sch_generic.c 15.9 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
/*
 * net/sched/sch_generic.c	Generic packet scheduler routines.
 *
 *		This program is free software; you can redistribute it and/or
 *		modify it under the terms of the GNU General Public License
 *		as published by the Free Software Foundation; either version
 *		2 of the License, or (at your option) any later version.
 *
 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 *              Jamal Hadi Salim, <hadi@cyberus.ca> 990601
 *              - Ingress support
 */

#include <linux/bitops.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/netdevice.h>
#include <linux/skbuff.h>
#include <linux/rtnetlink.h>
#include <linux/init.h>
#include <linux/rcupdate.h>
#include <linux/list.h>
#include <net/pkt_sched.h>

/* Main transmission queue. */

31
/* Modifications to data participating in scheduling must be protected with
32
 * queue->lock spinlock.
33 34 35
 *
 * The idea is the following:
 * - enqueue, dequeue are serialized via top level device
36
 *   spinlock queue->lock.
37
 * - ingress filtering is serialized via top level device
38
 *   spinlock dev->rx_queue.lock.
39
 * - updates to tree and tree walking are only done under the rtnl mutex.
L
Linus Torvalds 已提交
40 41 42
 */

void qdisc_lock_tree(struct net_device *dev)
43
	__acquires(dev->tx_queue.lock)
44
	__acquires(dev->rx_queue.lock)
L
Linus Torvalds 已提交
45
{
46
	spin_lock_bh(&dev->tx_queue.lock);
47
	spin_lock(&dev->rx_queue.lock);
L
Linus Torvalds 已提交
48
}
49
EXPORT_SYMBOL(qdisc_lock_tree);
L
Linus Torvalds 已提交
50 51

void qdisc_unlock_tree(struct net_device *dev)
52
	__releases(dev->rx_queue.lock)
53
	__releases(dev->tx_queue.lock)
L
Linus Torvalds 已提交
54
{
55
	spin_unlock(&dev->rx_queue.lock);
56
	spin_unlock_bh(&dev->tx_queue.lock);
L
Linus Torvalds 已提交
57
}
58
EXPORT_SYMBOL(qdisc_unlock_tree);
L
Linus Torvalds 已提交
59

60 61 62 63 64
static inline int qdisc_qlen(struct Qdisc *q)
{
	return q->q.qlen;
}

65
static inline int dev_requeue_skb(struct sk_buff *skb,
66
				  struct netdev_queue *dev_queue,
67
				  struct Qdisc *q)
68 69
{
	if (unlikely(skb->next))
70
		dev_queue->gso_skb = skb;
71 72
	else
		q->ops->requeue(skb, q);
73

74
	netif_schedule_queue(dev_queue);
75 76 77
	return 0;
}

78 79
static inline struct sk_buff *dequeue_skb(struct netdev_queue *dev_queue,
					  struct Qdisc *q)
80
{
81
	struct sk_buff *skb;
82

83 84
	if ((skb = dev_queue->gso_skb))
		dev_queue->gso_skb = NULL;
85 86 87 88 89 90
	else
		skb = q->dequeue(q);

	return skb;
}

91
static inline int handle_dev_cpu_collision(struct sk_buff *skb,
92
					   struct netdev_queue *dev_queue,
93
					   struct Qdisc *q)
94
{
95
	struct net_device *dev = dev_queue->dev;
96
	int ret;
97

98 99 100 101 102 103 104
	if (unlikely(dev->xmit_lock_owner == smp_processor_id())) {
		/*
		 * Same CPU holding the lock. It may be a transient
		 * configuration error, when hard_start_xmit() recurses. We
		 * detect it by checking xmit owner and drop the packet when
		 * deadloop is detected. Return OK to try the next skb.
		 */
105
		kfree_skb(skb);
106 107 108 109 110 111 112 113 114 115
		if (net_ratelimit())
			printk(KERN_WARNING "Dead loop on netdevice %s, "
			       "fix it urgently!\n", dev->name);
		ret = qdisc_qlen(q);
	} else {
		/*
		 * Another cpu is holding lock, requeue & delay xmits for
		 * some time.
		 */
		__get_cpu_var(netdev_rx_stat).cpu_collision++;
116
		ret = dev_requeue_skb(skb, dev_queue, q);
117 118
	}

119
	return ret;
120 121
}

122
/*
123
 * NOTE: Called under queue->lock with locally disabled BH.
124 125
 *
 * __LINK_STATE_QDISC_RUNNING guarantees only one CPU can process this
126
 * device at a time. queue->lock serializes queue accesses for
127
 * this device AND txq->qdisc pointer itself.
128 129 130
 *
 *  netif_tx_lock serializes accesses to device driver.
 *
131
 *  queue->lock and netif_tx_lock are mutually exclusive,
132 133 134 135 136 137 138 139 140
 *  if one is grabbed, another must be free.
 *
 * Note, that this procedure can be called by a watchdog timer
 *
 * Returns to the caller:
 *				0  - queue is empty or throttled.
 *				>0 - queue is not empty.
 *
 */
141
static inline int qdisc_restart(struct netdev_queue *txq)
L
Linus Torvalds 已提交
142
{
143
	struct Qdisc *q = txq->qdisc;
144
	int ret = NETDEV_TX_BUSY;
145 146
	struct net_device *dev;
	struct sk_buff *skb;
L
Linus Torvalds 已提交
147

148
	/* Dequeue packet */
149
	if (unlikely((skb = dequeue_skb(txq, q)) == NULL))
150
		return 0;
151

152 153

	/* And release queue */
154
	spin_unlock(&txq->lock);
155

156 157
	dev = txq->dev;

158
	HARD_TX_LOCK(dev, smp_processor_id());
159 160
	if (!netif_subqueue_stopped(dev, skb))
		ret = dev_hard_start_xmit(skb, dev);
161
	HARD_TX_UNLOCK(dev);
162

163 164
	spin_lock(&txq->lock);
	q = txq->qdisc;
165

166 167 168 169 170 171 172 173
	switch (ret) {
	case NETDEV_TX_OK:
		/* Driver sent out skb successfully */
		ret = qdisc_qlen(q);
		break;

	case NETDEV_TX_LOCKED:
		/* Driver try lock failed */
174
		ret = handle_dev_cpu_collision(skb, txq, q);
175 176 177 178 179 180 181 182
		break;

	default:
		/* Driver returned NETDEV_TX_BUSY - requeue skb */
		if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
			printk(KERN_WARNING "BUG %s code %d qlen %d\n",
			       dev->name, ret, q->q.qlen);

183
		ret = dev_requeue_skb(skb, txq, q);
184 185
		break;
	}
186

187
	return ret;
L
Linus Torvalds 已提交
188 189
}

190
void __qdisc_run(struct netdev_queue *txq)
H
Herbert Xu 已提交
191
{
192
	struct net_device *dev = txq->dev;
193 194
	unsigned long start_time = jiffies;

195
	while (qdisc_restart(txq)) {
196 197 198 199 200 201 202 203 204
		if (netif_queue_stopped(dev))
			break;

		/*
		 * Postpone processing if
		 * 1. another process needs the CPU;
		 * 2. we've been doing it for too long.
		 */
		if (need_resched() || jiffies != start_time) {
205
			netif_schedule_queue(txq);
206
			break;
207 208
		}
	}
H
Herbert Xu 已提交
209 210 211 212

	clear_bit(__LINK_STATE_QDISC_RUNNING, &dev->state);
}

L
Linus Torvalds 已提交
213 214 215
static void dev_watchdog(unsigned long arg)
{
	struct net_device *dev = (struct net_device *)arg;
216
	struct netdev_queue *txq = &dev->tx_queue;
L
Linus Torvalds 已提交
217

H
Herbert Xu 已提交
218
	netif_tx_lock(dev);
219
	if (txq->qdisc != &noop_qdisc) {
L
Linus Torvalds 已提交
220 221 222 223
		if (netif_device_present(dev) &&
		    netif_running(dev) &&
		    netif_carrier_ok(dev)) {
			if (netif_queue_stopped(dev) &&
224 225 226 227
			    time_after(jiffies, dev->trans_start + dev->watchdog_timeo)) {

				printk(KERN_INFO "NETDEV WATCHDOG: %s: transmit timed out\n",
				       dev->name);
L
Linus Torvalds 已提交
228
				dev->tx_timeout(dev);
229
				WARN_ON_ONCE(1);
L
Linus Torvalds 已提交
230
			}
231
			if (!mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + dev->watchdog_timeo)))
L
Linus Torvalds 已提交
232 233 234
				dev_hold(dev);
		}
	}
H
Herbert Xu 已提交
235
	netif_tx_unlock(dev);
L
Linus Torvalds 已提交
236 237 238 239 240 241 242 243 244

	dev_put(dev);
}

void __netdev_watchdog_up(struct net_device *dev)
{
	if (dev->tx_timeout) {
		if (dev->watchdog_timeo <= 0)
			dev->watchdog_timeo = 5*HZ;
245 246
		if (!mod_timer(&dev->watchdog_timer,
			       round_jiffies(jiffies + dev->watchdog_timeo)))
L
Linus Torvalds 已提交
247 248 249 250 251 252 253 254 255 256 257
			dev_hold(dev);
	}
}

static void dev_watchdog_up(struct net_device *dev)
{
	__netdev_watchdog_up(dev);
}

static void dev_watchdog_down(struct net_device *dev)
{
H
Herbert Xu 已提交
258
	netif_tx_lock_bh(dev);
L
Linus Torvalds 已提交
259
	if (del_timer(&dev->watchdog_timer))
260
		dev_put(dev);
H
Herbert Xu 已提交
261
	netif_tx_unlock_bh(dev);
L
Linus Torvalds 已提交
262 263
}

264 265 266 267 268 269
/**
 *	netif_carrier_on - set carrier
 *	@dev: network device
 *
 * Device has detected that carrier.
 */
270 271
void netif_carrier_on(struct net_device *dev)
{
J
Jeff Garzik 已提交
272
	if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
273
		linkwatch_fire_event(dev);
J
Jeff Garzik 已提交
274 275 276
		if (netif_running(dev))
			__netdev_watchdog_up(dev);
	}
277
}
278
EXPORT_SYMBOL(netif_carrier_on);
279

280 281 282 283 284 285
/**
 *	netif_carrier_off - clear carrier
 *	@dev: network device
 *
 * Device has detected loss of carrier.
 */
286 287 288 289 290
void netif_carrier_off(struct net_device *dev)
{
	if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state))
		linkwatch_fire_event(dev);
}
291
EXPORT_SYMBOL(netif_carrier_off);
292

L
Linus Torvalds 已提交
293 294 295 296 297
/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
   under all circumstances. It is difficult to invent anything faster or
   cheaper.
 */

298
static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
L
Linus Torvalds 已提交
299 300 301 302 303
{
	kfree_skb(skb);
	return NET_XMIT_CN;
}

304
static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
L
Linus Torvalds 已提交
305 306 307 308
{
	return NULL;
}

309
static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
L
Linus Torvalds 已提交
310 311
{
	if (net_ratelimit())
312 313
		printk(KERN_DEBUG "%s deferred output. It is buggy.\n",
		       skb->dev->name);
L
Linus Torvalds 已提交
314 315 316 317
	kfree_skb(skb);
	return NET_XMIT_CN;
}

318
struct Qdisc_ops noop_qdisc_ops __read_mostly = {
L
Linus Torvalds 已提交
319 320 321 322 323 324 325 326 327 328 329 330
	.id		=	"noop",
	.priv_size	=	0,
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
	.requeue	=	noop_requeue,
	.owner		=	THIS_MODULE,
};

struct Qdisc noop_qdisc = {
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
	.flags		=	TCQ_F_BUILTIN,
331
	.ops		=	&noop_qdisc_ops,
L
Linus Torvalds 已提交
332 333
	.list		=	LIST_HEAD_INIT(noop_qdisc.list),
};
334
EXPORT_SYMBOL(noop_qdisc);
L
Linus Torvalds 已提交
335

336
static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
L
Linus Torvalds 已提交
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
	.id		=	"noqueue",
	.priv_size	=	0,
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
	.requeue	=	noop_requeue,
	.owner		=	THIS_MODULE,
};

static struct Qdisc noqueue_qdisc = {
	.enqueue	=	NULL,
	.dequeue	=	noop_dequeue,
	.flags		=	TCQ_F_BUILTIN,
	.ops		=	&noqueue_qdisc_ops,
	.list		=	LIST_HEAD_INIT(noqueue_qdisc.list),
};


static const u8 prio2band[TC_PRIO_MAX+1] =
	{ 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 };

/* 3-band FIFO queue: old style, but should be a bit faster than
   generic prio+fifo combination.
 */

361 362
#define PFIFO_FAST_BANDS 3

363 364
static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
					     struct Qdisc *qdisc)
L
Linus Torvalds 已提交
365 366
{
	struct sk_buff_head *list = qdisc_priv(qdisc);
367 368
	return list + prio2band[skb->priority & TC_PRIO_MAX];
}
L
Linus Torvalds 已提交
369

370
static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
371 372
{
	struct sk_buff_head *list = prio2list(skb, qdisc);
L
Linus Torvalds 已提交
373

374
	if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
L
Linus Torvalds 已提交
375
		qdisc->q.qlen++;
376
		return __qdisc_enqueue_tail(skb, qdisc, list);
L
Linus Torvalds 已提交
377
	}
378 379

	return qdisc_drop(skb, qdisc);
L
Linus Torvalds 已提交
380 381
}

382
static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
L
Linus Torvalds 已提交
383 384 385 386
{
	int prio;
	struct sk_buff_head *list = qdisc_priv(qdisc);

387 388
	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
		if (!skb_queue_empty(list + prio)) {
L
Linus Torvalds 已提交
389
			qdisc->q.qlen--;
390
			return __qdisc_dequeue_head(qdisc, list + prio);
L
Linus Torvalds 已提交
391 392
		}
	}
393

L
Linus Torvalds 已提交
394 395 396
	return NULL;
}

397
static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
L
Linus Torvalds 已提交
398 399
{
	qdisc->q.qlen++;
400
	return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc));
L
Linus Torvalds 已提交
401 402
}

403
static void pfifo_fast_reset(struct Qdisc* qdisc)
L
Linus Torvalds 已提交
404 405 406 407
{
	int prio;
	struct sk_buff_head *list = qdisc_priv(qdisc);

408
	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
409 410 411
		__qdisc_reset_queue(qdisc, list + prio);

	qdisc->qstats.backlog = 0;
L
Linus Torvalds 已提交
412 413 414 415 416
	qdisc->q.qlen = 0;
}

static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
{
417
	struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
L
Linus Torvalds 已提交
418 419

	memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1);
420
	NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
L
Linus Torvalds 已提交
421 422
	return skb->len;

423
nla_put_failure:
L
Linus Torvalds 已提交
424 425 426
	return -1;
}

427
static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
L
Linus Torvalds 已提交
428
{
429
	int prio;
L
Linus Torvalds 已提交
430 431
	struct sk_buff_head *list = qdisc_priv(qdisc);

432 433
	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
		skb_queue_head_init(list + prio);
L
Linus Torvalds 已提交
434 435 436 437

	return 0;
}

438
static struct Qdisc_ops pfifo_fast_ops __read_mostly = {
L
Linus Torvalds 已提交
439
	.id		=	"pfifo_fast",
440
	.priv_size	=	PFIFO_FAST_BANDS * sizeof(struct sk_buff_head),
L
Linus Torvalds 已提交
441 442 443 444 445 446 447 448 449
	.enqueue	=	pfifo_fast_enqueue,
	.dequeue	=	pfifo_fast_dequeue,
	.requeue	=	pfifo_fast_requeue,
	.init		=	pfifo_fast_init,
	.reset		=	pfifo_fast_reset,
	.dump		=	pfifo_fast_dump,
	.owner		=	THIS_MODULE,
};

450
struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
451
			  struct Qdisc_ops *ops)
L
Linus Torvalds 已提交
452 453 454
{
	void *p;
	struct Qdisc *sch;
455 456
	unsigned int size;
	int err = -ENOBUFS;
L
Linus Torvalds 已提交
457 458

	/* ensure that the Qdisc and the private data are 32-byte aligned */
459 460
	size = QDISC_ALIGN(sizeof(*sch));
	size += ops->priv_size + (QDISC_ALIGNTO - 1);
L
Linus Torvalds 已提交
461

462
	p = kzalloc(size, GFP_KERNEL);
L
Linus Torvalds 已提交
463
	if (!p)
464 465 466
		goto errout;
	sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
	sch->padded = (char *) sch - (char *) p;
L
Linus Torvalds 已提交
467 468 469 470 471 472

	INIT_LIST_HEAD(&sch->list);
	skb_queue_head_init(&sch->q);
	sch->ops = ops;
	sch->enqueue = ops->enqueue;
	sch->dequeue = ops->dequeue;
473
	sch->dev_queue = dev_queue;
474
	dev_hold(qdisc_dev(sch));
L
Linus Torvalds 已提交
475
	atomic_set(&sch->refcnt, 1);
476 477 478

	return sch;
errout:
479
	return ERR_PTR(err);
480 481
}

482 483 484
struct Qdisc * qdisc_create_dflt(struct net_device *dev,
				 struct netdev_queue *dev_queue,
				 struct Qdisc_ops *ops,
485
				 unsigned int parentid)
486 487
{
	struct Qdisc *sch;
488

489
	sch = qdisc_alloc(dev_queue, ops);
490 491
	if (IS_ERR(sch))
		goto errout;
492
	sch->parent = parentid;
493

L
Linus Torvalds 已提交
494 495 496
	if (!ops->init || ops->init(sch, NULL) == 0)
		return sch;

497
	qdisc_destroy(sch);
498
errout:
L
Linus Torvalds 已提交
499 500
	return NULL;
}
501
EXPORT_SYMBOL(qdisc_create_dflt);
L
Linus Torvalds 已提交
502

503
/* Under queue->lock and BH! */
L
Linus Torvalds 已提交
504 505 506

void qdisc_reset(struct Qdisc *qdisc)
{
507
	const struct Qdisc_ops *ops = qdisc->ops;
L
Linus Torvalds 已提交
508 509 510 511

	if (ops->reset)
		ops->reset(qdisc);
}
512
EXPORT_SYMBOL(qdisc_reset);
L
Linus Torvalds 已提交
513

514
/* this is the rcu callback function to clean up a qdisc when there
L
Linus Torvalds 已提交
515 516 517 518 519 520 521 522
 * are no further references to it */

static void __qdisc_destroy(struct rcu_head *head)
{
	struct Qdisc *qdisc = container_of(head, struct Qdisc, q_rcu);
	kfree((char *) qdisc - qdisc->padded);
}

523
/* Under queue->lock and BH! */
L
Linus Torvalds 已提交
524 525 526

void qdisc_destroy(struct Qdisc *qdisc)
{
527
	const struct Qdisc_ops  *ops = qdisc->ops;
L
Linus Torvalds 已提交
528 529

	if (qdisc->flags & TCQ_F_BUILTIN ||
530
	    !atomic_dec_and_test(&qdisc->refcnt))
L
Linus Torvalds 已提交
531 532
		return;

533 534 535 536 537 538
	list_del(&qdisc->list);
	gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
	if (ops->reset)
		ops->reset(qdisc);
	if (ops->destroy)
		ops->destroy(qdisc);
L
Linus Torvalds 已提交
539

540
	module_put(ops->owner);
541
	dev_put(qdisc_dev(qdisc));
L
Linus Torvalds 已提交
542 543
	call_rcu(&qdisc->q_rcu, __qdisc_destroy);
}
544
EXPORT_SYMBOL(qdisc_destroy);
L
Linus Torvalds 已提交
545 546 547

void dev_activate(struct net_device *dev)
{
548 549
	struct netdev_queue *txq = &dev->tx_queue;

L
Linus Torvalds 已提交
550 551 552 553 554 555
	/* No queueing discipline is attached to device;
	   create default one i.e. pfifo_fast for devices,
	   which need queueing and noqueue_qdisc for
	   virtual interfaces
	 */

556
	if (txq->qdisc_sleeping == &noop_qdisc) {
L
Linus Torvalds 已提交
557 558
		struct Qdisc *qdisc;
		if (dev->tx_queue_len) {
559
			qdisc = qdisc_create_dflt(dev, txq,
560
						  &pfifo_fast_ops,
561
						  TC_H_ROOT);
L
Linus Torvalds 已提交
562 563 564 565
			if (qdisc == NULL) {
				printk(KERN_INFO "%s: activation failed\n", dev->name);
				return;
			}
566
			list_add_tail(&qdisc->list, &txq->qdisc_list);
L
Linus Torvalds 已提交
567 568 569
		} else {
			qdisc =  &noqueue_qdisc;
		}
570
		txq->qdisc_sleeping = qdisc;
L
Linus Torvalds 已提交
571 572
	}

573 574 575 576
	if (!netif_carrier_ok(dev))
		/* Delay activation until next carrier-on event */
		return;

577 578 579
	spin_lock_bh(&txq->lock);
	rcu_assign_pointer(txq->qdisc, txq->qdisc_sleeping);
	if (txq->qdisc != &noqueue_qdisc) {
L
Linus Torvalds 已提交
580 581 582
		dev->trans_start = jiffies;
		dev_watchdog_up(dev);
	}
583 584 585
	spin_unlock_bh(&txq->lock);
}

586
static void dev_deactivate_queue(struct netdev_queue *dev_queue,
587 588
				 struct Qdisc *qdisc_default)
{
589 590 591 592
	struct Qdisc *qdisc;
	struct sk_buff *skb;

	spin_lock_bh(&dev_queue->lock);
593

594
	qdisc = dev_queue->qdisc;
595 596 597 598
	if (qdisc) {
		dev_queue->qdisc = qdisc_default;
		qdisc_reset(qdisc);
	}
599 600 601 602 603 604
	skb = dev_queue->gso_skb;
	dev_queue->gso_skb = NULL;

	spin_unlock_bh(&dev_queue->lock);

	kfree_skb(skb);
L
Linus Torvalds 已提交
605 606 607 608
}

void dev_deactivate(struct net_device *dev)
{
609
	int running;
L
Linus Torvalds 已提交
610

611
	dev_deactivate_queue(&dev->tx_queue, &noop_qdisc);
612

L
Linus Torvalds 已提交
613 614
	dev_watchdog_down(dev);

615
	/* Wait for outstanding qdisc-less dev_queue_xmit calls. */
616
	synchronize_rcu();
L
Linus Torvalds 已提交
617

618
	/* Wait for outstanding qdisc_run calls. */
619 620 621 622 623 624 625 626
	do {
		while (test_bit(__LINK_STATE_QDISC_RUNNING, &dev->state))
			yield();

		/*
		 * Double-check inside queue lock to ensure that all effects
		 * of the queue run are visible when we return.
		 */
627
		spin_lock_bh(&dev->tx_queue.lock);
628
		running = test_bit(__LINK_STATE_QDISC_RUNNING, &dev->state);
629
		spin_unlock_bh(&dev->tx_queue.lock);
630 631 632 633 634 635 636 637 638 639

		/*
		 * The running flag should never be set at this point because
		 * we've already set dev->qdisc to noop_qdisc *inside* the same
		 * pair of spin locks.  That is, if any qdisc_run starts after
		 * our initial test it should see the noop_qdisc and then
		 * clear the RUNNING bit before dropping the queue lock.  So
		 * if it is set here then we've found a bug.
		 */
	} while (WARN_ON_ONCE(running));
L
Linus Torvalds 已提交
640 641
}

642 643 644 645 646 647 648 649 650
static void dev_init_scheduler_queue(struct net_device *dev,
				     struct netdev_queue *dev_queue,
				     struct Qdisc *qdisc)
{
	dev_queue->qdisc = qdisc;
	dev_queue->qdisc_sleeping = qdisc;
	INIT_LIST_HEAD(&dev_queue->qdisc_list);
}

L
Linus Torvalds 已提交
651 652 653
void dev_init_scheduler(struct net_device *dev)
{
	qdisc_lock_tree(dev);
654 655
	dev_init_scheduler_queue(dev, &dev->tx_queue, &noop_qdisc);
	dev_init_scheduler_queue(dev, &dev->rx_queue, NULL);
L
Linus Torvalds 已提交
656 657
	qdisc_unlock_tree(dev);

658
	setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
L
Linus Torvalds 已提交
659 660
}

661 662 663
static void dev_shutdown_scheduler_queue(struct net_device *dev,
					 struct netdev_queue *dev_queue,
					 struct Qdisc *qdisc_default)
L
Linus Torvalds 已提交
664
{
665 666 667 668 669
	struct Qdisc *qdisc = dev_queue->qdisc_sleeping;

	if (qdisc) {
		dev_queue->qdisc = qdisc_default;
		dev_queue->qdisc_sleeping = qdisc_default;
L
Linus Torvalds 已提交
670 671

		qdisc_destroy(qdisc);
672
	}
673 674 675 676 677 678 679
}

void dev_shutdown(struct net_device *dev)
{
	qdisc_lock_tree(dev);
	dev_shutdown_scheduler_queue(dev, &dev->tx_queue, &noop_qdisc);
	dev_shutdown_scheduler_queue(dev, &dev->rx_queue, NULL);
L
Linus Torvalds 已提交
680 681 682
	BUG_TRAP(!timer_pending(&dev->watchdog_timer));
	qdisc_unlock_tree(dev);
}