sch_generic.c 24.1 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
/*
 * 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>
27
#include <linux/slab.h>
28
#include <linux/if_vlan.h>
29
#include <net/sch_generic.h>
L
Linus Torvalds 已提交
30
#include <net/pkt_sched.h>
E
Eric Dumazet 已提交
31
#include <net/dst.h>
L
Linus Torvalds 已提交
32

33 34 35 36
/* Qdisc to use by default */
const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
EXPORT_SYMBOL(default_qdisc_ops);

L
Linus Torvalds 已提交
37 38
/* Main transmission queue. */

39
/* Modifications to data participating in scheduling must be protected with
40
 * qdisc_lock(qdisc) spinlock.
41 42
 *
 * The idea is the following:
43 44
 * - enqueue, dequeue are serialized via qdisc root lock
 * - ingress filtering is also serialized via qdisc root lock
45
 * - updates to tree and tree walking are only done under the rtnl mutex.
L
Linus Torvalds 已提交
46 47
 */

48
static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
49
{
50
	q->gso_skb = skb;
51
	q->qstats.requeues++;
52
	q->q.qlen++;	/* it's still part of the queue */
53
	__netif_schedule(q);
54

55 56 57
	return 0;
}

58 59
static void try_bulk_dequeue_skb(struct Qdisc *q,
				 struct sk_buff *skb,
60 61
				 const struct netdev_queue *txq,
				 int *packets)
62
{
63
	int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
64 65

	while (bytelimit > 0) {
66
		struct sk_buff *nskb = q->dequeue(q);
67

68
		if (!nskb)
69 70
			break;

71 72 73
		bytelimit -= nskb->len; /* covers GSO len */
		skb->next = nskb;
		skb = nskb;
74
		(*packets)++; /* GSO counts as one pkt */
75
	}
76
	skb->next = NULL;
77 78 79 80 81
}

/* Note that dequeue_skb can possibly return a SKB list (via skb->next).
 * A requeued skb (via q->gso_skb) can also be a SKB list.
 */
82 83
static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
				   int *packets)
84
{
85
	struct sk_buff *skb = q->gso_skb;
86
	const struct netdev_queue *txq = q->dev_queue;
87

88
	*packets = 1;
89
	*validate = true;
90 91
	if (unlikely(skb)) {
		/* check the reason of requeuing without tx lock first */
92
		txq = skb_get_tx_queue(txq->dev, skb);
93
		if (!netif_xmit_frozen_or_stopped(txq)) {
94
			q->gso_skb = NULL;
95 96
			q->q.qlen--;
		} else
97
			skb = NULL;
98 99
		/* skb in gso_skb were already validated */
		*validate = false;
100
	} else {
101 102
		if (!(q->flags & TCQ_F_ONETXQUEUE) ||
		    !netif_xmit_frozen_or_stopped(txq)) {
103
			skb = q->dequeue(q);
104
			if (skb && qdisc_may_bulk(q))
105
				try_bulk_dequeue_skb(q, skb, txq, packets);
106
		}
107
	}
108 109 110
	return skb;
}

111
static inline int handle_dev_cpu_collision(struct sk_buff *skb,
112
					   struct netdev_queue *dev_queue,
113
					   struct Qdisc *q)
114
{
115
	int ret;
116

117
	if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
118 119 120 121 122 123
		/*
		 * 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.
		 */
124
		kfree_skb_list(skb);
125 126
		net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n",
				     dev_queue->dev->name);
127 128 129 130 131 132
		ret = qdisc_qlen(q);
	} else {
		/*
		 * Another cpu is holding lock, requeue & delay xmits for
		 * some time.
		 */
E
Eric Dumazet 已提交
133
		__this_cpu_inc(softnet_data.cpu_collision);
134
		ret = dev_requeue_skb(skb, q);
135 136
	}

137
	return ret;
138 139
}

140
/*
141 142 143
 * Transmit possibly several skbs, and handle the return status as
 * required. Holding the __QDISC___STATE_RUNNING bit guarantees that
 * only one CPU can execute this function.
144 145 146 147 148
 *
 * Returns to the caller:
 *				0  - queue is empty or throttled.
 *				>0 - queue is not empty.
 */
149 150
int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
		    struct net_device *dev, struct netdev_queue *txq,
151
		    spinlock_t *root_lock, bool validate)
L
Linus Torvalds 已提交
152
{
153
	int ret = NETDEV_TX_BUSY;
154 155 156

	/* And release qdisc */
	spin_unlock(root_lock);
157

158 159 160
	/* Note that we validate skb (GSO, checksum, ...) outside of locks */
	if (validate)
		skb = validate_xmit_skb_list(skb, dev);
161

162 163 164 165
	if (skb) {
		HARD_TX_LOCK(dev, txq, smp_processor_id());
		if (!netif_xmit_frozen_or_stopped(txq))
			skb = dev_hard_start_xmit(skb, dev, txq, &ret);
166

167 168
		HARD_TX_UNLOCK(dev, txq);
	}
169
	spin_lock(root_lock);
170

171 172
	if (dev_xmit_complete(ret)) {
		/* Driver sent out skb successfully or skb was consumed */
173
		ret = qdisc_qlen(q);
174
	} else if (ret == NETDEV_TX_LOCKED) {
175
		/* Driver try lock failed */
176
		ret = handle_dev_cpu_collision(skb, txq, q);
177
	} else {
178
		/* Driver returned NETDEV_TX_BUSY - requeue skb */
179 180 181
		if (unlikely(ret != NETDEV_TX_BUSY))
			net_warn_ratelimited("BUG %s code %d qlen %d\n",
					     dev->name, ret, q->q.qlen);
182

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

186
	if (ret && netif_xmit_frozen_or_stopped(txq))
187 188
		ret = 0;

189
	return ret;
L
Linus Torvalds 已提交
190 191
}

192 193 194
/*
 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
 *
Y
Ying Xue 已提交
195
 * __QDISC___STATE_RUNNING guarantees only one CPU can process
196 197 198 199 200 201 202 203 204 205 206 207 208 209 210
 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
 * this queue.
 *
 *  netif_tx_lock serializes accesses to device driver.
 *
 *  qdisc_lock(q) and netif_tx_lock are mutually exclusive,
 *  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.
 *
 */
211
static inline int qdisc_restart(struct Qdisc *q, int *packets)
212 213 214 215 216
{
	struct netdev_queue *txq;
	struct net_device *dev;
	spinlock_t *root_lock;
	struct sk_buff *skb;
217
	bool validate;
218 219

	/* Dequeue packet */
220
	skb = dequeue_skb(q, &validate, packets);
221 222
	if (unlikely(!skb))
		return 0;
223

224 225
	root_lock = qdisc_lock(q);
	dev = qdisc_dev(q);
226
	txq = skb_get_tx_queue(dev, skb);
227

228
	return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
229 230
}

231
void __qdisc_run(struct Qdisc *q)
H
Herbert Xu 已提交
232
{
J
jamal 已提交
233
	int quota = weight_p;
234
	int packets;
235

236
	while (qdisc_restart(q, &packets)) {
237
		/*
J
jamal 已提交
238 239 240
		 * Ordered by possible occurrence: Postpone processing if
		 * 1. we've exceeded packet quota
		 * 2. another process needs the CPU;
241
		 */
242 243
		quota -= packets;
		if (quota <= 0 || need_resched()) {
244
			__netif_schedule(q);
245
			break;
246 247
		}
	}
H
Herbert Xu 已提交
248

249
	qdisc_run_end(q);
H
Herbert Xu 已提交
250 251
}

252 253
unsigned long dev_trans_start(struct net_device *dev)
{
254
	unsigned long val, res;
255 256
	unsigned int i;

257 258 259
	if (is_vlan_dev(dev))
		dev = vlan_dev_real_dev(dev);
	res = dev->trans_start;
260 261 262 263 264 265
	for (i = 0; i < dev->num_tx_queues; i++) {
		val = netdev_get_tx_queue(dev, i)->trans_start;
		if (val && time_after(val, res))
			res = val;
	}
	dev->trans_start = res;
266

267 268 269 270
	return res;
}
EXPORT_SYMBOL(dev_trans_start);

L
Linus Torvalds 已提交
271 272 273 274
static void dev_watchdog(unsigned long arg)
{
	struct net_device *dev = (struct net_device *)arg;

H
Herbert Xu 已提交
275
	netif_tx_lock(dev);
276
	if (!qdisc_tx_is_noop(dev)) {
L
Linus Torvalds 已提交
277 278 279
		if (netif_device_present(dev) &&
		    netif_running(dev) &&
		    netif_carrier_ok(dev)) {
280
			int some_queue_timedout = 0;
281
			unsigned int i;
282
			unsigned long trans_start;
283 284 285 286 287

			for (i = 0; i < dev->num_tx_queues; i++) {
				struct netdev_queue *txq;

				txq = netdev_get_tx_queue(dev, i);
288 289 290 291
				/*
				 * old device drivers set dev->trans_start
				 */
				trans_start = txq->trans_start ? : dev->trans_start;
292
				if (netif_xmit_stopped(txq) &&
293 294 295
				    time_after(jiffies, (trans_start +
							 dev->watchdog_timeo))) {
					some_queue_timedout = 1;
296
					txq->trans_timeout++;
297 298 299
					break;
				}
			}
300

301 302
			if (some_queue_timedout) {
				WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
303
				       dev->name, netdev_drivername(dev), i);
304
				dev->netdev_ops->ndo_tx_timeout(dev);
L
Linus Torvalds 已提交
305
			}
306 307 308
			if (!mod_timer(&dev->watchdog_timer,
				       round_jiffies(jiffies +
						     dev->watchdog_timeo)))
L
Linus Torvalds 已提交
309 310 311
				dev_hold(dev);
		}
	}
H
Herbert Xu 已提交
312
	netif_tx_unlock(dev);
L
Linus Torvalds 已提交
313 314 315 316 317 318

	dev_put(dev);
}

void __netdev_watchdog_up(struct net_device *dev)
{
319
	if (dev->netdev_ops->ndo_tx_timeout) {
L
Linus Torvalds 已提交
320 321
		if (dev->watchdog_timeo <= 0)
			dev->watchdog_timeo = 5*HZ;
322 323
		if (!mod_timer(&dev->watchdog_timer,
			       round_jiffies(jiffies + dev->watchdog_timeo)))
L
Linus Torvalds 已提交
324 325 326 327 328 329 330 331 332 333 334
			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
335
	netif_tx_lock_bh(dev);
L
Linus Torvalds 已提交
336
	if (del_timer(&dev->watchdog_timer))
337
		dev_put(dev);
H
Herbert Xu 已提交
338
	netif_tx_unlock_bh(dev);
L
Linus Torvalds 已提交
339 340
}

341 342 343 344 345 346
/**
 *	netif_carrier_on - set carrier
 *	@dev: network device
 *
 * Device has detected that carrier.
 */
347 348
void netif_carrier_on(struct net_device *dev)
{
J
Jeff Garzik 已提交
349
	if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
350 351
		if (dev->reg_state == NETREG_UNINITIALIZED)
			return;
352
		atomic_inc(&dev->carrier_changes);
353
		linkwatch_fire_event(dev);
J
Jeff Garzik 已提交
354 355 356
		if (netif_running(dev))
			__netdev_watchdog_up(dev);
	}
357
}
358
EXPORT_SYMBOL(netif_carrier_on);
359

360 361 362 363 364 365
/**
 *	netif_carrier_off - clear carrier
 *	@dev: network device
 *
 * Device has detected loss of carrier.
 */
366 367
void netif_carrier_off(struct net_device *dev)
{
368 369 370
	if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
		if (dev->reg_state == NETREG_UNINITIALIZED)
			return;
371
		atomic_inc(&dev->carrier_changes);
372
		linkwatch_fire_event(dev);
373
	}
374
}
375
EXPORT_SYMBOL(netif_carrier_off);
376

L
Linus Torvalds 已提交
377 378 379 380 381
/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
   under all circumstances. It is difficult to invent anything faster or
   cheaper.
 */

382
static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
L
Linus Torvalds 已提交
383 384 385 386 387
{
	kfree_skb(skb);
	return NET_XMIT_CN;
}

388
static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
L
Linus Torvalds 已提交
389 390 391 392
{
	return NULL;
}

393
struct Qdisc_ops noop_qdisc_ops __read_mostly = {
L
Linus Torvalds 已提交
394 395 396 397
	.id		=	"noop",
	.priv_size	=	0,
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
398
	.peek		=	noop_dequeue,
L
Linus Torvalds 已提交
399 400 401
	.owner		=	THIS_MODULE,
};

402 403
static struct netdev_queue noop_netdev_queue = {
	.qdisc		=	&noop_qdisc,
404
	.qdisc_sleeping	=	&noop_qdisc,
405 406
};

L
Linus Torvalds 已提交
407 408 409 410
struct Qdisc noop_qdisc = {
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
	.flags		=	TCQ_F_BUILTIN,
411
	.ops		=	&noop_qdisc_ops,
L
Linus Torvalds 已提交
412
	.list		=	LIST_HEAD_INIT(noop_qdisc.list),
413
	.q.lock		=	__SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
414
	.dev_queue	=	&noop_netdev_queue,
415
	.busylock	=	__SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
L
Linus Torvalds 已提交
416
};
417
EXPORT_SYMBOL(noop_qdisc);
L
Linus Torvalds 已提交
418

P
Phil Sutter 已提交
419 420 421 422 423 424 425 426 427 428
static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt)
{
	/* register_qdisc() assigns a default of noop_enqueue if unset,
	 * but __dev_queue_xmit() treats noqueue only as such
	 * if this is NULL - so clear it here. */
	qdisc->enqueue = NULL;
	return 0;
}

struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
L
Linus Torvalds 已提交
429 430
	.id		=	"noqueue",
	.priv_size	=	0,
P
Phil Sutter 已提交
431
	.init		=	noqueue_init,
L
Linus Torvalds 已提交
432 433
	.enqueue	=	noop_enqueue,
	.dequeue	=	noop_dequeue,
434
	.peek		=	noop_dequeue,
L
Linus Torvalds 已提交
435 436 437
	.owner		=	THIS_MODULE,
};

438 439 440
static struct Qdisc noqueue_qdisc;
static struct netdev_queue noqueue_netdev_queue = {
	.qdisc		=	&noqueue_qdisc,
441
	.qdisc_sleeping	=	&noqueue_qdisc,
442 443
};

L
Linus Torvalds 已提交
444 445 446 447 448 449
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),
450 451
	.q.lock		=	__SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
	.dev_queue	=	&noqueue_netdev_queue,
452
	.busylock	=	__SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock),
L
Linus Torvalds 已提交
453 454 455
};


E
Eric Dumazet 已提交
456 457 458
static const u8 prio2band[TC_PRIO_MAX + 1] = {
	1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
};
459 460 461 462 463 464 465

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

#define PFIFO_FAST_BANDS 3

466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
/*
 * Private data for a pfifo_fast scheduler containing:
 * 	- queues for the three band
 * 	- bitmap indicating which of the bands contain skbs
 */
struct pfifo_fast_priv {
	u32 bitmap;
	struct sk_buff_head q[PFIFO_FAST_BANDS];
};

/*
 * Convert a bitmap to the first band number where an skb is queued, where:
 * 	bitmap=0 means there are no skbs on any band.
 * 	bitmap=1 means there is an skb on band 0.
 *	bitmap=7 means there are skbs on all 3 bands, etc.
 */
static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};

static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
					     int band)
486
{
487
	return priv->q + band;
488 489
}

E
Eric Dumazet 已提交
490
static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
491
{
492 493 494 495
	if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
		int band = prio2band[skb->priority & TC_PRIO_MAX];
		struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
		struct sk_buff_head *list = band2list(priv, band);
L
Linus Torvalds 已提交
496

497
		priv->bitmap |= (1 << band);
498
		qdisc->q.qlen++;
499
		return __qdisc_enqueue_tail(skb, qdisc, list);
500
	}
501 502

	return qdisc_drop(skb, qdisc);
L
Linus Torvalds 已提交
503 504
}

E
Eric Dumazet 已提交
505
static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
L
Linus Torvalds 已提交
506
{
507 508
	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
	int band = bitmap2band[priv->bitmap];
L
Linus Torvalds 已提交
509

510 511 512 513 514 515 516 517 518
	if (likely(band >= 0)) {
		struct sk_buff_head *list = band2list(priv, band);
		struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);

		qdisc->q.qlen--;
		if (skb_queue_empty(list))
			priv->bitmap &= ~(1 << band);

		return skb;
519
	}
520

L
Linus Torvalds 已提交
521 522 523
	return NULL;
}

E
Eric Dumazet 已提交
524
static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
525
{
526 527 528 529 530
	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
	int band = bitmap2band[priv->bitmap];

	if (band >= 0) {
		struct sk_buff_head *list = band2list(priv, band);
531

532
		return skb_peek(list);
533 534 535 536 537
	}

	return NULL;
}

E
Eric Dumazet 已提交
538
static void pfifo_fast_reset(struct Qdisc *qdisc)
L
Linus Torvalds 已提交
539
{
540
	int prio;
541
	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
542 543

	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
544
		__qdisc_reset_queue(qdisc, band2list(priv, prio));
545

546
	priv->bitmap = 0;
547
	qdisc->qstats.backlog = 0;
548
	qdisc->q.qlen = 0;
L
Linus Torvalds 已提交
549 550
}

551 552 553 554
static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
{
	struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };

E
Eric Dumazet 已提交
555
	memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
556 557
	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
		goto nla_put_failure;
558 559 560 561 562 563 564 565 566
	return skb->len;

nla_put_failure:
	return -1;
}

static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
{
	int prio;
567
	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
568 569

	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
570
		__skb_queue_head_init(band2list(priv, prio));
571

572 573
	/* Can by-pass the queue discipline */
	qdisc->flags |= TCQ_F_CAN_BYPASS;
574 575 576
	return 0;
}

577
struct Qdisc_ops pfifo_fast_ops __read_mostly = {
578
	.id		=	"pfifo_fast",
579
	.priv_size	=	sizeof(struct pfifo_fast_priv),
580 581
	.enqueue	=	pfifo_fast_enqueue,
	.dequeue	=	pfifo_fast_dequeue,
582
	.peek		=	pfifo_fast_peek,
583 584 585
	.init		=	pfifo_fast_init,
	.reset		=	pfifo_fast_reset,
	.dump		=	pfifo_fast_dump,
L
Linus Torvalds 已提交
586 587 588
	.owner		=	THIS_MODULE,
};

589 590
static struct lock_class_key qdisc_tx_busylock;

591
struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
592
			  const struct Qdisc_ops *ops)
L
Linus Torvalds 已提交
593 594 595
{
	void *p;
	struct Qdisc *sch;
E
Eric Dumazet 已提交
596
	unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
597
	int err = -ENOBUFS;
598
	struct net_device *dev = dev_queue->dev;
L
Linus Torvalds 已提交
599

600 601 602
	p = kzalloc_node(size, GFP_KERNEL,
			 netdev_queue_numa_node_read(dev_queue));

L
Linus Torvalds 已提交
603
	if (!p)
604 605
		goto errout;
	sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
E
Eric Dumazet 已提交
606 607 608 609 610 611 612 613 614 615
	/* if we got non aligned memory, ask more and do alignment ourself */
	if (sch != p) {
		kfree(p);
		p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
				 netdev_queue_numa_node_read(dev_queue));
		if (!p)
			goto errout;
		sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
		sch->padded = (char *) sch - (char *) p;
	}
L
Linus Torvalds 已提交
616 617
	INIT_LIST_HEAD(&sch->list);
	skb_queue_head_init(&sch->q);
618

619
	spin_lock_init(&sch->busylock);
620 621 622
	lockdep_set_class(&sch->busylock,
			  dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);

L
Linus Torvalds 已提交
623 624 625
	sch->ops = ops;
	sch->enqueue = ops->enqueue;
	sch->dequeue = ops->dequeue;
626
	sch->dev_queue = dev_queue;
627
	dev_hold(dev);
L
Linus Torvalds 已提交
628
	atomic_set(&sch->refcnt, 1);
629 630 631

	return sch;
errout:
632
	return ERR_PTR(err);
633 634
}

635
struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
636 637
				const struct Qdisc_ops *ops,
				unsigned int parentid)
638 639
{
	struct Qdisc *sch;
640

641 642 643
	if (!try_module_get(ops->owner))
		goto errout;

644
	sch = qdisc_alloc(dev_queue, ops);
645 646
	if (IS_ERR(sch))
		goto errout;
647
	sch->parent = parentid;
648

L
Linus Torvalds 已提交
649 650 651
	if (!ops->init || ops->init(sch, NULL) == 0)
		return sch;

652
	qdisc_destroy(sch);
653
errout:
L
Linus Torvalds 已提交
654 655
	return NULL;
}
656
EXPORT_SYMBOL(qdisc_create_dflt);
L
Linus Torvalds 已提交
657

658
/* Under qdisc_lock(qdisc) and BH! */
L
Linus Torvalds 已提交
659 660 661

void qdisc_reset(struct Qdisc *qdisc)
{
662
	const struct Qdisc_ops *ops = qdisc->ops;
L
Linus Torvalds 已提交
663 664 665

	if (ops->reset)
		ops->reset(qdisc);
666

667
	if (qdisc->gso_skb) {
668
		kfree_skb_list(qdisc->gso_skb);
669 670 671
		qdisc->gso_skb = NULL;
		qdisc->q.qlen = 0;
	}
L
Linus Torvalds 已提交
672
}
673
EXPORT_SYMBOL(qdisc_reset);
L
Linus Torvalds 已提交
674

E
Eric Dumazet 已提交
675 676 677 678
static void qdisc_rcu_free(struct rcu_head *head)
{
	struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);

679 680 681
	if (qdisc_is_percpu_stats(qdisc))
		free_percpu(qdisc->cpu_bstats);

E
Eric Dumazet 已提交
682 683 684
	kfree((char *) qdisc - qdisc->padded);
}

685
void qdisc_destroy(struct Qdisc *qdisc)
L
Linus Torvalds 已提交
686
{
687 688
	const struct Qdisc_ops  *ops = qdisc->ops;

689 690 691 692
	if (qdisc->flags & TCQ_F_BUILTIN ||
	    !atomic_dec_and_test(&qdisc->refcnt))
		return;

693
#ifdef CONFIG_NET_SCHED
694 695
	qdisc_list_del(qdisc);

E
Eric Dumazet 已提交
696
	qdisc_put_stab(rtnl_dereference(qdisc->stab));
697
#endif
698 699 700 701 702 703 704 705 706
	gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
	if (ops->reset)
		ops->reset(qdisc);
	if (ops->destroy)
		ops->destroy(qdisc);

	module_put(ops->owner);
	dev_put(qdisc_dev(qdisc));

707
	kfree_skb_list(qdisc->gso_skb);
E
Eric Dumazet 已提交
708 709 710 711 712
	/*
	 * gen_estimator est_timer() might access qdisc->q.lock,
	 * wait a RCU grace period before freeing qdisc.
	 */
	call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
L
Linus Torvalds 已提交
713
}
714
EXPORT_SYMBOL(qdisc_destroy);
L
Linus Torvalds 已提交
715

716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739
/* Attach toplevel qdisc to device queue. */
struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
			      struct Qdisc *qdisc)
{
	struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
	spinlock_t *root_lock;

	root_lock = qdisc_lock(oqdisc);
	spin_lock_bh(root_lock);

	/* Prune old scheduler */
	if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
		qdisc_reset(oqdisc);

	/* ... and graft new one */
	if (qdisc == NULL)
		qdisc = &noop_qdisc;
	dev_queue->qdisc_sleeping = qdisc;
	rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);

	spin_unlock_bh(root_lock);

	return oqdisc;
}
740
EXPORT_SYMBOL(dev_graft_qdisc);
741

742 743 744 745
static void attach_one_default_qdisc(struct net_device *dev,
				     struct netdev_queue *dev_queue,
				     void *_unused)
{
E
Eric Dumazet 已提交
746
	struct Qdisc *qdisc = &noqueue_qdisc;
747

748
	if (!(dev->priv_flags & IFF_NO_QUEUE)) {
749
		qdisc = qdisc_create_dflt(dev_queue,
750
					  default_qdisc_ops, TC_H_ROOT);
751
		if (!qdisc) {
E
Eric Dumazet 已提交
752
			netdev_info(dev, "activation failed\n");
753 754
			return;
		}
755 756
		if (!netif_is_multiqueue(dev))
			qdisc->flags |= TCQ_F_ONETXQUEUE;
757 758 759 760
	}
	dev_queue->qdisc_sleeping = qdisc;
}

761 762 763 764 765 766 767
static void attach_default_qdiscs(struct net_device *dev)
{
	struct netdev_queue *txq;
	struct Qdisc *qdisc;

	txq = netdev_get_tx_queue(dev, 0);

768 769
	if (!netif_is_multiqueue(dev) ||
	    dev->priv_flags & IFF_NO_QUEUE) {
770 771 772 773
		netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
		dev->qdisc = txq->qdisc_sleeping;
		atomic_inc(&dev->qdisc->refcnt);
	} else {
774
		qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
775 776
		if (qdisc) {
			dev->qdisc = qdisc;
777
			qdisc->ops->attach(qdisc);
778 779 780 781
		}
	}
}

782 783 784 785
static void transition_one_qdisc(struct net_device *dev,
				 struct netdev_queue *dev_queue,
				 void *_need_watchdog)
{
786
	struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
787 788
	int *need_watchdog_p = _need_watchdog;

789 790 791
	if (!(new_qdisc->flags & TCQ_F_BUILTIN))
		clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);

792
	rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
793 794
	if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
		dev_queue->trans_start = 0;
795
		*need_watchdog_p = 1;
796
	}
797 798
}

L
Linus Torvalds 已提交
799 800
void dev_activate(struct net_device *dev)
{
801
	int need_watchdog;
802

L
Linus Torvalds 已提交
803
	/* No queueing discipline is attached to device;
804 805
	 * create default one for devices, which need queueing
	 * and noqueue_qdisc for virtual interfaces
L
Linus Torvalds 已提交
806 807
	 */

808 809
	if (dev->qdisc == &noop_qdisc)
		attach_default_qdiscs(dev);
810

811 812 813 814
	if (!netif_carrier_ok(dev))
		/* Delay activation until next carrier-on event */
		return;

815 816
	need_watchdog = 0;
	netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
817 818
	if (dev_ingress_queue(dev))
		transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
819 820

	if (need_watchdog) {
L
Linus Torvalds 已提交
821 822 823
		dev->trans_start = jiffies;
		dev_watchdog_up(dev);
	}
824
}
825
EXPORT_SYMBOL(dev_activate);
826

827 828 829
static void dev_deactivate_queue(struct net_device *dev,
				 struct netdev_queue *dev_queue,
				 void *_qdisc_default)
830
{
831
	struct Qdisc *qdisc_default = _qdisc_default;
832 833
	struct Qdisc *qdisc;

834
	qdisc = rtnl_dereference(dev_queue->qdisc);
835
	if (qdisc) {
836 837
		spin_lock_bh(qdisc_lock(qdisc));

838 839 840
		if (!(qdisc->flags & TCQ_F_BUILTIN))
			set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);

841
		rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
842
		qdisc_reset(qdisc);
843

844
		spin_unlock_bh(qdisc_lock(qdisc));
845
	}
L
Linus Torvalds 已提交
846 847
}

848
static bool some_qdisc_is_busy(struct net_device *dev)
849 850 851 852 853
{
	unsigned int i;

	for (i = 0; i < dev->num_tx_queues; i++) {
		struct netdev_queue *dev_queue;
854
		spinlock_t *root_lock;
855
		struct Qdisc *q;
856 857 858
		int val;

		dev_queue = netdev_get_tx_queue(dev, i);
859
		q = dev_queue->qdisc_sleeping;
860
		root_lock = qdisc_lock(q);
861

862
		spin_lock_bh(root_lock);
863

864
		val = (qdisc_is_running(q) ||
865
		       test_bit(__QDISC_STATE_SCHED, &q->state));
866

867
		spin_unlock_bh(root_lock);
868 869 870 871 872 873 874

		if (val)
			return true;
	}
	return false;
}

875 876 877 878 879 880 881
/**
 * 	dev_deactivate_many - deactivate transmissions on several devices
 * 	@head: list of devices to deactivate
 *
 *	This function returns only when all outstanding transmissions
 *	have completed, unless all devices are in dismantle phase.
 */
882
void dev_deactivate_many(struct list_head *head)
L
Linus Torvalds 已提交
883
{
884
	struct net_device *dev;
885
	bool sync_needed = false;
886

887
	list_for_each_entry(dev, head, close_list) {
888 889 890 891 892 893 894
		netdev_for_each_tx_queue(dev, dev_deactivate_queue,
					 &noop_qdisc);
		if (dev_ingress_queue(dev))
			dev_deactivate_queue(dev, dev_ingress_queue(dev),
					     &noop_qdisc);

		dev_watchdog_down(dev);
895
		sync_needed |= !dev->dismantle;
896
	}
L
Linus Torvalds 已提交
897

898 899 900 901 902 903
	/* Wait for outstanding qdisc-less dev_queue_xmit calls.
	 * This is avoided if all devices are in dismantle phase :
	 * Caller will call synchronize_net() for us
	 */
	if (sync_needed)
		synchronize_net();
L
Linus Torvalds 已提交
904

905
	/* Wait for outstanding qdisc_run calls. */
906
	list_for_each_entry(dev, head, close_list)
907 908 909 910 911 912 913 914
		while (some_qdisc_is_busy(dev))
			yield();
}

void dev_deactivate(struct net_device *dev)
{
	LIST_HEAD(single);

915
	list_add(&dev->close_list, &single);
916
	dev_deactivate_many(&single);
917
	list_del(&single);
L
Linus Torvalds 已提交
918
}
919
EXPORT_SYMBOL(dev_deactivate);
L
Linus Torvalds 已提交
920

921 922
static void dev_init_scheduler_queue(struct net_device *dev,
				     struct netdev_queue *dev_queue,
923
				     void *_qdisc)
924
{
925 926
	struct Qdisc *qdisc = _qdisc;

927
	rcu_assign_pointer(dev_queue->qdisc, qdisc);
928 929 930
	dev_queue->qdisc_sleeping = qdisc;
}

L
Linus Torvalds 已提交
931 932
void dev_init_scheduler(struct net_device *dev)
{
933
	dev->qdisc = &noop_qdisc;
934
	netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
935 936
	if (dev_ingress_queue(dev))
		dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
L
Linus Torvalds 已提交
937

938
	setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
L
Linus Torvalds 已提交
939 940
}

941 942 943
static void shutdown_scheduler_queue(struct net_device *dev,
				     struct netdev_queue *dev_queue,
				     void *_qdisc_default)
L
Linus Torvalds 已提交
944
{
945
	struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
946
	struct Qdisc *qdisc_default = _qdisc_default;
947 948

	if (qdisc) {
949
		rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
950
		dev_queue->qdisc_sleeping = qdisc_default;
L
Linus Torvalds 已提交
951 952

		qdisc_destroy(qdisc);
953
	}
954 955 956 957
}

void dev_shutdown(struct net_device *dev)
{
958
	netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
959 960
	if (dev_ingress_queue(dev))
		shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
961 962 963
	qdisc_destroy(dev->qdisc);
	dev->qdisc = &noop_qdisc;

964
	WARN_ON(timer_pending(&dev->watchdog_timer));
L
Linus Torvalds 已提交
965
}
966

967
void psched_ratecfg_precompute(struct psched_ratecfg *r,
968 969
			       const struct tc_ratespec *conf,
			       u64 rate64)
970
{
971 972
	memset(r, 0, sizeof(*r));
	r->overhead = conf->overhead;
973
	r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
974
	r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
975 976
	r->mult = 1;
	/*
977 978 979 980 981 982 983 984 985 986 987
	 * The deal here is to replace a divide by a reciprocal one
	 * in fast path (a reciprocal divide is a multiply and a shift)
	 *
	 * Normal formula would be :
	 *  time_in_ns = (NSEC_PER_SEC * len) / rate_bps
	 *
	 * We compute mult/shift to use instead :
	 *  time_in_ns = (len * mult) >> shift;
	 *
	 * We try to get the highest possible mult value for accuracy,
	 * but have to make sure no overflows will ever happen.
988
	 */
989 990 991 992 993 994
	if (r->rate_bytes_ps > 0) {
		u64 factor = NSEC_PER_SEC;

		for (;;) {
			r->mult = div64_u64(factor, r->rate_bytes_ps);
			if (r->mult & (1U << 31) || factor & (1ULL << 63))
995
				break;
996 997
			factor <<= 1;
			r->shift++;
998 999 1000 1001
		}
	}
}
EXPORT_SYMBOL(psched_ratecfg_precompute);