net/sched/sch_generic.c

   1 /*
   2  * net/sched/sch_generic.c      Generic packet scheduler routines.
   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  *              Jamal Hadi Salim, <hadi@cyberus.ca> 990601
  11  *              - Ingress support
  12  */
  13
  14 #include <linux/bitops.h>
  15 #include <linux/module.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/sched.h>
  19 #include <linux/string.h>
  20 #include <linux/errno.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/rtnetlink.h>
  24 #include <linux/init.h>
  25 #include <linux/rcupdate.h>
  26 #include <linux/list.h>
  27 #include <net/pkt_sched.h>
  28
  29 /* Main transmission queue. */
  30
  31 /* Modifications to data participating in scheduling must be protected with
  32  * qdisc_lock(qdisc) spinlock.
  33  *
  34  * The idea is the following:
  35  * - enqueue, dequeue are serialized via qdisc root lock
  36  * - ingress filtering is also serialized via qdisc root lock
  37  * - updates to tree and tree walking are only done under the rtnl mutex.
  38  */
  39
  40 static inline int qdisc_qlen(struct Qdisc *q)
  41 {
  42         return q->q.qlen;
  43 }
  44
  45 static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
  46 {
  47         q->gso_skb = skb;
  48         q->qstats.requeues++;
  49         __netif_schedule(q);
  50
  51         return 0;
  52 }
  53
  54 static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
  55 {
  56         struct sk_buff *skb = q->gso_skb;
  57
  58         if (unlikely(skb)) {
  59                 struct net_device *dev = qdisc_dev(q);
  60                 struct netdev_queue *txq;
  61
  62                 /* check the reason of requeuing without tx lock first */
  63                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  64                 if (!netif_tx_queue_stopped(txq) && !netif_tx_queue_frozen(txq))
  65                         q->gso_skb = NULL;
  66                 else
  67                         skb = NULL;
  68         } else {
  69                 skb = q->dequeue(q);
  70         }
  71
  72         return skb;
  73 }
  74
  75 static inline int handle_dev_cpu_collision(struct sk_buff *skb,
  76                                            struct netdev_queue *dev_queue,
  77                                            struct Qdisc *q)
  78 {
  79         int ret;
  80
  81         if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
  82                 /*
  83                  * Same CPU holding the lock. It may be a transient
  84                  * configuration error, when hard_start_xmit() recurses. We
  85                  * detect it by checking xmit owner and drop the packet when
  86                  * deadloop is detected. Return OK to try the next skb.
  87                  */
  88                 kfree_skb(skb);
  89                 if (net_ratelimit())
  90                         printk(KERN_WARNING "Dead loop on netdevice %s, "
  91                                "fix it urgently!\n", dev_queue->dev->name);
  92                 ret = qdisc_qlen(q);
  93         } else {
  94                 /*
  95                  * Another cpu is holding lock, requeue & delay xmits for
  96                  * some time.
  97                  */
  98                 __get_cpu_var(netdev_rx_stat).cpu_collision++;
  99                 ret = dev_requeue_skb(skb, q);
 100         }
 101
 102         return ret;
 103 }
 104
 105 /*
 106  * NOTE: Called under qdisc_lock(q) with locally disabled BH.
 107  *
 108  * __QDISC_STATE_RUNNING guarantees only one CPU can process
 109  * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
 110  * this queue.
 111  *
 112  *  netif_tx_lock serializes accesses to device driver.
 113  *
 114  *  qdisc_lock(q) and netif_tx_lock are mutually exclusive,
 115  *  if one is grabbed, another must be free.
 116  *
 117  * Note, that this procedure can be called by a watchdog timer
 118  *
 119  * Returns to the caller:
 120  *                              0  - queue is empty or throttled.
 121  *                              >0 - queue is not empty.
 122  *
 123  */
 124 static inline int qdisc_restart(struct Qdisc *q)
 125 {
 126         struct netdev_queue *txq;
 127         int ret = NETDEV_TX_BUSY;
 128         struct net_device *dev;
 129         spinlock_t *root_lock;
 130         struct sk_buff *skb;
 131
 132         /* Dequeue packet */
 133         if (unlikely((skb = dequeue_skb(q)) == NULL))
 134                 return 0;
 135
 136         root_lock = qdisc_lock(q);
 137
 138         /* And release qdisc */
 139         spin_unlock(root_lock);
 140
 141         dev = qdisc_dev(q);
 142         txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 143
 144         HARD_TX_LOCK(dev, txq, smp_processor_id());
 145         if (!netif_tx_queue_stopped(txq) &&
 146             !netif_tx_queue_frozen(txq))
 147                 ret = dev_hard_start_xmit(skb, dev, txq);
 148         HARD_TX_UNLOCK(dev, txq);
 149
 150         spin_lock(root_lock);
 151
 152         switch (ret) {
 153         case NETDEV_TX_OK:
 154                 /* Driver sent out skb successfully */
 155                 ret = qdisc_qlen(q);
 156                 break;
 157
 158         case NETDEV_TX_LOCKED:
 159                 /* Driver try lock failed */
 160                 ret = handle_dev_cpu_collision(skb, txq, q);
 161                 break;
 162
 163         default:
 164                 /* Driver returned NETDEV_TX_BUSY - requeue skb */
 165                 if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
 166                         printk(KERN_WARNING "BUG %s code %d qlen %d\n",
 167                                dev->name, ret, q->q.qlen);
 168
 169                 ret = dev_requeue_skb(skb, q);
 170                 break;
 171         }
 172
 173         if (ret && (netif_tx_queue_stopped(txq) ||
 174                     netif_tx_queue_frozen(txq)))
 175                 ret = 0;
 176
 177         return ret;
 178 }
 179
 180 void __qdisc_run(struct Qdisc *q)
 181 {
 182         unsigned long start_time = jiffies;
 183
 184         while (qdisc_restart(q)) {
 185                 /*
 186                  * Postpone processing if
 187                  * 1. another process needs the CPU;
 188                  * 2. we've been doing it for too long.
 189                  */
 190                 if (need_resched() || jiffies != start_time) {
 191                         __netif_schedule(q);
 192                         break;
 193                 }
 194         }
 195
 196         clear_bit(__QDISC_STATE_RUNNING, &q->state);
 197 }
 198
 199 unsigned long dev_trans_start(struct net_device *dev)
 200 {
 201         unsigned long val, res = dev->trans_start;
 202         unsigned int i;
 203
 204         for (i = 0; i < dev->num_tx_queues; i++) {
 205                 val = netdev_get_tx_queue(dev, i)->trans_start;
 206                 if (val && time_after(val, res))
 207                         res = val;
 208         }
 209         dev->trans_start = res;
 210         return res;
 211 }
 212 EXPORT_SYMBOL(dev_trans_start);
 213
 214 static void dev_watchdog(unsigned long arg)
 215 {
 216         struct net_device *dev = (struct net_device *)arg;
 217
 218         netif_tx_lock(dev);
 219         if (!qdisc_tx_is_noop(dev)) {
 220                 if (netif_device_present(dev) &&
 221                     netif_running(dev) &&
 222                     netif_carrier_ok(dev)) {
 223                         int some_queue_timedout = 0;
 224                         unsigned int i;
 225                         unsigned long trans_start;
 226
 227                         for (i = 0; i < dev->num_tx_queues; i++) {
 228                                 struct netdev_queue *txq;
 229
 230                                 txq = netdev_get_tx_queue(dev, i);
 231                                 /*
 232                                  * old device drivers set dev->trans_start
 233                                  */
 234                                 trans_start = txq->trans_start ? : dev->trans_start;
 235                                 if (netif_tx_queue_stopped(txq) &&
 236                                     time_after(jiffies, (trans_start +
 237                                                          dev->watchdog_timeo))) {
 238                                         some_queue_timedout = 1;
 239                                         break;
 240                                 }
 241                         }
 242
 243                         if (some_queue_timedout) {
 244                                 char drivername[64];
 245                                 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
 246                                        dev->name, netdev_drivername(dev, drivername, 64), i);
 247                                 dev->netdev_ops->ndo_tx_timeout(dev);
 248                         }
 249                         if (!mod_timer(&dev->watchdog_timer,
 250                                        round_jiffies(jiffies +
 251                                                      dev->watchdog_timeo)))
 252                                 dev_hold(dev);
 253                 }
 254         }
 255         netif_tx_unlock(dev);
 256
 257         dev_put(dev);
 258 }
 259
 260 void __netdev_watchdog_up(struct net_device *dev)
 261 {
 262         if (dev->netdev_ops->ndo_tx_timeout) {
 263                 if (dev->watchdog_timeo <= 0)
 264                         dev->watchdog_timeo = 5*HZ;
 265                 if (!mod_timer(&dev->watchdog_timer,
 266                                round_jiffies(jiffies + dev->watchdog_timeo)))
 267                         dev_hold(dev);
 268         }
 269 }
 270
 271 static void dev_watchdog_up(struct net_device *dev)
 272 {
 273         __netdev_watchdog_up(dev);
 274 }
 275
 276 static void dev_watchdog_down(struct net_device *dev)
 277 {
 278         netif_tx_lock_bh(dev);
 279         if (del_timer(&dev->watchdog_timer))
 280                 dev_put(dev);
 281         netif_tx_unlock_bh(dev);
 282 }
 283
 284 /**
 285  *      netif_carrier_on - set carrier
 286  *      @dev: network device
 287  *
 288  * Device has detected that carrier.
 289  */
 290 void netif_carrier_on(struct net_device *dev)
 291 {
 292         if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
 293                 if (dev->reg_state == NETREG_UNINITIALIZED)
 294                         return;
 295                 linkwatch_fire_event(dev);
 296                 if (netif_running(dev))
 297                         __netdev_watchdog_up(dev);
 298         }
 299 }
 300 EXPORT_SYMBOL(netif_carrier_on);
 301
 302 /**
 303  *      netif_carrier_off - clear carrier
 304  *      @dev: network device
 305  *
 306  * Device has detected loss of carrier.
 307  */
 308 void netif_carrier_off(struct net_device *dev)
 309 {
 310         if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
 311                 if (dev->reg_state == NETREG_UNINITIALIZED)
 312                         return;
 313                 linkwatch_fire_event(dev);
 314         }
 315 }
 316 EXPORT_SYMBOL(netif_carrier_off);
 317
 318 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
 319    under all circumstances. It is difficult to invent anything faster or
 320    cheaper.
 321  */
 322
 323 static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
 324 {
 325         kfree_skb(skb);
 326         return NET_XMIT_CN;
 327 }
 328
 329 static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
 330 {
 331         return NULL;
 332 }
 333
 334 struct Qdisc_ops noop_qdisc_ops __read_mostly = {
 335         .id             =       "noop",
 336         .priv_size      =       0,
 337         .enqueue        =       noop_enqueue,
 338         .dequeue        =       noop_dequeue,
 339         .peek           =       noop_dequeue,
 340         .owner          =       THIS_MODULE,
 341 };
 342
 343 static struct netdev_queue noop_netdev_queue = {
 344         .qdisc          =       &noop_qdisc,
 345         .qdisc_sleeping =       &noop_qdisc,
 346 };
 347
 348 struct Qdisc noop_qdisc = {
 349         .enqueue        =       noop_enqueue,
 350         .dequeue        =       noop_dequeue,
 351         .flags          =       TCQ_F_BUILTIN,
 352         .ops            =       &noop_qdisc_ops,
 353         .list           =       LIST_HEAD_INIT(noop_qdisc.list),
 354         .q.lock         =       __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
 355         .dev_queue      =       &noop_netdev_queue,
 356 };
 357 EXPORT_SYMBOL(noop_qdisc);
 358
 359 static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
 360         .id             =       "noqueue",
 361         .priv_size      =       0,
 362         .enqueue        =       noop_enqueue,
 363         .dequeue        =       noop_dequeue,
 364         .peek           =       noop_dequeue,
 365         .owner          =       THIS_MODULE,
 366 };
 367
 368 static struct Qdisc noqueue_qdisc;
 369 static struct netdev_queue noqueue_netdev_queue = {
 370         .qdisc          =       &noqueue_qdisc,
 371         .qdisc_sleeping =       &noqueue_qdisc,
 372 };
 373
 374 static struct Qdisc noqueue_qdisc = {
 375         .enqueue        =       NULL,
 376         .dequeue        =       noop_dequeue,
 377         .flags          =       TCQ_F_BUILTIN,
 378         .ops            =       &noqueue_qdisc_ops,
 379         .list           =       LIST_HEAD_INIT(noqueue_qdisc.list),
 380         .q.lock         =       __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
 381         .dev_queue      =       &noqueue_netdev_queue,
 382 };
 383
 384
 385 static const u8 prio2band[TC_PRIO_MAX+1] =
 386         { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 };
 387
 388 /* 3-band FIFO queue: old style, but should be a bit faster than
 389    generic prio+fifo combination.
 390  */
 391
 392 #define PFIFO_FAST_BANDS 3
 393
 394 static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
 395                                              struct Qdisc *qdisc)
 396 {
 397         struct sk_buff_head *list = qdisc_priv(qdisc);
 398         return list + prio2band[skb->priority & TC_PRIO_MAX];
 399 }
 400
 401 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
 402 {
 403         struct sk_buff_head *list = prio2list(skb, qdisc);
 404
 405         if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
 406                 qdisc->q.qlen++;
 407                 return __qdisc_enqueue_tail(skb, qdisc, list);
 408         }
 409
 410         return qdisc_drop(skb, qdisc);
 411 }
 412
 413 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
 414 {
 415         int prio;
 416         struct sk_buff_head *list = qdisc_priv(qdisc);
 417
 418         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
 419                 if (!skb_queue_empty(list + prio)) {
 420                         qdisc->q.qlen--;
 421                         return __qdisc_dequeue_head(qdisc, list + prio);
 422                 }
 423         }
 424
 425         return NULL;
 426 }
 427
 428 static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc)
 429 {
 430         int prio;
 431         struct sk_buff_head *list = qdisc_priv(qdisc);
 432
 433         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
 434                 if (!skb_queue_empty(list + prio))
 435                         return skb_peek(list + prio);
 436         }
 437
 438         return NULL;
 439 }
 440
 441 static void pfifo_fast_reset(struct Qdisc* qdisc)
 442 {
 443         int prio;
 444         struct sk_buff_head *list = qdisc_priv(qdisc);
 445
 446         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
 447                 __qdisc_reset_queue(qdisc, list + prio);
 448
 449         qdisc->qstats.backlog = 0;
 450         qdisc->q.qlen = 0;
 451 }
 452
 453 static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
 454 {
 455         struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
 456
 457         memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1);
 458         NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 459         return skb->len;
 460
 461 nla_put_failure:
 462         return -1;
 463 }
 464
 465 static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
 466 {
 467         int prio;
 468         struct sk_buff_head *list = qdisc_priv(qdisc);
 469
 470         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
 471                 skb_queue_head_init(list + prio);
 472
 473         return 0;
 474 }
 475
 476 static struct Qdisc_ops pfifo_fast_ops __read_mostly = {
 477         .id             =       "pfifo_fast",
 478         .priv_size      =       PFIFO_FAST_BANDS * sizeof(struct sk_buff_head),
 479         .enqueue        =       pfifo_fast_enqueue,
 480         .dequeue        =       pfifo_fast_dequeue,
 481         .peek           =       pfifo_fast_peek,
 482         .init           =       pfifo_fast_init,
 483         .reset          =       pfifo_fast_reset,
 484         .dump           =       pfifo_fast_dump,
 485         .owner          =       THIS_MODULE,
 486 };
 487
 488 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 489                           struct Qdisc_ops *ops)
 490 {
 491         void *p;
 492         struct Qdisc *sch;
 493         unsigned int size;
 494         int err = -ENOBUFS;
 495
 496         /* ensure that the Qdisc and the private data are 32-byte aligned */
 497         size = QDISC_ALIGN(sizeof(*sch));
 498         size += ops->priv_size + (QDISC_ALIGNTO - 1);
 499
 500         p = kzalloc(size, GFP_KERNEL);
 501         if (!p)
 502                 goto errout;
 503         sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
 504         sch->padded = (char *) sch - (char *) p;
 505
 506         INIT_LIST_HEAD(&sch->list);
 507         skb_queue_head_init(&sch->q);
 508         sch->ops = ops;
 509         sch->enqueue = ops->enqueue;
 510         sch->dequeue = ops->dequeue;
 511         sch->dev_queue = dev_queue;
 512         dev_hold(qdisc_dev(sch));
 513         atomic_set(&sch->refcnt, 1);
 514
 515         return sch;
 516 errout:
 517         return ERR_PTR(err);
 518 }
 519
 520 struct Qdisc * qdisc_create_dflt(struct net_device *dev,
 521                                  struct netdev_queue *dev_queue,
 522                                  struct Qdisc_ops *ops,
 523                                  unsigned int parentid)
 524 {
 525         struct Qdisc *sch;
 526
 527         sch = qdisc_alloc(dev_queue, ops);
 528         if (IS_ERR(sch))
 529                 goto errout;
 530         sch->parent = parentid;
 531
 532         if (!ops->init || ops->init(sch, NULL) == 0)
 533                 return sch;
 534
 535         qdisc_destroy(sch);
 536 errout:
 537         return NULL;
 538 }
 539 EXPORT_SYMBOL(qdisc_create_dflt);
 540
 541 /* Under qdisc_lock(qdisc) and BH! */
 542
 543 void qdisc_reset(struct Qdisc *qdisc)
 544 {
 545         const struct Qdisc_ops *ops = qdisc->ops;
 546
 547         if (ops->reset)
 548                 ops->reset(qdisc);
 549
 550         kfree_skb(qdisc->gso_skb);
 551         qdisc->gso_skb = NULL;
 552 }
 553 EXPORT_SYMBOL(qdisc_reset);
 554
 555 void qdisc_destroy(struct Qdisc *qdisc)
 556 {
 557         const struct Qdisc_ops  *ops = qdisc->ops;
 558
 559         if (qdisc->flags & TCQ_F_BUILTIN ||
 560             !atomic_dec_and_test(&qdisc->refcnt))
 561                 return;
 562
 563 #ifdef CONFIG_NET_SCHED
 564         qdisc_list_del(qdisc);
 565
 566         qdisc_put_stab(qdisc->stab);
 567 #endif
 568         gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
 569         if (ops->reset)
 570                 ops->reset(qdisc);
 571         if (ops->destroy)
 572                 ops->destroy(qdisc);
 573
 574         module_put(ops->owner);
 575         dev_put(qdisc_dev(qdisc));
 576
 577         kfree_skb(qdisc->gso_skb);
 578         kfree((char *) qdisc - qdisc->padded);
 579 }
 580 EXPORT_SYMBOL(qdisc_destroy);
 581
 582 static bool dev_all_qdisc_sleeping_noop(struct net_device *dev)
 583 {
 584         unsigned int i;
 585
 586         for (i = 0; i < dev->num_tx_queues; i++) {
 587                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 588
 589                 if (txq->qdisc_sleeping != &noop_qdisc)
 590                         return false;
 591         }
 592         return true;
 593 }
 594
 595 static void attach_one_default_qdisc(struct net_device *dev,
 596                                      struct netdev_queue *dev_queue,
 597                                      void *_unused)
 598 {
 599         struct Qdisc *qdisc;
 600
 601         if (dev->tx_queue_len) {
 602                 qdisc = qdisc_create_dflt(dev, dev_queue,
 603                                           &pfifo_fast_ops, TC_H_ROOT);
 604                 if (!qdisc) {
 605                         printk(KERN_INFO "%s: activation failed\n", dev->name);
 606                         return;
 607                 }
 608         } else {
 609                 qdisc =  &noqueue_qdisc;
 610         }
 611         dev_queue->qdisc_sleeping = qdisc;
 612 }
 613
 614 static void transition_one_qdisc(struct net_device *dev,
 615                                  struct netdev_queue *dev_queue,
 616                                  void *_need_watchdog)
 617 {
 618         struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
 619         int *need_watchdog_p = _need_watchdog;
 620
 621         if (!(new_qdisc->flags & TCQ_F_BUILTIN))
 622                 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
 623
 624         rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
 625         if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
 626                 dev_queue->trans_start = 0;
 627                 *need_watchdog_p = 1;
 628         }
 629 }
 630
 631 void dev_activate(struct net_device *dev)
 632 {
 633         int need_watchdog;
 634
 635         /* No queueing discipline is attached to device;
 636            create default one i.e. pfifo_fast for devices,
 637            which need queueing and noqueue_qdisc for
 638            virtual interfaces
 639          */
 640
 641         if (dev_all_qdisc_sleeping_noop(dev))
 642                 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
 643
 644         if (!netif_carrier_ok(dev))
 645                 /* Delay activation until next carrier-on event */
 646                 return;
 647
 648         need_watchdog = 0;
 649         netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
 650         transition_one_qdisc(dev, &dev->rx_queue, NULL);
 651
 652         if (need_watchdog) {
 653                 dev->trans_start = jiffies;
 654                 dev_watchdog_up(dev);
 655         }
 656 }
 657
 658 static void dev_deactivate_queue(struct net_device *dev,
 659                                  struct netdev_queue *dev_queue,
 660                                  void *_qdisc_default)
 661 {
 662         struct Qdisc *qdisc_default = _qdisc_default;
 663         struct Qdisc *qdisc;
 664
 665         qdisc = dev_queue->qdisc;
 666         if (qdisc) {
 667                 spin_lock_bh(qdisc_lock(qdisc));
 668
 669                 if (!(qdisc->flags & TCQ_F_BUILTIN))
 670                         set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
 671
 672                 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
 673                 qdisc_reset(qdisc);
 674
 675                 spin_unlock_bh(qdisc_lock(qdisc));
 676         }
 677 }
 678
 679 static bool some_qdisc_is_busy(struct net_device *dev)
 680 {
 681         unsigned int i;
 682
 683         for (i = 0; i < dev->num_tx_queues; i++) {
 684                 struct netdev_queue *dev_queue;
 685                 spinlock_t *root_lock;
 686                 struct Qdisc *q;
 687                 int val;
 688
 689                 dev_queue = netdev_get_tx_queue(dev, i);
 690                 q = dev_queue->qdisc_sleeping;
 691                 root_lock = qdisc_lock(q);
 692
 693                 spin_lock_bh(root_lock);
 694
 695                 val = (test_bit(__QDISC_STATE_RUNNING, &q->state) ||
 696                        test_bit(__QDISC_STATE_SCHED, &q->state));
 697
 698                 spin_unlock_bh(root_lock);
 699
 700                 if (val)
 701                         return true;
 702         }
 703         return false;
 704 }
 705
 706 void dev_deactivate(struct net_device *dev)
 707 {
 708         netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc);
 709         dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc);
 710
 711         dev_watchdog_down(dev);
 712
 713         /* Wait for outstanding qdisc-less dev_queue_xmit calls. */
 714         synchronize_rcu();
 715
 716         /* Wait for outstanding qdisc_run calls. */
 717         while (some_qdisc_is_busy(dev))
 718                 yield();
 719 }
 720
 721 static void dev_init_scheduler_queue(struct net_device *dev,
 722                                      struct netdev_queue *dev_queue,
 723                                      void *_qdisc)
 724 {
 725         struct Qdisc *qdisc = _qdisc;
 726
 727         dev_queue->qdisc = qdisc;
 728         dev_queue->qdisc_sleeping = qdisc;
 729 }
 730
 731 void dev_init_scheduler(struct net_device *dev)
 732 {
 733         netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
 734         dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
 735
 736         setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
 737 }
 738
 739 static void shutdown_scheduler_queue(struct net_device *dev,
 740                                      struct netdev_queue *dev_queue,
 741                                      void *_qdisc_default)
 742 {
 743         struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
 744         struct Qdisc *qdisc_default = _qdisc_default;
 745
 746         if (qdisc) {
 747                 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
 748                 dev_queue->qdisc_sleeping = qdisc_default;
 749
 750                 qdisc_destroy(qdisc);
 751         }
 752 }
 753
 754 void dev_shutdown(struct net_device *dev)
 755 {
 756         netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
 757         shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
 758         WARN_ON(timer_pending(&dev->watchdog_timer));
 759 }