root(dev->qdisc, handle);
 301         if (q)
 302                 goto out;
 303
 304         if (dev_ingress_queue(dev))
 305                 q = qdisc_match_from_root(
 306                         dev_ingress_queue(dev)->qdisc_sleeping,
 307                         handle);
 308 out:
 309         return q;
 310 }
 311
 312 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
 313 {
 314         unsigned long cl;
 315         struct Qdisc *leaf;
 316         const struct Qdisc_class_ops *cops = p->ops->cl_ops;
 317
 318         if (cops == NULL)
 319                 return NULL;
 320         cl = cops->get(p, classid);
 321
 322         if (cl == 0)
 323                 return NULL;
 324         leaf = cops->leaf(p, cl);
 325         cops->put(p, cl);
 326         return leaf;
 327 }
 328
 329 /* Find queueing discipline by name */
 330
 331 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
 332 {
 333         struct Qdisc_ops *q = NULL;
 334
 335         if (kind) {
 336                 read_lock(&qdisc_mod_lock);
 337                 for (q = qdisc_base; q; q = q->next) {
 338                         if (nla_strcmp(kind, q->id) == 0) {
 339                                 if (!try_module_get(q->owner))
 340                                         q = NULL;
 341                                 break;
 342                         }
 343                 }
 344                 read_unlock(&qdisc_mod_lock);
 345         }
 346         return q;
 347 }
 348
 349 /* The linklayer setting were not transferred from iproute2, in older
 350  * versions, and the rate tables lookup systems have been dropped in
 351  * the kernel. To keep backward compatible with older iproute2 tc
 352  * utils, we detect the linklayer setting by detecting if the rate
 353  * table were modified.
 354  *
 355  * For linklayer ATM table entries, the rate table will be aligned to
 356  * 48 bytes, thus some table entries will contain the same value.  The
 357  * mpu (min packet unit) is also encoded into the old rate table, thus
 358  * starting from the mpu, we find low and high table entries for
 359  * mapping this cell.  If these entries contain the same value, when
 360  * the rate tables have been modified for linklayer ATM.
 361  *
 362  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
 363  * and then roundup to the next cell, calc the table entry one below,
 364  * and compare.
 365  */
 366 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
 367 {
 368         int low       = roundup(r->mpu, 48);
 369         int high      = roundup(low+1, 48);
 370         int cell_low  = low >> r->cell_log;
 371         int cell_high = (high >> r->cell_log) - 1;
 372
 373         /* rtab is too inaccurate at rates > 100Mbit/s */
 374         if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
 375                 pr_debug("TC linklayer: Giving up ATM detection\n");
 376                 return TC_LINKLAYER_ETHERNET;
 377         }
 378
 379         if ((cell_high > cell_low) && (cell_high < 256)
 380             && (rtab[cell_low] == rtab[cell_high])) {
 381                 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
 382                          cell_low, cell_high, rtab[cell_high]);
 383                 return TC_LINKLAYER_ATM;
 384         }
 385         return TC_LINKLAYER_ETHERNET;
 386 }
 387
 388 static struct qdisc_rate_table *qdisc_rtab_list;
 389
 390 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
 391 {
 392         struct qdisc_rate_table *rtab;
 393
 394         if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
 395             nla_len(tab) != TC_RTAB_SIZE)
 396                 return NULL;
 397
 398         for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
 399                 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
 400                     !memcmp(&rtab->data, nla_data(tab), 1024)) {
 401                         rtab->refcnt++;
 402                         return rtab;
 403                 }
 404         }
 405
 406         rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
 407         if (rtab) {
 408                 rtab->rate = *r;
 409                 rtab->refcnt = 1;
 410                 memcpy(rtab->data, nla_data(tab), 1024);
 411                 if (r->linklayer == TC_LINKLAYER_UNAWARE)
 412                         r->linklayer = __detect_linklayer(r, rtab->data);
 413                 rtab->next = qdisc_rtab_list;
 414                 qdisc_rtab_list = rtab;
 415         }
 416         return rtab;
 417 }
 418 EXPORT_SYMBOL(qdisc_get_rtab);
 419
 420 void qdisc_put_rtab(struct qdisc_rate_table *tab)
 421 {
 422         struct qdisc_rate_table *rtab, **rtabp;
 423
 424         if (!tab || --tab->refcnt)
 425                 return;
 426
 427         for (rtabp = &qdisc_rtab_list;
 428              (rtab = *rtabp) != NULL;
 429              rtabp = &rtab->next) {
 430                 if (rtab == tab) {
 431                         *rtabp = rtab->next;
 432                         kfree(rtab);
 433                         return;
 434                 }
 435         }
 436 }
 437 EXPORT_SYMBOL(qdisc_put_rtab);
 438
 439 static LIST_HEAD(qdisc_stab_list);
 440 static DEFINE_SPINLOCK(qdisc_stab_lock);
 441
 442 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
 443         [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
 444         [TCA_STAB_DATA] = { .type = NLA_BINARY },
 445 };
 446
 447 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
 448 {
 449         struct nlattr *tb[TCA_STAB_MAX + 1];
 450         struct qdisc_size_table *stab;
 451         struct tc_sizespec *s;
 452         unsigned int tsize = 0;
 453         u16 *tab = NULL;
 454         int err;
 455
 456         err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
 457         if (err < 0)
 458                 return ERR_PTR(err);
 459         if (!tb[TCA_STAB_BASE])
 460                 return ERR_PTR(-EINVAL);
 461
 462         s = nla_data(tb[TCA_STAB_BASE]);
 463
 464         if (s->tsize > 0) {
 465                 if (!tb[TCA_STAB_DATA])
 466                         return ERR_PTR(-EINVAL);
 467                 tab = nla_data(tb[TCA_STAB_DATA]);
 468                 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
 469         }
 470
 471         if (tsize != s->tsize || (!tab && tsize > 0))
 472                 return ERR_PTR(-EINVAL);
 473
 474         spin_lock(&qdisc_stab_lock);
 475
 476         list_for_each_entry(stab, &qdisc_stab_list, list) {
 477                 if (memcmp(&stab->szopts, s, sizeof(*s)))
 478                         continue;
 479                 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
 480                         continue;
 481                 stab->refcnt++;
 482                 spin_unlock(&qdisc_stab_lock);
 483                 return stab;
 484         }
 485
 486         spin_unlock(&qdisc_stab_lock);
 487
 488         stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
 489         if (!stab)
 490                 return ERR_PTR(-ENOMEM);
 491
 492         stab->refcnt = 1;
 493         stab->szopts = *s;
 494         if (tsize > 0)
 495                 memcpy(stab->data, tab, tsize * sizeof(u16));
 496
 497         spin_lock(&qdisc_stab_lock);
 498         list_add_tail(&stab->list, &qdisc_stab_list);
 499         spin_unlock(&qdisc_stab_lock);
 500
 501         return stab;
 502 }
 503
 504 static void stab_kfree_rcu(struct rcu_head *head)
 505 {
 506         kfree(container_of(head, struct qdisc_size_table, rcu));
 507 }
 508
 509 void qdisc_put_stab(struct qdisc_size_table *tab)
 510 {
 511         if (!tab)
 512                 return;
 513
 514         spin_lock(&qdisc_stab_lock);
 515
 516         if (--tab->refcnt == 0) {
 517                 list_del(&tab->list);
 518                 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
 519         }
 520
 521         spin_unlock(&qdisc_stab_lock);
 522 }
 523 EXPORT_SYMBOL(qdisc_put_stab);
 524
 525 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
 526 {
 527         struct nlattr *nest;
 528
 529         nest = nla_nest_start(skb, TCA_STAB);
 530         if (nest == NULL)
 531                 goto nla_put_failure;
 532         if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
 533                 goto nla_put_failure;
 534         nla_nest_end(skb, nest);
 535
 536         return skb->len;
 537
 538 nla_put_failure:
 539         return -1;
 540 }
 541
 542 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
 543 {
 544         int pkt_len, slot;
 545
 546         pkt_len = skb->len + stab->szopts.overhead;
 547         if (unlikely(!stab->szopts.tsize))
 548                 goto out;
 549
 550         slot = pkt_len + stab->szopts.cell_align;
 551         if (unlikely(slot < 0))
 552                 slot = 0;
 553
 554         slot >>= stab->szopts.cell_log;
 555         if (likely(slot < stab->szopts.tsize))
 556                 pkt_len = stab->data[slot];
 557         else
 558                 pkt_len = stab->data[stab->szopts.tsize - 1] *
 559                                 (slot / stab->szopts.tsize) +
 560                                 stab->data[slot % stab->szopts.tsize];
 561
 562         pkt_len <<= stab->szopts.size_log;
 563 out:
 564         if (unlikely(pkt_len < 1))
 565                 pkt_len = 1;
 566         qdisc_skb_cb(skb)->pkt_len = pkt_len;
 567 }
 568 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
 569
 570 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
 571 {
 572         if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
 573                 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
 574                         txt, qdisc->ops->id, qdisc->handle >> 16);
 575                 qdisc->flags |= TCQ_F_WARN_NONWC;
 576         }
 577 }
 578 EXPORT_SYMBOL(qdisc_warn_nonwc);
 579
 580 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
 581 {
 582         struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
 583                                                  timer);
 584
 585         rcu_read_lock();
 586         qdisc_unthrottled(wd->qdisc);
 587         __netif_schedule(qdisc_root(wd->qdisc));
 588         rcu_read_unlock();
 589
 590         return HRTIMER_NORESTART;
 591 }
 592
 593 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
 594 {
 595         hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
 596         wd->timer.function = qdisc_watchdog;
 597         wd->qdisc = qdisc;
 598 }
 599 EXPORT_SYMBOL(qdisc_watchdog_init);
 600
 601 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle)
 602 {
 603         if (test_bit(__QDISC_STATE_DEACTIVATED,
 604                      &qdisc_root_sleeping(wd->qdisc)->state))
 605                 return;
 606
 607         if (throttle)
 608                 qdisc_throttled(wd->qdisc);
 609
 610         hrtimer_start(&wd->timer,
 611                       ns_to_ktime(expires),
 612                       HRTIMER_MODE_ABS_PINNED);
 613 }
 614 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
 615
 616 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
 617 {
 618         hrtimer_cancel(&wd->timer);
 619         qdisc_unthrottled(wd->qdisc);
 620 }
 621 EXPORT_SYMBOL(qdisc_watchdog_cancel);
 622
 623 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
 624 {
 625         unsigned int size = n * sizeof(struct hlist_head), i;
 626         struct hlist_head *h;
 627
 628         if (size <= PAGE_SIZE)
 629                 h = kmalloc(size, GFP_KERNEL);
 630         else
 631                 h = (struct hlist_head *)
 632                         __get_free_pages(GFP_KERNEL, get_order(size));
 633
 634         if (h != NULL) {
 635                 for (i = 0; i < n; i++)
 636                         INIT_HLIST_HEAD(&h[i]);
 637         }
 638         return h;
 639 }
 640
 641 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
 642 {
 643         unsigned int size = n * sizeof(struct hlist_head);
 644
 645         if (size <= PAGE_SIZE)
 646                 kfree(h);
 647         else
 648                 free_pages((unsigned long)h, get_order(size));
 649 }
 650
 651 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
 652 {
 653         struct Qdisc_class_common *cl;
 654         struct hlist_node *next;
 655         struct hlist_head *nhash, *ohash;
 656         unsigned int nsize, nmask, osize;
 657         unsigned int i, h;
 658
 659         /* Rehash when load factor exceeds 0.75 */
 660         if (clhash->hashelems * 4 <= clhash->hashsize * 3)
 661                 return;
 662         nsize = clhash->hashsize * 2;
 663         nmask = nsize - 1;
 664         nhash = qdisc_class_hash_alloc(nsize);
 665         if (nhash == NULL)
 666                 return;
 667
 668         ohash = clhash->hash;
 669         osize = clhash->hashsize;
 670
 671         sch_tree_lock(sch);
 672         for (i = 0; i < osize; i++) {
 673                 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
 674                         h = qdisc_class_hash(cl->classid, nmask);
 675                         hlist_add_head(&cl->hnode, &nhash[h]);
 676                 }
 677         }
 678         clhash->hash     = nhash;
 679         clhash->hashsize = nsize;
 680         clhash->hashmask = nmask;
 681         sch_tree_unlock(sch);
 682
 683         qdisc_class_hash_free(ohash, osize);
 684 }
 685 EXPORT_SYMBOL(qdisc_class_hash_grow);
 686
 687 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
 688 {
 689         unsigned int size = 4;
 690
 691         clhash->hash = qdisc_class_hash_alloc(size);
 692         if (clhash->hash == NULL)
 693                 return -ENOMEM;
 694         clhash->hashsize  = size;
 695         clhash->hashmask  = size - 1;
 696         clhash->hashelems = 0;
 697         return 0;
 698 }
 699 EXPORT_SYMBOL(qdisc_class_hash_init);
 700
 701 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
 702 {
 703         qdisc_class_hash_free(clhash->hash, clhash->hashsize);
 704 }
 705 EXPORT_SYMBOL(qdisc_class_hash_destroy);
 706
 707 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
 708                              struct Qdisc_class_common *cl)
 709 {
 710         unsigned int h;
 711
 712         INIT_HLIST_NODE(&cl->hnode);
 713         h = qdisc_class_hash(cl->classid, clhash->hashmask);
 714         hlist_add_head(&cl->hnode, &clhash->hash[h]);
 715         clhash->hashelems++;
 716 }
 717 EXPORT_SYMBOL(qdisc_class_hash_insert);
 718
 719 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
 720                              struct Qdisc_class_common *cl)
 721 {
 722         hlist_del(&cl->hnode);
 723         clhash->hashelems--;
 724 }
 725 EXPORT_SYMBOL(qdisc_class_hash_remove);
 726
 727 /* Allocate an unique handle from space managed by kernel
 728  * Possible range is [8000-FFFF]:0000 (0x8000 values)
 729  */
 730 static u32 qdisc_alloc_handle(struct net_device *dev)
 731 {
 732         int i = 0x8000;
 733         static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
 734
 735         do {
 736                 autohandle += TC_H_MAKE(0x10000U, 0);
 737                 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
 738                         autohandle = TC_H_MAKE(0x80000000U, 0);
 739                 if (!qdisc_lookup(dev, autohandle))
 740                         return autohandle;
 741                 cond_resched();
 742         } while (--i > 0);
 743
 744         return 0;
 745 }
 746
 747 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
 748                                unsigned int len)
 749 {
 750         const struct Qdisc_class_ops *cops;
 751         unsigned long cl;
 752         u32 parentid;
 753         int drops;
 754
 755         if (n == 0 && len == 0)
 756                 return;
 757         drops = max_t(int, n, 0);
 758         rcu_read_lock();
 759         while ((parentid = sch->parent)) {
 760                 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
 761                         break;
 762
 763                 if (sch->flags & TCQ_F_NOPARENT)
 764                         break;
 765                 /* TODO: perform the search on a per txq basis */
 766                 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
 767                 if (sch == NULL) {
 768                         WARN_ON_ONCE(parentid != TC_H_ROOT);
 769                         break;
 770                 }
 771                 cops = sch->ops->cl_ops;
 772                 if (cops->qlen_notify) {
 773                         cl = cops->get(sch, parentid);
 774                         cops->qlen_notify(sch, cl);
 775                         cops->put(sch, cl);
 776                 }
 777                 sch->q.qlen -= n;
 778                 sch->qstats.backlog -= len;
 779                 __qdisc_qstats_drop(sch, drops);
 780         }
 781         rcu_read_unlock();
 782 }
 783 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
 784
 785 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
 786                                struct nlmsghdr *n, u32 clid,
 787                                struct Qdisc *old, struct Qdisc *new)
 788 {
 789         if (new || old)
 790                 qdisc_notify(net, skb, n, clid, old, new);
 791
 792         if (old)
 793                 qdisc_destroy(old);
 794 }
 795
 796 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
 797  * to device "dev".
 798  *
 799  * When appropriate send a netlink notification using 'skb'
 800  * and "n".
 801  *
 802  * On success, destroy old qdisc.
 803  */
 804
 805 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
 806                        struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
 807                        struct Qdisc *new, struct Qdisc *old)
 808 {
 809         struct Qdisc *q = old;
 810         struct net *net = dev_net(dev);
 811         int err = 0;
 812
 813         if (parent == NULL) {
 814                 unsigned int i, num_q, ingress;
 815
 816                 ingress = 0;
 817                 num_q = dev->num_tx_queues;
 818                 if ((q && q->flags & TCQ_F_INGRESS) ||
 819                     (new && new->flags & TCQ_F_INGRESS)) {
 820                         num_q = 1;
 821                         ingress = 1;
 822                         if (!dev_ingress_queue(dev))
 823                                 return -ENOENT;
 824                 }
 825
 826                 if (dev->flags & IFF_UP)
 827                         dev_deactivate(dev);
 828
 829                 if (new && new->ops->attach)
 830                         goto skip;
 831
 832                 for (i = 0; i < num_q; i++) {
 833                         struct netdev_queue *dev_queue = dev_ingress_queue(dev);
 834
 835                         if (!ingress)
 836                                 dev_queue = netdev_get_tx_queue(dev, i);
 837
 838                         old = dev_graft_qdisc(dev_queue, new);
 839                         if (new && i > 0)
 840                                 atomic_inc(&new->refcnt);
 841
 842                         if (!ingress)
 843                                 qdisc_destroy(old);
 844                 }
 845
 846 skip:
 847                 if (!ingress) {
 848                         notify_and_destroy(net, skb, n, classid,
 849                                            dev->qdisc, new);
 850                         if (new && !new->ops->attach)
 851                                 atomic_inc(&new->refcnt);
 852                         dev->qdisc = new ? : &noop_qdisc;
 853
 854                         if (new && new->ops->attach)
 855                                 new->ops->attach(new);
 856                 } else {
 857                         notify_and_destroy(net, skb, n, classid, old, new);
 858                 }
 859
 860                 if (dev->flags & IFF_UP)
 861                         dev_activate(dev);
 862         } else {
 863                 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
 864
 865                 err = -EOPNOTSUPP;
 866                 if (cops && cops->graft) {
 867                         unsigned long cl = cops->get(parent, classid);
 868                         if (cl) {
 869                                 err = cops->graft(parent, cl, new, &old);
 870                                 cops->put(parent, cl);
 871                         } else
 872                                 err = -ENOENT;
 873                 }
 874                 if (!err)
 875                         notify_and_destroy(net, skb, n, classid, old, new);
 876         }
 877         return err;
 878 }
 879
 880 /* lockdep annotation is needed for ingress; egress gets it only for name */
 881 static struct lock_class_key qdisc_tx_lock;
 882 static struct lock_class_key qdisc_rx_lock;
 883
 884 /*
 885    Allocate and initialize new qdisc.
 886
 887    Parameters are passed via opt.
 888  */
 889
 890 static struct Qdisc *
 891 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
 892              struct Qdisc *p, u32 parent, u32 handle,
 893              struct nlattr **tca, int *errp)
 894 {
 895         int err;
 896         struct nlattr *kind = tca[TCA_KIND];
 897         struct Qdisc *sch;
 898         struct Qdisc_ops *ops;
 899         struct qdisc_size_table *stab;
 900
 901         ops = qdisc_lookup_ops(kind);
 902 #ifdef CONFIG_MODULES
 903         if (ops == NULL && kind != NULL) {
 904                 char name[IFNAMSIZ];
 905                 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
 906                         /* We dropped the RTNL semaphore in order to
 907                          * perform the module load.  So, even if we
 908                          * succeeded in loading the module we have to
 909                          * tell the caller to replay the request.  We
 910                          * indicate this using -EAGAIN.
 911                          * We replay the request because the device may
 912                          * go away in the mean time.
 913                          */
 914                         rtnl_unlock();
 915                         request_module("sch_%s", name);
 916                         rtnl_lock();
 917                         ops = qdisc_lookup_ops(kind);
 918                         if (ops != NULL) {
 919                                 /* We will try again qdisc_lookup_ops,
 920                                  * so don't keep a reference.
 921                                  */
 922                                 module_put(ops->owner);
 923                                 err = -EAGAIN;
 924                                 goto err_out;
 925                         }
 926                 }
 927         }
 928 #endif
 929
 930         err = -ENOENT;
 931         if (ops == NULL)
 932                 goto err_out;
 933
 934         sch = qdisc_alloc(dev_queue, ops);
 935         if (IS_ERR(sch)) {
 936                 err = PTR_ERR(sch);
 937                 goto err_out2;
 938         }
 939
 940         sch->parent = parent;
 941
 942         if (handle == TC_H_INGRESS) {
 943                 sch->flags |= TCQ_F_INGRESS;
 944                 handle = TC_H_MAKE(TC_H_INGRESS, 0);
 945                 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
 946         } else {
 947                 if (handle == 0) {
 948                         handle = qdisc_alloc_handle(dev);
 949                         err = -ENOMEM;
 950                         if (handle == 0)
 951                                 goto err_out3;
 952                 }
 953                 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
 954                 if (!netif_is_multiqueue(dev))
 955                         sch->flags |= TCQ_F_ONETXQUEUE;
 956         }
 957
 958         sch->handle = handle;
 959
 960         if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
 961                 if (qdisc_is_percpu_stats(sch)) {
 962                         sch->cpu_bstats =
 963                                 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
 964                         if (!sch->cpu_bstats)
 965                                 goto err_out4;
 966
 967                         sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
 968                         if (!sch->cpu_qstats)
 969                                 goto err_out4;
 970                 }
 971
 972                 if (tca[TCA_STAB]) {
 973                         stab = qdisc_get_stab(tca[TCA_STAB]);
 974                         if (IS_ERR(stab)) {
 975                                 err = PTR_ERR(stab);
 976                                 goto err_out4;
 977                         }
 978                         rcu_assign_pointer(sch->stab, stab);
 979                 }
 980                 if (tca[TCA_RATE]) {
 981                         spinlock_t *root_lock;
 982
 983                         err = -EOPNOTSUPP;
 984                         if (sch->flags & TCQ_F_MQROOT)
 985                                 goto err_out4;
 986
 987                         if ((sch->parent != TC_H_ROOT) &&
 988                             !(sch->flags & TCQ_F_INGRESS) &&
 989                             (!p || !(p->flags & TCQ_F_MQROOT)))
 990                                 root_lock = qdisc_root_sleeping_lock(sch);
 991                         else
 992                                 root_lock = qdisc_lock(sch);
 993
 994                         err = gen_new_estimator(&sch->bstats,
 995                                                 sch->cpu_bstats,
 996                                                 &sch->rate_est,
 997                                                 root_lock,
 998                                                 tca[TCA_RATE]);
 999                         if (err)
1000                                 goto err_out4;
1001                 }
1002
1003                 qdisc_list_add(sch);
1004
1005                 return sch;
1006         }
1007 err_out3:
1008         dev_put(dev);
1009         kfree((char *) sch - sch->padded);
1010 err_out2:
1011         module_put(ops->owner);
1012 err_out:
1013         *errp = err;
1014         return NULL;
1015
1016 err_out4:
1017         free_percpu(sch->cpu_bstats);
1018         free_percpu(sch->cpu_qstats);
1019         /*
1020          * Any broken qdiscs that would require a ops->reset() here?
1021          * The qdisc was never in action so it shouldn't be necessary.
1022          */
1023         qdisc_put_stab(rtnl_dereference(sch->stab));
1024         if (ops->destroy)
1025                 ops->destroy(sch);
1026         goto err_out3;
1027 }
1028
1029 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1030 {
1031         struct qdisc_size_table *ostab, *stab = NULL;
1032         int err = 0;
1033
1034         if (tca[TCA_OPTIONS]) {
1035                 if (sch->ops->change == NULL)
1036                         return -EINVAL;
1037                 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1038                 if (err)
1039                         return err;
1040         }
1041
1042         if (tca[TCA_STAB]) {
1043                 stab = qdisc_get_stab(tca[TCA_STAB]);
1044                 if (IS_ERR(stab))
1045                         return PTR_ERR(stab);
1046         }
1047
1048         ostab = rtnl_dereference(sch->stab);
1049         rcu_assign_pointer(sch->stab, stab);
1050         qdisc_put_stab(ostab);
1051
1052         if (tca[TCA_RATE]) {
1053                 /* NB: ignores errors from replace_estimator
1054                    because change can't be undone. */
1055                 if (sch->flags & TCQ_F_MQROOT)
1056                         goto out;
1057                 gen_replace_estimator(&sch->bstats,
1058                                       sch->cpu_bstats,
1059                                       &sch->rate_est,
1060                                       qdisc_root_sleeping_lock(sch),
1061                                       tca[TCA_RATE]);
1062         }
1063 out:
1064         return 0;
1065 }
1066
1067 struct check_loop_arg {
1068         struct qdisc_walker     w;
1069         struct Qdisc            *p;
1070         int                     depth;
1071 };
1072
1073 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1074
1075 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1076 {
1077         struct check_loop_arg   arg;
1078
1079         if (q->ops->cl_ops == NULL)
1080                 return 0;
1081
1082         arg.w.stop = arg.w.skip = arg.w.count = 0;
1083         arg.w.fn = check_loop_fn;
1084         arg.depth = depth;
1085         arg.p = p;
1086         q->ops->cl_ops->walk(q, &arg.w);
1087         return arg.w.stop ? -ELOOP : 0;
1088 }
1089
1090 static int
1091 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1092 {
1093         struct Qdisc *leaf;
1094         const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1095         struct check_loop_arg *arg = (struct check_loop_arg *)w;
1096
1097         leaf = cops->leaf(q, cl);
1098         if (leaf) {
1099                 if (leaf == arg->p || arg->depth > 7)
1100                         return -ELOOP;
1101                 return check_loop(leaf, arg->p, arg->depth + 1);
1102         }
1103         return 0;
1104 }
1105
1106 /*
1107  * Delete/get qdisc.
1108  */
1109
1110 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1111 {
1112         struct net *net = sock_net(skb->sk);
1113         struct tcmsg *tcm = nlmsg_data(n);
1114         struct nlattr *tca[TCA_MAX + 1];
1115         struct net_device *dev;
1116         u32 clid;
1117         struct Qdisc *q = NULL;
1118         struct Qdisc *p = NULL;
1119         int err;
1120
1121         if ((n->nlmsg_type != RTM_GETQDISC) &&
1122             !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1123                 return -EPERM;
1124
1125         err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1126         if (err < 0)
1127                 return err;
1128
1129         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1130         if (!dev)
1131                 return -ENODEV;
1132
1133         clid = tcm->tcm_parent;
1134         if (clid) {
1135                 if (clid != TC_H_ROOT) {
1136                         if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1137                                 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1138                                 if (!p)
1139                                         return -ENOENT;
1140                                 q = qdisc_leaf(p, clid);
1141                         } else if (dev_ingress_queue(dev)) {
1142                                 q = dev_ingress_queue(dev)->qdisc_sleeping;
1143                         }
1144                 } else {
1145                         q = dev->qdisc;
1146                 }
1147                 if (!q)
1148                         return -ENOENT;
1149
1150                 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1151                         return -EINVAL;
1152         } else {
1153                 q = qdisc_lookup(dev, tcm->tcm_handle);
1154                 if (!q)
1155                         return -ENOENT;
1156         }
1157
1158         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1159                 return -EINVAL;
1160
1161         if (n->nlmsg_type == RTM_DELQDISC) {
1162                 if (!clid)
1163                         return -EINVAL;
1164                 if (q->handle == 0)
1165                         return -ENOENT;
1166                 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1167                 if (err != 0)
1168                         return err;
1169         } else {
1170                 qdisc_notify(net, skb, n, clid, NULL, q);
1171         }
1172         return 0;
1173 }
1174
1175 /*
1176  * Create/change qdisc.
1177  */
1178
1179 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1180 {
1181         struct net *net = sock_net(skb->sk);
1182         struct tcmsg *tcm;
1183         struct nlattr *tca[TCA_MAX + 1];
1184         struct net_device *dev;
1185         u32 clid;
1186         struct Qdisc *q, *p;
1187         int err;
1188
1189         if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1190                 return -EPERM;
1191
1192 replay:
1193         /* Reinit, just in case something touches this. */
1194         err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1195         if (err < 0)
1196                 return err;
1197
1198         tcm = nlmsg_data(n);
1199         clid = tcm->tcm_parent;
1200         q = p = NULL;
1201
1202         dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1203         if (!dev)
1204                 return -ENODEV;
1205
1206
1207         if (clid) {
1208                 if (clid != TC_H_ROOT) {
1209                         if (clid != TC_H_INGRESS) {
1210                                 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1211                                 if (!p)
1212                                         return -ENOENT;
1213                                 q = qdisc_leaf(p, clid);
1214                         } else if (dev_ingress_queue_create(dev)) {
1215                                 q = dev_ingress_queue(dev)->qdisc_sleeping;
1216                         }
1217                 } else {
1218                         q = dev->qdisc;
1219                 }
1220
1221                 /* It may be default qdisc, ignore it */
1222                 if (q && q->handle == 0)
1223                         q = NULL;
1224
1225                 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1226                         if (tcm->tcm_handle) {
1227                                 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1228                                         return -EEXIST;
1229                                 if (TC_H_MIN(tcm->tcm_handle))
1230                                         return -EINVAL;
1231                                 q = qdisc_lookup(dev, tcm->tcm_handle);
1232                                 if (!q)
1233                                         goto create_n_graft;
1234                                 if (n->nlmsg_flags & NLM_F_EXCL)
1235                                         return -EEXIST;
1236                                 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1237                                         return -EINVAL;
1238                                 if (q == p ||
1239                                     (p && check_loop(q, p, 0)))
1240                                         return -ELOOP;
1241                                 atomic_inc(&q->refcnt);
1242                                 goto graft;
1243                         } else {
1244                                 if (!q)
1245                                         goto create_n_graft;
1246
1247                                 /* This magic test requires explanation.
1248                                  *
1249                                  *   We know, that some child q is already
1250                                  *   attached to this parent and have choice:
1251                                  *   either to change it or to create/graft new one.
1252                                  *
1253                                  *   1. We are allowed to create/graft only
1254                                  *   if CREATE and REPLACE flags are set.
1255                                  *
1256                                  *   2. If EXCL is set, requestor wanted to say,
1257                                  *   that qdisc tcm_handle is not expected
1258                                  *   to exist, so that we choose create/graft too.
1259                                  *
1260                                  *   3. The last case is when no flags are set.
1261                                  *   Alas, it is sort of hole in API, we
1262                                  *   cannot decide what to do unambiguously.
1263                                  *   For now we select create/graft, if
1264                                  *   user gave KIND, which does not match existing.
1265                                  */
1266                                 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1267                                     (n->nlmsg_flags & NLM_F_REPLACE) &&
1268                                     ((n->nlmsg_flags & NLM_F_EXCL) ||
1269                                      (tca[TCA_KIND] &&
1270                                       nla_strcmp(tca[TCA_KIND], q->ops->id))))
1271                                         goto create_n_graft;
1272                         }
1273                 }
1274         } else {
1275                 if (!tcm->tcm_handle)
1276                         return -EINVAL;
1277                 q = qdisc_lookup(dev, tcm->tcm_handle);
1278         }
1279
1280         /* Change qdisc parameters */
1281         if (q == NULL)
1282                 return -ENOENT;
1283         if (n->nlmsg_flags & NLM_F_EXCL)
1284                 return -EEXIST;
1285         if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1286                 return -EINVAL;
1287         err = qdisc_change(q, tca);
1288         if (err == 0)
1289                 qdisc_notify(net, skb, n, clid, NULL, q);
1290         return err;
1291
1292 create_n_graft:
1293         if (!(n->nlmsg_flags & NLM_F_CREATE))
1294                 return -ENOENT;
1295         if (clid == TC_H_INGRESS) {
1296                 if (dev_ingress_queue(dev))
1297                         q = qdisc_create(dev, dev_ingress_queue(dev), p,
1298                                          tcm->tcm_parent, tcm->tcm_parent,
1299                                          tca, &err);
1300                 else
1301                         err = -ENOENT;
1302         } else {
1303                 struct netdev_queue *dev_queue;
1304
1305                 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1306                         dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1307                 else if (p)
1308                         dev_queue = p->dev_queue;
1309                 else
1310                         dev_queue = netdev_get_tx_queue(dev, 0);
1311
1312                 q = qdisc_create(dev, dev_queue, p,
1313                                  tcm->tcm_parent, tcm->tcm_handle,
1314                                  tca, &err);
1315         }
1316         if (q == NULL) {
1317                 if (err == -EAGAIN)
1318                         goto replay;
1319                 return err;
1320         }
1321
1322 graft:
1323         err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1324         if (err) {
1325                 if (q)
1326                         qdisc_destroy(q);
1327                 return err;
1328         }
1329
1330         return 0;
1331 }
1332
1333 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1334                          u32 portid, u32 seq, u16 flags, int event)
1335 {
1336         struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1337         struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1338         struct tcmsg *tcm;
1339         struct nlmsghdr  *nlh;
1340         unsigned char *b = skb_tail_pointer(skb);
1341         struct gnet_dump d;
1342         struct qdisc_size_table *stab;
1343         __u32 qlen;
1344
1345         cond_resched();
1346         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1347         if (!nlh)
1348                 goto out_nlmsg_trim;
1349         tcm = nlmsg_data(nlh);
1350         tcm->tcm_family = AF_UNSPEC;
1351         tcm->tcm__pad1 = 0;
1352         tcm->tcm__pad2 = 0;
1353         tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1354         tcm->tcm_parent = clid;
1355         tcm->tcm_handle = q->handle;
1356         tcm->tcm_info = atomic_read(&q->refcnt);
1357         if (nla_put_string(skb, TCA_KIND, q->ops->id))
1358                 goto nla_put_failure;
1359         if (q->ops->dump && q->ops->dump(q, skb) < 0)
1360                 goto nla_put_failure;
1361         qlen = q->q.qlen;
1362
1363         stab = rtnl_dereference(q->stab);
1364         if (stab && qdisc_dump_stab(skb, stab) < 0)
1365                 goto nla_put_failure;
1366
1367         if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1368                                          qdisc_root_sleeping_lock(q), &d) < 0)
1369                 goto nla_put_failure;
1370
1371         if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1372                 goto nla_put_failure;
1373
1374         if (qdisc_is_percpu_stats(q)) {
1375                 cpu_bstats = q->cpu_bstats;
1376                 cpu_qstats = q->cpu_qstats;
1377         }
1378
1379         if (gnet_stats_copy_basic(&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                                          qdisc_root_sleeping_lock(q), &d) < 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 -1;
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);