elevator.c 24.4 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
/*
 *  Block device elevator/IO-scheduler.
 *
 *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
 *
6
 * 30042000 Jens Axboe <axboe@kernel.dk> :
Linus Torvalds's avatar
Linus Torvalds committed
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
 *
 * Split the elevator a bit so that it is possible to choose a different
 * one or even write a new "plug in". There are three pieces:
 * - elevator_fn, inserts a new request in the queue list
 * - elevator_merge_fn, decides whether a new buffer can be merged with
 *   an existing request
 * - elevator_dequeue_fn, called when a request is taken off the active list
 *
 * 20082000 Dave Jones <davej@suse.de> :
 * Removed tests for max-bomb-segments, which was breaking elvtune
 *  when run without -bN
 *
 * Jens:
 * - Rework again to work with bio instead of buffer_heads
 * - loose bi_dev comparisons, partition handling is right now
 * - completely modularize elevator setup and teardown
 *
 */
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/blkdev.h>
#include <linux/elevator.h>
#include <linux/bio.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/init.h>
#include <linux/compiler.h>
34
#include <linux/delay.h>
35
#include <linux/blktrace_api.h>
36
#include <trace/block.h>
37
#include <linux/hash.h>
38
#include <linux/uaccess.h>
Linus Torvalds's avatar
Linus Torvalds committed
39

40
41
#include "blk.h"

Linus Torvalds's avatar
Linus Torvalds committed
42
43
44
static DEFINE_SPINLOCK(elv_list_lock);
static LIST_HEAD(elv_list);

45
46
DEFINE_TRACE(block_rq_abort);

47
48
49
50
51
/*
 * Merge hash stuff.
 */
static const int elv_hash_shift = 6;
#define ELV_HASH_BLOCK(sec)	((sec) >> 3)
52
53
#define ELV_HASH_FN(sec)	\
		(hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
54
#define ELV_HASH_ENTRIES	(1 << elv_hash_shift)
55
#define rq_hash_key(rq)		(blk_rq_pos(rq) + blk_rq_sectors(rq))
56

57
58
59
DEFINE_TRACE(block_rq_insert);
DEFINE_TRACE(block_rq_issue);

60
61
62
63
64
65
/*
 * Query io scheduler to see if the current process issuing bio may be
 * merged with rq.
 */
static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
{
66
	struct request_queue *q = rq->q;
67
	struct elevator_queue *e = q->elevator;
68
69
70
71
72
73
74

	if (e->ops->elevator_allow_merge_fn)
		return e->ops->elevator_allow_merge_fn(q, rq, bio);

	return 1;
}

Linus Torvalds's avatar
Linus Torvalds committed
75
76
77
/*
 * can we safely merge with this request?
 */
78
int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
79
80
81
82
{
	if (!rq_mergeable(rq))
		return 0;

83
84
85
86
87
88
	/*
	 * Don't merge file system requests and discard requests
	 */
	if (bio_discard(bio) != bio_discard(rq->bio))
		return 0;

Linus Torvalds's avatar
Linus Torvalds committed
89
90
91
92
93
94
95
	/*
	 * different data direction or already started, don't merge
	 */
	if (bio_data_dir(bio) != rq_data_dir(rq))
		return 0;

	/*
96
	 * must be same device and not a special request
Linus Torvalds's avatar
Linus Torvalds committed
97
	 */
98
	if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
99
100
		return 0;

101
102
103
104
105
106
	/*
	 * only merge integrity protected bio into ditto rq
	 */
	if (bio_integrity(bio) != blk_integrity_rq(rq))
		return 0;

107
108
	if (!elv_iosched_allow_merge(rq, bio))
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
109

110
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
111
112
113
}
EXPORT_SYMBOL(elv_rq_merge_ok);

114
static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
115
116
117
118
119
120
121
{
	int ret = ELEVATOR_NO_MERGE;

	/*
	 * we can merge and sequence is ok, check if it's possible
	 */
	if (elv_rq_merge_ok(__rq, bio)) {
122
		if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
Linus Torvalds's avatar
Linus Torvalds committed
123
			ret = ELEVATOR_BACK_MERGE;
124
		else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
Linus Torvalds's avatar
Linus Torvalds committed
125
126
127
128
129
130
131
132
			ret = ELEVATOR_FRONT_MERGE;
	}

	return ret;
}

static struct elevator_type *elevator_find(const char *name)
{
133
	struct elevator_type *e;
Linus Torvalds's avatar
Linus Torvalds committed
134

135
	list_for_each_entry(e, &elv_list, list) {
136
137
		if (!strcmp(e->elevator_name, name))
			return e;
Linus Torvalds's avatar
Linus Torvalds committed
138
139
	}

140
	return NULL;
Linus Torvalds's avatar
Linus Torvalds committed
141
142
143
144
145
146
147
148
149
}

static void elevator_put(struct elevator_type *e)
{
	module_put(e->elevator_owner);
}

static struct elevator_type *elevator_get(const char *name)
{
150
	struct elevator_type *e;
Linus Torvalds's avatar
Linus Torvalds committed
151

152
	spin_lock(&elv_list_lock);
153
154

	e = elevator_find(name);
155
156
157
158
159
160
161
162
163
164
	if (!e) {
		char elv[ELV_NAME_MAX + strlen("-iosched")];

		spin_unlock(&elv_list_lock);

		if (!strcmp(name, "anticipatory"))
			sprintf(elv, "as-iosched");
		else
			sprintf(elv, "%s-iosched", name);

165
		request_module("%s", elv);
166
167
168
169
		spin_lock(&elv_list_lock);
		e = elevator_find(name);
	}

170
171
172
	if (e && !try_module_get(e->elevator_owner))
		e = NULL;

173
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
174
175
176
177

	return e;
}

178
179
static void *elevator_init_queue(struct request_queue *q,
				 struct elevator_queue *eq)
Linus Torvalds's avatar
Linus Torvalds committed
180
{
181
	return eq->ops->elevator_init_fn(q);
Jens Axboe's avatar
Jens Axboe committed
182
}
Linus Torvalds's avatar
Linus Torvalds committed
183

184
static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboe's avatar
Jens Axboe committed
185
186
			   void *data)
{
Linus Torvalds's avatar
Linus Torvalds committed
187
	q->elevator = eq;
Jens Axboe's avatar
Jens Axboe committed
188
	eq->elevator_data = data;
Linus Torvalds's avatar
Linus Torvalds committed
189
190
191
192
}

static char chosen_elevator[16];

193
static int __init elevator_setup(char *str)
Linus Torvalds's avatar
Linus Torvalds committed
194
{
195
196
197
198
	/*
	 * Be backwards-compatible with previous kernels, so users
	 * won't get the wrong elevator.
	 */
199
	if (!strcmp(str, "as"))
200
		strcpy(chosen_elevator, "anticipatory");
201
	else
202
		strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
203
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
204
205
206
207
}

__setup("elevator=", elevator_setup);

208
209
static struct kobj_type elv_ktype;

210
static struct elevator_queue *elevator_alloc(struct request_queue *q,
211
				  struct elevator_type *e)
212
{
213
	struct elevator_queue *eq;
214
215
	int i;

216
	eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
217
218
219
220
221
	if (unlikely(!eq))
		goto err;

	eq->ops = &e->ops;
	eq->elevator_type = e;
222
	kobject_init(&eq->kobj, &elv_ktype);
223
224
	mutex_init(&eq->sysfs_lock);

225
226
	eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
					GFP_KERNEL, q->node);
227
228
229
230
231
232
	if (!eq->hash)
		goto err;

	for (i = 0; i < ELV_HASH_ENTRIES; i++)
		INIT_HLIST_HEAD(&eq->hash[i]);

233
	return eq;
234
235
236
237
err:
	kfree(eq);
	elevator_put(e);
	return NULL;
238
239
240
241
}

static void elevator_release(struct kobject *kobj)
{
242
	struct elevator_queue *e;
243

244
	e = container_of(kobj, struct elevator_queue, kobj);
245
	elevator_put(e->elevator_type);
246
	kfree(e->hash);
247
248
249
	kfree(e);
}

250
int elevator_init(struct request_queue *q, char *name)
Linus Torvalds's avatar
Linus Torvalds committed
251
252
253
254
{
	struct elevator_type *e = NULL;
	struct elevator_queue *eq;
	int ret = 0;
Jens Axboe's avatar
Jens Axboe committed
255
	void *data;
Linus Torvalds's avatar
Linus Torvalds committed
256

257
258
259
260
261
	INIT_LIST_HEAD(&q->queue_head);
	q->last_merge = NULL;
	q->end_sector = 0;
	q->boundary_rq = NULL;

262
263
264
265
266
	if (name) {
		e = elevator_get(name);
		if (!e)
			return -EINVAL;
	}
Linus Torvalds's avatar
Linus Torvalds committed
267

268
269
270
271
272
273
	if (!e && *chosen_elevator) {
		e = elevator_get(chosen_elevator);
		if (!e)
			printk(KERN_ERR "I/O scheduler %s not found\n",
							chosen_elevator);
	}
274

275
276
277
278
279
280
281
282
	if (!e) {
		e = elevator_get(CONFIG_DEFAULT_IOSCHED);
		if (!e) {
			printk(KERN_ERR
				"Default I/O scheduler not found. " \
				"Using noop.\n");
			e = elevator_get("noop");
		}
283
284
	}

285
	eq = elevator_alloc(q, e);
286
	if (!eq)
Linus Torvalds's avatar
Linus Torvalds committed
287
288
		return -ENOMEM;

Jens Axboe's avatar
Jens Axboe committed
289
290
	data = elevator_init_queue(q, eq);
	if (!data) {
291
		kobject_put(&eq->kobj);
Jens Axboe's avatar
Jens Axboe committed
292
293
		return -ENOMEM;
	}
Linus Torvalds's avatar
Linus Torvalds committed
294

Jens Axboe's avatar
Jens Axboe committed
295
	elevator_attach(q, eq, data);
Linus Torvalds's avatar
Linus Torvalds committed
296
297
	return ret;
}
298
299
EXPORT_SYMBOL(elevator_init);

300
void elevator_exit(struct elevator_queue *e)
Linus Torvalds's avatar
Linus Torvalds committed
301
{
302
	mutex_lock(&e->sysfs_lock);
Linus Torvalds's avatar
Linus Torvalds committed
303
304
	if (e->ops->elevator_exit_fn)
		e->ops->elevator_exit_fn(e);
305
306
	e->ops = NULL;
	mutex_unlock(&e->sysfs_lock);
Linus Torvalds's avatar
Linus Torvalds committed
307

308
	kobject_put(&e->kobj);
Linus Torvalds's avatar
Linus Torvalds committed
309
}
310
311
EXPORT_SYMBOL(elevator_exit);

312
313
314
315
316
static inline void __elv_rqhash_del(struct request *rq)
{
	hlist_del_init(&rq->hash);
}

317
static void elv_rqhash_del(struct request_queue *q, struct request *rq)
318
319
320
321
322
{
	if (ELV_ON_HASH(rq))
		__elv_rqhash_del(rq);
}

323
static void elv_rqhash_add(struct request_queue *q, struct request *rq)
324
{
325
	struct elevator_queue *e = q->elevator;
326
327
328
329
330

	BUG_ON(ELV_ON_HASH(rq));
	hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
}

331
static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
332
333
334
335
336
{
	__elv_rqhash_del(rq);
	elv_rqhash_add(q, rq);
}

337
static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
338
{
339
	struct elevator_queue *e = q->elevator;
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
	struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
	struct hlist_node *entry, *next;
	struct request *rq;

	hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
		BUG_ON(!ELV_ON_HASH(rq));

		if (unlikely(!rq_mergeable(rq))) {
			__elv_rqhash_del(rq);
			continue;
		}

		if (rq_hash_key(rq) == offset)
			return rq;
	}

	return NULL;
}

359
360
361
362
363
364
365
366
367
368
369
370
371
372
/*
 * RB-tree support functions for inserting/lookup/removal of requests
 * in a sorted RB tree.
 */
struct request *elv_rb_add(struct rb_root *root, struct request *rq)
{
	struct rb_node **p = &root->rb_node;
	struct rb_node *parent = NULL;
	struct request *__rq;

	while (*p) {
		parent = *p;
		__rq = rb_entry(parent, struct request, rb_node);

373
		if (blk_rq_pos(rq) < blk_rq_pos(__rq))
374
			p = &(*p)->rb_left;
375
		else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
			p = &(*p)->rb_right;
		else
			return __rq;
	}

	rb_link_node(&rq->rb_node, parent, p);
	rb_insert_color(&rq->rb_node, root);
	return NULL;
}
EXPORT_SYMBOL(elv_rb_add);

void elv_rb_del(struct rb_root *root, struct request *rq)
{
	BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
	rb_erase(&rq->rb_node, root);
	RB_CLEAR_NODE(&rq->rb_node);
}
EXPORT_SYMBOL(elv_rb_del);

struct request *elv_rb_find(struct rb_root *root, sector_t sector)
{
	struct rb_node *n = root->rb_node;
	struct request *rq;

	while (n) {
		rq = rb_entry(n, struct request, rb_node);

403
		if (sector < blk_rq_pos(rq))
404
			n = n->rb_left;
405
		else if (sector > blk_rq_pos(rq))
406
407
408
409
410
411
412
413
414
			n = n->rb_right;
		else
			return rq;
	}

	return NULL;
}
EXPORT_SYMBOL(elv_rb_find);

415
416
/*
 * Insert rq into dispatch queue of q.  Queue lock must be held on
417
 * entry.  rq is sort instead into the dispatch queue. To be used by
418
 * specific elevators.
419
 */
420
void elv_dispatch_sort(struct request_queue *q, struct request *rq)
421
422
423
{
	sector_t boundary;
	struct list_head *entry;
424
	int stop_flags;
425

426
427
	if (q->last_merge == rq)
		q->last_merge = NULL;
428
429
430

	elv_rqhash_del(q, rq);

431
	q->nr_sorted--;
432

Jens Axboe's avatar
Jens Axboe committed
433
	boundary = q->end_sector;
434
	stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
435
436
437
	list_for_each_prev(entry, &q->queue_head) {
		struct request *pos = list_entry_rq(entry);

438
439
		if (blk_discard_rq(rq) != blk_discard_rq(pos))
			break;
440
441
		if (rq_data_dir(rq) != rq_data_dir(pos))
			break;
442
		if (pos->cmd_flags & stop_flags)
443
			break;
444
445
		if (blk_rq_pos(rq) >= boundary) {
			if (blk_rq_pos(pos) < boundary)
446
447
				continue;
		} else {
448
			if (blk_rq_pos(pos) >= boundary)
449
450
				break;
		}
451
		if (blk_rq_pos(rq) >= blk_rq_pos(pos))
452
453
454
455
456
			break;
	}

	list_add(&rq->queuelist, entry);
}
457
458
EXPORT_SYMBOL(elv_dispatch_sort);

459
/*
460
461
462
 * Insert rq into dispatch queue of q.  Queue lock must be held on
 * entry.  rq is added to the back of the dispatch queue. To be used by
 * specific elevators.
463
464
465
466
467
468
469
470
471
472
473
474
475
476
 */
void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
{
	if (q->last_merge == rq)
		q->last_merge = NULL;

	elv_rqhash_del(q, rq);

	q->nr_sorted--;

	q->end_sector = rq_end_sector(rq);
	q->boundary_rq = rq;
	list_add_tail(&rq->queuelist, &q->queue_head);
}
477
478
EXPORT_SYMBOL(elv_dispatch_add_tail);

479
int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds's avatar
Linus Torvalds committed
480
{
481
	struct elevator_queue *e = q->elevator;
482
	struct request *__rq;
483
484
	int ret;

485
486
487
	/*
	 * First try one-hit cache.
	 */
488
489
490
491
492
493
494
	if (q->last_merge) {
		ret = elv_try_merge(q->last_merge, bio);
		if (ret != ELEVATOR_NO_MERGE) {
			*req = q->last_merge;
			return ret;
		}
	}
Linus Torvalds's avatar
Linus Torvalds committed
495

496
497
498
	if (blk_queue_nomerges(q))
		return ELEVATOR_NO_MERGE;

499
500
501
502
503
504
505
506
507
	/*
	 * See if our hash lookup can find a potential backmerge.
	 */
	__rq = elv_rqhash_find(q, bio->bi_sector);
	if (__rq && elv_rq_merge_ok(__rq, bio)) {
		*req = __rq;
		return ELEVATOR_BACK_MERGE;
	}

Linus Torvalds's avatar
Linus Torvalds committed
508
509
510
511
512
513
	if (e->ops->elevator_merge_fn)
		return e->ops->elevator_merge_fn(q, req, bio);

	return ELEVATOR_NO_MERGE;
}

514
void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds's avatar
Linus Torvalds committed
515
{
516
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
517
518

	if (e->ops->elevator_merged_fn)
519
		e->ops->elevator_merged_fn(q, rq, type);
520

521
522
	if (type == ELEVATOR_BACK_MERGE)
		elv_rqhash_reposition(q, rq);
523

524
	q->last_merge = rq;
Linus Torvalds's avatar
Linus Torvalds committed
525
526
}

527
void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds's avatar
Linus Torvalds committed
528
529
			     struct request *next)
{
530
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
531
532
533

	if (e->ops->elevator_merge_req_fn)
		e->ops->elevator_merge_req_fn(q, rq, next);
534

535
536
537
538
	elv_rqhash_reposition(q, rq);
	elv_rqhash_del(q, next);

	q->nr_sorted--;
539
	q->last_merge = rq;
Linus Torvalds's avatar
Linus Torvalds committed
540
541
}

542
void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
543
544
545
546
547
{
	/*
	 * it already went through dequeue, we need to decrement the
	 * in_flight count again
	 */
548
	if (blk_account_rq(rq)) {
549
		q->in_flight[rq_is_sync(rq)]--;
550
551
		if (blk_sorted_rq(rq))
			elv_deactivate_rq(q, rq);
552
	}
Linus Torvalds's avatar
Linus Torvalds committed
553

554
	rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds's avatar
Linus Torvalds committed
555

556
	elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds's avatar
Linus Torvalds committed
557
558
}

559
void elv_drain_elevator(struct request_queue *q)
560
561
562
563
564
565
566
567
568
569
570
571
572
{
	static int printed;
	while (q->elevator->ops->elevator_dispatch_fn(q, 1))
		;
	if (q->nr_sorted == 0)
		return;
	if (printed++ < 10) {
		printk(KERN_ERR "%s: forced dispatching is broken "
		       "(nr_sorted=%u), please report this\n",
		       q->elevator->elevator_type->elevator_name, q->nr_sorted);
	}
}

Jens Axboe's avatar
Jens Axboe committed
573
574
575
/*
 * Call with queue lock held, interrupts disabled
 */
576
void elv_quiesce_start(struct request_queue *q)
Jens Axboe's avatar
Jens Axboe committed
577
{
578
579
580
	if (!q->elevator)
		return;

Jens Axboe's avatar
Jens Axboe committed
581
582
583
584
585
586
587
	queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);

	/*
	 * make sure we don't have any requests in flight
	 */
	elv_drain_elevator(q);
	while (q->rq.elvpriv) {
Tejun Heo's avatar
Tejun Heo committed
588
		__blk_run_queue(q);
Jens Axboe's avatar
Jens Axboe committed
589
590
591
592
593
594
595
		spin_unlock_irq(q->queue_lock);
		msleep(10);
		spin_lock_irq(q->queue_lock);
		elv_drain_elevator(q);
	}
}

596
void elv_quiesce_end(struct request_queue *q)
Jens Axboe's avatar
Jens Axboe committed
597
598
599
600
{
	queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
}

601
void elv_insert(struct request_queue *q, struct request *rq, int where)
Linus Torvalds's avatar
Linus Torvalds committed
602
{
603
604
	struct list_head *pos;
	unsigned ordseq;
605
	int unplug_it = 1;
606

607
	trace_block_rq_insert(q, rq);
608

Linus Torvalds's avatar
Linus Torvalds committed
609
610
	rq->q = q;

611
612
	switch (where) {
	case ELEVATOR_INSERT_FRONT:
613
		rq->cmd_flags |= REQ_SOFTBARRIER;
614
615
616
617
618

		list_add(&rq->queuelist, &q->queue_head);
		break;

	case ELEVATOR_INSERT_BACK:
619
		rq->cmd_flags |= REQ_SOFTBARRIER;
620
		elv_drain_elevator(q);
621
622
623
624
625
626
627
628
629
630
631
		list_add_tail(&rq->queuelist, &q->queue_head);
		/*
		 * We kick the queue here for the following reasons.
		 * - The elevator might have returned NULL previously
		 *   to delay requests and returned them now.  As the
		 *   queue wasn't empty before this request, ll_rw_blk
		 *   won't run the queue on return, resulting in hang.
		 * - Usually, back inserted requests won't be merged
		 *   with anything.  There's no point in delaying queue
		 *   processing.
		 */
Tejun Heo's avatar
Tejun Heo committed
632
		__blk_run_queue(q);
633
634
635
		break;

	case ELEVATOR_INSERT_SORT:
636
		BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
637
		rq->cmd_flags |= REQ_SORTED;
638
		q->nr_sorted++;
639
640
641
642
643
644
		if (rq_mergeable(rq)) {
			elv_rqhash_add(q, rq);
			if (!q->last_merge)
				q->last_merge = rq;
		}

645
646
647
648
649
650
		/*
		 * Some ioscheds (cfq) run q->request_fn directly, so
		 * rq cannot be accessed after calling
		 * elevator_add_req_fn.
		 */
		q->elevator->ops->elevator_add_req_fn(q, rq);
651
652
		break;

653
654
655
656
657
658
	case ELEVATOR_INSERT_REQUEUE:
		/*
		 * If ordered flush isn't in progress, we do front
		 * insertion; otherwise, requests should be requeued
		 * in ordseq order.
		 */
659
		rq->cmd_flags |= REQ_SOFTBARRIER;
660

661
662
663
664
665
666
		/*
		 * Most requeues happen because of a busy condition,
		 * don't force unplug of the queue for that case.
		 */
		unplug_it = 0;

667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
		if (q->ordseq == 0) {
			list_add(&rq->queuelist, &q->queue_head);
			break;
		}

		ordseq = blk_ordered_req_seq(rq);

		list_for_each(pos, &q->queue_head) {
			struct request *pos_rq = list_entry_rq(pos);
			if (ordseq <= blk_ordered_req_seq(pos_rq))
				break;
		}

		list_add_tail(&rq->queuelist, pos);
		break;

683
684
	default:
		printk(KERN_ERR "%s: bad insertion point %d\n",
685
		       __func__, where);
686
687
688
		BUG();
	}

689
	if (unplug_it && blk_queue_plugged(q)) {
690
		int nrq = q->rq.count[BLK_RW_SYNC] + q->rq.count[BLK_RW_ASYNC]
691
				- queue_in_flight(q);
692
693
694
695

		if (nrq >= q->unplug_thresh)
			__generic_unplug_device(q);
	}
Linus Torvalds's avatar
Linus Torvalds committed
696
697
}

698
void __elv_add_request(struct request_queue *q, struct request *rq, int where,
699
700
701
		       int plug)
{
	if (q->ordcolor)
702
		rq->cmd_flags |= REQ_ORDERED_COLOR;
703

704
	if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
		/*
		 * toggle ordered color
		 */
		if (blk_barrier_rq(rq))
			q->ordcolor ^= 1;

		/*
		 * barriers implicitly indicate back insertion
		 */
		if (where == ELEVATOR_INSERT_SORT)
			where = ELEVATOR_INSERT_BACK;

		/*
		 * this request is scheduling boundary, update
		 * end_sector
		 */
721
		if (blk_fs_request(rq) || blk_discard_rq(rq)) {
722
723
724
			q->end_sector = rq_end_sector(rq);
			q->boundary_rq = rq;
		}
725
726
	} else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
		    where == ELEVATOR_INSERT_SORT)
727
728
729
730
731
732
733
		where = ELEVATOR_INSERT_BACK;

	if (plug)
		blk_plug_device(q);

	elv_insert(q, rq, where);
}
734
735
EXPORT_SYMBOL(__elv_add_request);

736
void elv_add_request(struct request_queue *q, struct request *rq, int where,
Linus Torvalds's avatar
Linus Torvalds committed
737
738
739
740
741
742
743
744
		     int plug)
{
	unsigned long flags;

	spin_lock_irqsave(q->queue_lock, flags);
	__elv_add_request(q, rq, where, plug);
	spin_unlock_irqrestore(q->queue_lock, flags);
}
745
746
EXPORT_SYMBOL(elv_add_request);

747
int elv_queue_empty(struct request_queue *q)
Linus Torvalds's avatar
Linus Torvalds committed
748
{
749
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
750

751
752
753
	if (!list_empty(&q->queue_head))
		return 0;

Linus Torvalds's avatar
Linus Torvalds committed
754
755
756
	if (e->ops->elevator_queue_empty_fn)
		return e->ops->elevator_queue_empty_fn(q);

757
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
758
}
759
760
EXPORT_SYMBOL(elv_queue_empty);

761
struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
762
{
763
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
764
765
766
767
768
769

	if (e->ops->elevator_latter_req_fn)
		return e->ops->elevator_latter_req_fn(q, rq);
	return NULL;
}

770
struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
771
{
772
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
773
774
775
776
777
778

	if (e->ops->elevator_former_req_fn)
		return e->ops->elevator_former_req_fn(q, rq);
	return NULL;
}

779
int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds's avatar
Linus Torvalds committed
780
{
781
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
782
783

	if (e->ops->elevator_set_req_fn)
784
		return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds's avatar
Linus Torvalds committed
785
786
787
788
789

	rq->elevator_private = NULL;
	return 0;
}

790
void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
791
{
792
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
793
794

	if (e->ops->elevator_put_req_fn)
795
		e->ops->elevator_put_req_fn(rq);
Linus Torvalds's avatar
Linus Torvalds committed
796
797
}

798
int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds's avatar
Linus Torvalds committed
799
{
800
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
801
802

	if (e->ops->elevator_may_queue_fn)
803
		return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds's avatar
Linus Torvalds committed
804
805
806
807

	return ELV_MQUEUE_MAY;
}

808
809
810
811
812
813
814
void elv_abort_queue(struct request_queue *q)
{
	struct request *rq;

	while (!list_empty(&q->queue_head)) {
		rq = list_entry_rq(q->queue_head.next);
		rq->cmd_flags |= REQ_QUIET;
815
		trace_block_rq_abort(q, rq);
816
		__blk_end_request_all(rq, -EIO);
817
818
819
820
	}
}
EXPORT_SYMBOL(elv_abort_queue);

821
void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds's avatar
Linus Torvalds committed
822
{
823
	struct elevator_queue *e = q->elevator;
Linus Torvalds's avatar
Linus Torvalds committed
824
825
826
827

	/*
	 * request is released from the driver, io must be done
	 */
828
	if (blk_account_rq(rq)) {
829
		q->in_flight[rq_is_sync(rq)]--;
830
831
832
		if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
			e->ops->elevator_completed_req_fn(q, rq);
	}
833

834
835
836
837
838
	/*
	 * Check if the queue is waiting for fs requests to be
	 * drained for flush sequence.
	 */
	if (unlikely(q->ordseq)) {
839
840
841
842
843
		struct request *next = NULL;

		if (!list_empty(&q->queue_head))
			next = list_entry_rq(q->queue_head.next);

844
		if (!queue_in_flight(q) &&
845
		    blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
846
		    (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
847
			blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Tejun Heo's avatar
Tejun Heo committed
848
			__blk_run_queue(q);
849
		}
850
	}
Linus Torvalds's avatar
Linus Torvalds committed
851
852
}

853
854
855
856
#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)

static ssize_t
elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
Linus Torvalds's avatar
Linus Torvalds committed
857
{
858
	struct elv_fs_entry *entry = to_elv(attr);
859
	struct elevator_queue *e;
860
861
862
863
864
	ssize_t error;

	if (!entry->show)
		return -EIO;

865
	e = container_of(kobj, struct elevator_queue, kobj);
866
867
868
869
870
	mutex_lock(&e->sysfs_lock);
	error = e->ops ? entry->show(e, page) : -ENOENT;
	mutex_unlock(&e->sysfs_lock);
	return error;
}
Linus Torvalds's avatar
Linus Torvalds committed
871

872
873
874
875
876
static ssize_t
elv_attr_store(struct kobject *kobj, struct attribute *attr,
	       const char *page, size_t length)
{
	struct elv_fs_entry *entry = to_elv(attr);
877
	struct elevator_queue *e;
878
	ssize_t error;
Linus Torvalds's avatar
Linus Torvalds committed
879

880
881
	if (!entry->store)
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
882

883
	e = container_of(kobj, struct elevator_queue, kobj);
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
	mutex_lock(&e->sysfs_lock);
	error = e->ops ? entry->store(e, page, length) : -ENOENT;
	mutex_unlock(&e->sysfs_lock);
	return error;
}

static struct sysfs_ops elv_sysfs_ops = {
	.show	= elv_attr_show,
	.store	= elv_attr_store,
};

static struct kobj_type elv_ktype = {
	.sysfs_ops	= &elv_sysfs_ops,
	.release	= elevator_release,
};

int elv_register_queue(struct request_queue *q)
{
902
	struct elevator_queue *e = q->elevator;
903
904
	int error;

905
	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
906
	if (!error) {
907
		struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
908
		if (attr) {
909
910
			while (attr->attr.name) {
				if (sysfs_create_file(&e->kobj, &attr->attr))
911
					break;
912
				attr++;
913
914
915
916
917
			}
		}
		kobject_uevent(&e->kobj, KOBJ_ADD);
	}
	return error;
Linus Torvalds's avatar
Linus Torvalds committed
918
919
}

920
static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboe's avatar
Jens Axboe committed
921
922
923
924
925
{
	kobject_uevent(&e->kobj, KOBJ_REMOVE);
	kobject_del(&e->kobj);
}

Linus Torvalds's avatar
Linus Torvalds committed
926
927
void elv_unregister_queue(struct request_queue *q)
{
Jens Axboe's avatar
Jens Axboe committed
928
929
	if (q)
		__elv_unregister_queue(q->elevator);
Linus Torvalds's avatar
Linus Torvalds committed
930
931
}

932
void elv_register(struct elevator_type *e)
Linus Torvalds's avatar
Linus Torvalds committed
933
{
934
	char *def = "";
935
936

	spin_lock(&elv_list_lock);
937
	BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds's avatar
Linus Torvalds committed
938
	list_add_tail(&e->list, &elv_list);
939
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
940

941
942
943
	if (!strcmp(e->elevator_name, chosen_elevator) ||
			(!*chosen_elevator &&
			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
944
945
				def = " (default)";

946
947
	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
								def);
Linus Torvalds's avatar
Linus Torvalds committed
948
949
950
951
952
}
EXPORT_SYMBOL_GPL(elv_register);

void elv_unregister(struct elevator_type *e)
{
953
954
955
956
957
	struct task_struct *g, *p;

	/*
	 * Iterate every thread in the process to remove the io contexts.
	 */
958
959
960
961
	if (e->ops.trim) {
		read_lock(&tasklist_lock);
		do_each_thread(g, p) {
			task_lock(p);
962
963
			if (p->io_context)
				e->ops.trim(p->io_context);
964
965
966
967
			task_unlock(p);
		} while_each_thread(g, p);
		read_unlock(&tasklist_lock);
	}
968

969
	spin_lock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
970
	list_del_init(&e->list);
971
	spin_unlock(&elv_list_lock);
Linus Torvalds's avatar
Linus Torvalds committed
972
973
974
975
976
977
978
}
EXPORT_SYMBOL_GPL(elv_unregister);

/*
 * switch to new_e io scheduler. be careful not to introduce deadlocks -
 * we don't free the old io scheduler, before we have allocated what we
 * need for the new one. this way we have a chance of going back to the old
979
 * one, if the new one fails init for some reason.
Linus Torvalds's avatar
Linus Torvalds committed
980
 */
981
static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds's avatar
Linus Torvalds committed
982
{
983
	struct elevator_queue *old_elevator, *e;
Jens Axboe's avatar
Jens Axboe committed
984
	void *data;
Linus Torvalds's avatar
Linus Torvalds committed
985

986
987
988
	/*
	 * Allocate new elevator
	 */
989
	e = elevator_alloc(q, new_e);
Linus Torvalds's avatar
Linus Torvalds committed
990
	if (!e)
991
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
992

Jens Axboe's avatar
Jens Axboe committed
993
994
995
996
997
998
	data = elevator_init_queue(q, e);
	if (!data) {
		kobject_put(&e->kobj);
		return 0;
	}

Linus Torvalds's avatar
Linus Torvalds committed
999
	/*
1000
	 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds's avatar
Linus Torvalds committed
1001
	 */
1002
	spin_lock_irq(q->queue_lock);
1003
	elv_quiesce_start(q);
1004

Linus Torvalds's avatar
Linus Torvalds committed
1005
	/*
Jens Axboe's avatar
Jens Axboe committed
1006
	 * Remember old elevator.
Linus Torvalds's avatar
Linus Torvalds committed
1007
1008
1009
1010
1011
1012
	 */
	old_elevator = q->elevator;

	/*
	 * attach and start new elevator
	 */
Jens Axboe's avatar
Jens Axboe committed
1013
1014
1015
1016
1017
	elevator_attach(q, e, data);

	spin_unlock_irq(q->queue_lock);

	__elv_unregister_queue(old_elevator);
Linus Torvalds's avatar
Linus Torvalds committed
1018
1019
1020
1021
1022

	if (elv_register_queue(q))
		goto fail_register;

	/*
1023
	 * finally exit old elevator and turn off BYPASS.
Linus Torvalds's avatar
Linus Torvalds committed
1024
1025
	 */
	elevator_exit(old_elevator);
1026
	spin_lock_irq(q->queue_lock);
1027
	elv_quiesce_end(q);
1028
1029
	spin_unlock_irq(q->queue_lock);

1030
1031
	blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);

1032
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
1033
1034
1035
1036
1037
1038
1039
1040
1041

fail_register:
	/*
	 * switch failed, exit the new io scheduler and reattach the old
	 * one again (along with re-adding the sysfs dir)
	 */
	elevator_exit(e);
	q->elevator = old_elevator;
	elv_register_queue(q);
1042
1043
1044
1045
1046

	spin_lock_irq(q->queue_lock);
	queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
	spin_unlock_irq(q->queue_lock);

1047
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
1048
1049
}

1050
1051
ssize_t elv_iosched_store(struct request_queue *q, const char *name,
			  size_t count)
Linus Torvalds's avatar
Linus Torvalds committed
1052
1053
1054
1055
{
	char elevator_name[ELV_NAME_MAX];
	struct elevator_type *e;

1056
1057
1058
	if (!q->elevator)
		return count;

1059
1060
	strlcpy(elevator_name, name, sizeof(elevator_name));
	strstrip(elevator_name);
Linus Torvalds's avatar
Linus Torvalds committed
1061
1062
1063
1064
1065
1066
1067

	e = elevator_get(elevator_name);
	if (!e) {
		printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
		return -EINVAL;
	}

1068
1069
	if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
		elevator_put(e);
Linus Torvalds's avatar
Linus Torvalds committed
1070
		return count;
1071
	}
Linus Torvalds's avatar
Linus Torvalds committed
1072

1073
	if (!elevator_switch(q, e))
1074
1075
		printk(KERN_ERR "elevator: switch to %s failed\n",
							elevator_name);
Linus Torvalds's avatar
Linus Torvalds committed
1076
1077
1078
	return count;
}

1079
ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds's avatar
Linus Torvalds committed
1080
{
1081
	struct elevator_queue *e = q->elevator;