elevator.c 23.0 KB
Newer Older
L
Linus Torvalds 已提交
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> :
L
Linus Torvalds 已提交
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/blktrace_api.h>
35
#include <linux/hash.h>
36
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
37

38 39
#include <trace/events/block.h>

J
Jens Axboe 已提交
40
#include "blk.h"
41
#include "blk-cgroup.h"
J
Jens Axboe 已提交
42

L
Linus Torvalds 已提交
43 44 45
static DEFINE_SPINLOCK(elv_list_lock);
static LIST_HEAD(elv_list);

46 47 48 49 50
/*
 * Merge hash stuff.
 */
static const int elv_hash_shift = 6;
#define ELV_HASH_BLOCK(sec)	((sec) >> 3)
51 52
#define ELV_HASH_FN(sec)	\
		(hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
53
#define ELV_HASH_ENTRIES	(1 << elv_hash_shift)
54
#define rq_hash_key(rq)		(blk_rq_pos(rq) + blk_rq_sectors(rq))
55

56 57 58 59 60 61
/*
 * 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)
{
62
	struct request_queue *q = rq->q;
J
Jens Axboe 已提交
63
	struct elevator_queue *e = q->elevator;
64

T
Tejun Heo 已提交
65 66
	if (e->type->ops.elevator_allow_merge_fn)
		return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
67 68 69 70

	return 1;
}

L
Linus Torvalds 已提交
71 72 73
/*
 * can we safely merge with this request?
 */
74
bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
L
Linus Torvalds 已提交
75
{
76
	if (!blk_rq_merge_ok(rq, bio))
77 78
		return 0;

79 80
	if (!elv_iosched_allow_merge(rq, bio))
		return 0;
L
Linus Torvalds 已提交
81

82
	return 1;
L
Linus Torvalds 已提交
83 84 85 86 87
}
EXPORT_SYMBOL(elv_rq_merge_ok);

static struct elevator_type *elevator_find(const char *name)
{
88
	struct elevator_type *e;
L
Linus Torvalds 已提交
89

90
	list_for_each_entry(e, &elv_list, list) {
91 92
		if (!strcmp(e->elevator_name, name))
			return e;
L
Linus Torvalds 已提交
93 94
	}

95
	return NULL;
L
Linus Torvalds 已提交
96 97 98 99 100 101 102 103 104
}

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

static struct elevator_type *elevator_get(const char *name)
{
105
	struct elevator_type *e;
L
Linus Torvalds 已提交
106

107
	spin_lock(&elv_list_lock);
108 109

	e = elevator_find(name);
110 111
	if (!e) {
		spin_unlock(&elv_list_lock);
K
Kees Cook 已提交
112
		request_module("%s-iosched", name);
113 114 115 116
		spin_lock(&elv_list_lock);
		e = elevator_find(name);
	}

117 118 119
	if (e && !try_module_get(e->elevator_owner))
		e = NULL;

120
	spin_unlock(&elv_list_lock);
L
Linus Torvalds 已提交
121 122 123 124

	return e;
}

125
static char chosen_elevator[ELV_NAME_MAX];
L
Linus Torvalds 已提交
126

127
static int __init elevator_setup(char *str)
L
Linus Torvalds 已提交
128
{
129 130 131 132
	/*
	 * Be backwards-compatible with previous kernels, so users
	 * won't get the wrong elevator.
	 */
133
	strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
134
	return 1;
L
Linus Torvalds 已提交
135 136 137 138
}

__setup("elevator=", elevator_setup);

139 140
static struct kobj_type elv_ktype;

J
Jens Axboe 已提交
141
static struct elevator_queue *elevator_alloc(struct request_queue *q,
142
				  struct elevator_type *e)
143
{
J
Jens Axboe 已提交
144
	struct elevator_queue *eq;
145 146
	int i;

J
Jens Axboe 已提交
147
	eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
148 149 150
	if (unlikely(!eq))
		goto err;

T
Tejun Heo 已提交
151
	eq->type = e;
152
	kobject_init(&eq->kobj, &elv_ktype);
153 154
	mutex_init(&eq->sysfs_lock);

155 156
	eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
					GFP_KERNEL, q->node);
157 158 159 160 161 162
	if (!eq->hash)
		goto err;

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

163
	return eq;
164 165 166 167
err:
	kfree(eq);
	elevator_put(e);
	return NULL;
168 169 170 171
}

static void elevator_release(struct kobject *kobj)
{
J
Jens Axboe 已提交
172
	struct elevator_queue *e;
173

J
Jens Axboe 已提交
174
	e = container_of(kobj, struct elevator_queue, kobj);
T
Tejun Heo 已提交
175
	elevator_put(e->type);
176
	kfree(e->hash);
177 178 179
	kfree(e);
}

180
int elevator_init(struct request_queue *q, char *name)
L
Linus Torvalds 已提交
181 182
{
	struct elevator_type *e = NULL;
183
	int err;
L
Linus Torvalds 已提交
184

185 186 187
	if (unlikely(q->elevator))
		return 0;

T
Tejun Heo 已提交
188 189 190 191 192
	INIT_LIST_HEAD(&q->queue_head);
	q->last_merge = NULL;
	q->end_sector = 0;
	q->boundary_rq = NULL;

193 194 195 196 197
	if (name) {
		e = elevator_get(name);
		if (!e)
			return -EINVAL;
	}
L
Linus Torvalds 已提交
198

199 200 201 202 203 204
	if (!e && *chosen_elevator) {
		e = elevator_get(chosen_elevator);
		if (!e)
			printk(KERN_ERR "I/O scheduler %s not found\n",
							chosen_elevator);
	}
205

206 207 208 209 210 211 212 213
	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");
		}
214 215
	}

216 217
	q->elevator = elevator_alloc(q, e);
	if (!q->elevator)
L
Linus Torvalds 已提交
218 219
		return -ENOMEM;

220
	err = e->ops.elevator_init_fn(q);
221
	if (err) {
222
		kobject_put(&q->elevator->kobj);
223
		return err;
J
Jens Axboe 已提交
224
	}
L
Linus Torvalds 已提交
225

226
	return 0;
L
Linus Torvalds 已提交
227
}
228 229
EXPORT_SYMBOL(elevator_init);

J
Jens Axboe 已提交
230
void elevator_exit(struct elevator_queue *e)
L
Linus Torvalds 已提交
231
{
232
	mutex_lock(&e->sysfs_lock);
T
Tejun Heo 已提交
233 234
	if (e->type->ops.elevator_exit_fn)
		e->type->ops.elevator_exit_fn(e);
235
	mutex_unlock(&e->sysfs_lock);
L
Linus Torvalds 已提交
236

237
	kobject_put(&e->kobj);
L
Linus Torvalds 已提交
238
}
239 240
EXPORT_SYMBOL(elevator_exit);

241 242 243 244 245
static inline void __elv_rqhash_del(struct request *rq)
{
	hlist_del_init(&rq->hash);
}

246
static void elv_rqhash_del(struct request_queue *q, struct request *rq)
247 248 249 250 251
{
	if (ELV_ON_HASH(rq))
		__elv_rqhash_del(rq);
}

252
static void elv_rqhash_add(struct request_queue *q, struct request *rq)
253
{
J
Jens Axboe 已提交
254
	struct elevator_queue *e = q->elevator;
255 256 257 258 259

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

260
static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
261 262 263 264 265
{
	__elv_rqhash_del(rq);
	elv_rqhash_add(q, rq);
}

266
static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
267
{
J
Jens Axboe 已提交
268
	struct elevator_queue *e = q->elevator;
269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
	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;
}

288 289 290 291
/*
 * RB-tree support functions for inserting/lookup/removal of requests
 * in a sorted RB tree.
 */
292
void elv_rb_add(struct rb_root *root, struct request *rq)
293 294 295 296 297 298 299 300 301
{
	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);

302
		if (blk_rq_pos(rq) < blk_rq_pos(__rq))
303
			p = &(*p)->rb_left;
304
		else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
			p = &(*p)->rb_right;
	}

	rb_link_node(&rq->rb_node, parent, p);
	rb_insert_color(&rq->rb_node, root);
}
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);

329
		if (sector < blk_rq_pos(rq))
330
			n = n->rb_left;
331
		else if (sector > blk_rq_pos(rq))
332 333 334 335 336 337 338 339 340
			n = n->rb_right;
		else
			return rq;
	}

	return NULL;
}
EXPORT_SYMBOL(elv_rb_find);

341 342
/*
 * Insert rq into dispatch queue of q.  Queue lock must be held on
U
Uwe Kleine-König 已提交
343
 * entry.  rq is sort instead into the dispatch queue. To be used by
344
 * specific elevators.
345
 */
346
void elv_dispatch_sort(struct request_queue *q, struct request *rq)
347 348 349
{
	sector_t boundary;
	struct list_head *entry;
350
	int stop_flags;
351

352 353
	if (q->last_merge == rq)
		q->last_merge = NULL;
354 355 356

	elv_rqhash_del(q, rq);

357
	q->nr_sorted--;
358

J
Jens Axboe 已提交
359
	boundary = q->end_sector;
C
Christoph Hellwig 已提交
360
	stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
361 362 363
	list_for_each_prev(entry, &q->queue_head) {
		struct request *pos = list_entry_rq(entry);

364 365
		if ((rq->cmd_flags & REQ_DISCARD) !=
		    (pos->cmd_flags & REQ_DISCARD))
366
			break;
367 368
		if (rq_data_dir(rq) != rq_data_dir(pos))
			break;
369
		if (pos->cmd_flags & stop_flags)
370
			break;
371 372
		if (blk_rq_pos(rq) >= boundary) {
			if (blk_rq_pos(pos) < boundary)
373 374
				continue;
		} else {
375
			if (blk_rq_pos(pos) >= boundary)
376 377
				break;
		}
378
		if (blk_rq_pos(rq) >= blk_rq_pos(pos))
379 380 381 382 383
			break;
	}

	list_add(&rq->queuelist, entry);
}
384 385
EXPORT_SYMBOL(elv_dispatch_sort);

386
/*
387 388 389
 * 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.
390 391 392 393 394 395 396 397 398 399 400 401 402 403
 */
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);
}
404 405
EXPORT_SYMBOL(elv_dispatch_add_tail);

406
int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
L
Linus Torvalds 已提交
407
{
J
Jens Axboe 已提交
408
	struct elevator_queue *e = q->elevator;
409
	struct request *__rq;
410 411
	int ret;

412 413 414 415 416 417 418 419 420
	/*
	 * Levels of merges:
	 * 	nomerges:  No merges at all attempted
	 * 	noxmerges: Only simple one-hit cache try
	 * 	merges:	   All merge tries attempted
	 */
	if (blk_queue_nomerges(q))
		return ELEVATOR_NO_MERGE;

421 422 423
	/*
	 * First try one-hit cache.
	 */
424 425
	if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
		ret = blk_try_merge(q->last_merge, bio);
426 427 428 429 430
		if (ret != ELEVATOR_NO_MERGE) {
			*req = q->last_merge;
			return ret;
		}
	}
L
Linus Torvalds 已提交
431

432
	if (blk_queue_noxmerges(q))
433 434
		return ELEVATOR_NO_MERGE;

435 436 437 438 439 440 441 442 443
	/*
	 * 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;
	}

T
Tejun Heo 已提交
444 445
	if (e->type->ops.elevator_merge_fn)
		return e->type->ops.elevator_merge_fn(q, req, bio);
L
Linus Torvalds 已提交
446 447 448 449

	return ELEVATOR_NO_MERGE;
}

450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476
/*
 * Attempt to do an insertion back merge. Only check for the case where
 * we can append 'rq' to an existing request, so we can throw 'rq' away
 * afterwards.
 *
 * Returns true if we merged, false otherwise
 */
static bool elv_attempt_insert_merge(struct request_queue *q,
				     struct request *rq)
{
	struct request *__rq;

	if (blk_queue_nomerges(q))
		return false;

	/*
	 * First try one-hit cache.
	 */
	if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
		return true;

	if (blk_queue_noxmerges(q))
		return false;

	/*
	 * See if our hash lookup can find a potential backmerge.
	 */
477 478 479
	__rq = elv_rqhash_find(q, blk_rq_pos(rq));
	if (__rq && blk_attempt_req_merge(q, __rq, rq))
		return true;
S
Shaohua Li 已提交
480

481
	return false;
482 483
}

484
void elv_merged_request(struct request_queue *q, struct request *rq, int type)
L
Linus Torvalds 已提交
485
{
J
Jens Axboe 已提交
486
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
487

T
Tejun Heo 已提交
488 489
	if (e->type->ops.elevator_merged_fn)
		e->type->ops.elevator_merged_fn(q, rq, type);
490

491 492
	if (type == ELEVATOR_BACK_MERGE)
		elv_rqhash_reposition(q, rq);
493

494
	q->last_merge = rq;
L
Linus Torvalds 已提交
495 496
}

497
void elv_merge_requests(struct request_queue *q, struct request *rq,
L
Linus Torvalds 已提交
498 499
			     struct request *next)
{
J
Jens Axboe 已提交
500
	struct elevator_queue *e = q->elevator;
501
	const int next_sorted = next->cmd_flags & REQ_SORTED;
L
Linus Torvalds 已提交
502

T
Tejun Heo 已提交
503 504
	if (next_sorted && e->type->ops.elevator_merge_req_fn)
		e->type->ops.elevator_merge_req_fn(q, rq, next);
505

506 507
	elv_rqhash_reposition(q, rq);

508 509 510 511 512
	if (next_sorted) {
		elv_rqhash_del(q, next);
		q->nr_sorted--;
	}

513
	q->last_merge = rq;
L
Linus Torvalds 已提交
514 515
}

D
Divyesh Shah 已提交
516 517 518 519 520
void elv_bio_merged(struct request_queue *q, struct request *rq,
			struct bio *bio)
{
	struct elevator_queue *e = q->elevator;

T
Tejun Heo 已提交
521 522
	if (e->type->ops.elevator_bio_merged_fn)
		e->type->ops.elevator_bio_merged_fn(q, rq, bio);
D
Divyesh Shah 已提交
523 524
}

525
void elv_requeue_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
526 527 528 529 530
{
	/*
	 * it already went through dequeue, we need to decrement the
	 * in_flight count again
	 */
531
	if (blk_account_rq(rq)) {
532
		q->in_flight[rq_is_sync(rq)]--;
533
		if (rq->cmd_flags & REQ_SORTED)
534
			elv_deactivate_rq(q, rq);
535
	}
L
Linus Torvalds 已提交
536

537
	rq->cmd_flags &= ~REQ_STARTED;
L
Linus Torvalds 已提交
538

539
	__elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
L
Linus Torvalds 已提交
540 541
}

542
void elv_drain_elevator(struct request_queue *q)
543 544
{
	static int printed;
T
Tejun Heo 已提交
545 546 547

	lockdep_assert_held(q->queue_lock);

T
Tejun Heo 已提交
548
	while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
549
		;
T
Tejun Heo 已提交
550
	if (q->nr_sorted && printed++ < 10) {
551 552
		printk(KERN_ERR "%s: forced dispatching is broken "
		       "(nr_sorted=%u), please report this\n",
T
Tejun Heo 已提交
553
		       q->elevator->type->elevator_name, q->nr_sorted);
554 555 556
	}
}

557
void __elv_add_request(struct request_queue *q, struct request *rq, int where)
L
Linus Torvalds 已提交
558
{
559
	trace_block_rq_insert(q, rq);
560

L
Linus Torvalds 已提交
561 562
	rq->q = q;

563 564
	if (rq->cmd_flags & REQ_SOFTBARRIER) {
		/* barriers are scheduling boundary, update end_sector */
565
		if (rq->cmd_type == REQ_TYPE_FS) {
566 567 568 569
			q->end_sector = rq_end_sector(rq);
			q->boundary_rq = rq;
		}
	} else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
570 571
		    (where == ELEVATOR_INSERT_SORT ||
		     where == ELEVATOR_INSERT_SORT_MERGE))
572 573
		where = ELEVATOR_INSERT_BACK;

574
	switch (where) {
575
	case ELEVATOR_INSERT_REQUEUE:
576
	case ELEVATOR_INSERT_FRONT:
577
		rq->cmd_flags |= REQ_SOFTBARRIER;
578 579 580 581
		list_add(&rq->queuelist, &q->queue_head);
		break;

	case ELEVATOR_INSERT_BACK:
582
		rq->cmd_flags |= REQ_SOFTBARRIER;
583
		elv_drain_elevator(q);
584 585 586 587 588 589 590 591 592 593 594
		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.
		 */
595
		__blk_run_queue(q);
596 597
		break;

598 599 600 601 602 603 604 605
	case ELEVATOR_INSERT_SORT_MERGE:
		/*
		 * If we succeed in merging this request with one in the
		 * queue already, we are done - rq has now been freed,
		 * so no need to do anything further.
		 */
		if (elv_attempt_insert_merge(q, rq))
			break;
606
	case ELEVATOR_INSERT_SORT:
607
		BUG_ON(rq->cmd_type != REQ_TYPE_FS);
608
		rq->cmd_flags |= REQ_SORTED;
609
		q->nr_sorted++;
610 611 612 613 614 615
		if (rq_mergeable(rq)) {
			elv_rqhash_add(q, rq);
			if (!q->last_merge)
				q->last_merge = rq;
		}

616 617 618 619 620
		/*
		 * Some ioscheds (cfq) run q->request_fn directly, so
		 * rq cannot be accessed after calling
		 * elevator_add_req_fn.
		 */
T
Tejun Heo 已提交
621
		q->elevator->type->ops.elevator_add_req_fn(q, rq);
622 623
		break;

624 625 626 627
	case ELEVATOR_INSERT_FLUSH:
		rq->cmd_flags |= REQ_SOFTBARRIER;
		blk_insert_flush(rq);
		break;
628 629
	default:
		printk(KERN_ERR "%s: bad insertion point %d\n",
630
		       __func__, where);
631 632
		BUG();
	}
L
Linus Torvalds 已提交
633
}
634 635
EXPORT_SYMBOL(__elv_add_request);

J
Jens Axboe 已提交
636
void elv_add_request(struct request_queue *q, struct request *rq, int where)
L
Linus Torvalds 已提交
637 638 639 640
{
	unsigned long flags;

	spin_lock_irqsave(q->queue_lock, flags);
J
Jens Axboe 已提交
641
	__elv_add_request(q, rq, where);
L
Linus Torvalds 已提交
642 643
	spin_unlock_irqrestore(q->queue_lock, flags);
}
644 645
EXPORT_SYMBOL(elv_add_request);

646
struct request *elv_latter_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
647
{
J
Jens Axboe 已提交
648
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
649

T
Tejun Heo 已提交
650 651
	if (e->type->ops.elevator_latter_req_fn)
		return e->type->ops.elevator_latter_req_fn(q, rq);
L
Linus Torvalds 已提交
652 653 654
	return NULL;
}

655
struct request *elv_former_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
656
{
J
Jens Axboe 已提交
657
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
658

T
Tejun Heo 已提交
659 660
	if (e->type->ops.elevator_former_req_fn)
		return e->type->ops.elevator_former_req_fn(q, rq);
L
Linus Torvalds 已提交
661 662 663
	return NULL;
}

664 665
int elv_set_request(struct request_queue *q, struct request *rq,
		    struct bio *bio, gfp_t gfp_mask)
L
Linus Torvalds 已提交
666
{
J
Jens Axboe 已提交
667
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
668

T
Tejun Heo 已提交
669
	if (e->type->ops.elevator_set_req_fn)
670
		return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
L
Linus Torvalds 已提交
671 672 673
	return 0;
}

674
void elv_put_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
675
{
J
Jens Axboe 已提交
676
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
677

T
Tejun Heo 已提交
678 679
	if (e->type->ops.elevator_put_req_fn)
		e->type->ops.elevator_put_req_fn(rq);
L
Linus Torvalds 已提交
680 681
}

682
int elv_may_queue(struct request_queue *q, int rw)
L
Linus Torvalds 已提交
683
{
J
Jens Axboe 已提交
684
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
685

T
Tejun Heo 已提交
686 687
	if (e->type->ops.elevator_may_queue_fn)
		return e->type->ops.elevator_may_queue_fn(q, rw);
L
Linus Torvalds 已提交
688 689 690 691

	return ELV_MQUEUE_MAY;
}

692 693 694 695
void elv_abort_queue(struct request_queue *q)
{
	struct request *rq;

696 697
	blk_abort_flushes(q);

698 699 700
	while (!list_empty(&q->queue_head)) {
		rq = list_entry_rq(q->queue_head.next);
		rq->cmd_flags |= REQ_QUIET;
701
		trace_block_rq_abort(q, rq);
702 703 704 705 706
		/*
		 * Mark this request as started so we don't trigger
		 * any debug logic in the end I/O path.
		 */
		blk_start_request(rq);
707
		__blk_end_request_all(rq, -EIO);
708 709 710 711
	}
}
EXPORT_SYMBOL(elv_abort_queue);

712
void elv_completed_request(struct request_queue *q, struct request *rq)
L
Linus Torvalds 已提交
713
{
J
Jens Axboe 已提交
714
	struct elevator_queue *e = q->elevator;
L
Linus Torvalds 已提交
715 716 717 718

	/*
	 * request is released from the driver, io must be done
	 */
719
	if (blk_account_rq(rq)) {
720
		q->in_flight[rq_is_sync(rq)]--;
721
		if ((rq->cmd_flags & REQ_SORTED) &&
T
Tejun Heo 已提交
722 723
		    e->type->ops.elevator_completed_req_fn)
			e->type->ops.elevator_completed_req_fn(q, rq);
724
	}
L
Linus Torvalds 已提交
725 726
}

727 728 729 730
#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)
L
Linus Torvalds 已提交
731
{
732
	struct elv_fs_entry *entry = to_elv(attr);
J
Jens Axboe 已提交
733
	struct elevator_queue *e;
734 735 736 737 738
	ssize_t error;

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

J
Jens Axboe 已提交
739
	e = container_of(kobj, struct elevator_queue, kobj);
740
	mutex_lock(&e->sysfs_lock);
T
Tejun Heo 已提交
741
	error = e->type ? entry->show(e, page) : -ENOENT;
742 743 744
	mutex_unlock(&e->sysfs_lock);
	return error;
}
L
Linus Torvalds 已提交
745

746 747 748 749 750
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);
J
Jens Axboe 已提交
751
	struct elevator_queue *e;
752
	ssize_t error;
L
Linus Torvalds 已提交
753

754 755
	if (!entry->store)
		return -EIO;
L
Linus Torvalds 已提交
756

J
Jens Axboe 已提交
757
	e = container_of(kobj, struct elevator_queue, kobj);
758
	mutex_lock(&e->sysfs_lock);
T
Tejun Heo 已提交
759
	error = e->type ? entry->store(e, page, length) : -ENOENT;
760 761 762 763
	mutex_unlock(&e->sysfs_lock);
	return error;
}

764
static const struct sysfs_ops elv_sysfs_ops = {
765 766 767 768 769 770 771 772 773
	.show	= elv_attr_show,
	.store	= elv_attr_store,
};

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

774
int elv_register_queue(struct request_queue *q)
775
{
776
	struct elevator_queue *e = q->elevator;
777 778
	int error;

779
	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
780
	if (!error) {
T
Tejun Heo 已提交
781
		struct elv_fs_entry *attr = e->type->elevator_attrs;
782
		if (attr) {
783 784
			while (attr->attr.name) {
				if (sysfs_create_file(&e->kobj, &attr->attr))
785
					break;
786
				attr++;
787 788 789
			}
		}
		kobject_uevent(&e->kobj, KOBJ_ADD);
790
		e->registered = 1;
791 792
	}
	return error;
L
Linus Torvalds 已提交
793
}
794
EXPORT_SYMBOL(elv_register_queue);
J
Jens Axboe 已提交
795

L
Linus Torvalds 已提交
796 797
void elv_unregister_queue(struct request_queue *q)
{
798 799 800 801 802 803 804
	if (q) {
		struct elevator_queue *e = q->elevator;

		kobject_uevent(&e->kobj, KOBJ_REMOVE);
		kobject_del(&e->kobj);
		e->registered = 0;
	}
L
Linus Torvalds 已提交
805
}
806
EXPORT_SYMBOL(elv_unregister_queue);
L
Linus Torvalds 已提交
807

808
int elv_register(struct elevator_type *e)
L
Linus Torvalds 已提交
809
{
810
	char *def = "";
811

812 813 814 815 816 817 818 819 820 821 822 823 824 825 826
	/* create icq_cache if requested */
	if (e->icq_size) {
		if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
		    WARN_ON(e->icq_align < __alignof__(struct io_cq)))
			return -EINVAL;

		snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
			 "%s_io_cq", e->elevator_name);
		e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
						 e->icq_align, 0, NULL);
		if (!e->icq_cache)
			return -ENOMEM;
	}

	/* register, don't allow duplicate names */
827
	spin_lock(&elv_list_lock);
828 829 830 831 832 833
	if (elevator_find(e->elevator_name)) {
		spin_unlock(&elv_list_lock);
		if (e->icq_cache)
			kmem_cache_destroy(e->icq_cache);
		return -EBUSY;
	}
L
Linus Torvalds 已提交
834
	list_add_tail(&e->list, &elv_list);
835
	spin_unlock(&elv_list_lock);
L
Linus Torvalds 已提交
836

837
	/* print pretty message */
838 839 840
	if (!strcmp(e->elevator_name, chosen_elevator) ||
			(!*chosen_elevator &&
			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
841 842
				def = " (default)";

843 844
	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
								def);
845
	return 0;
L
Linus Torvalds 已提交
846 847 848 849 850
}
EXPORT_SYMBOL_GPL(elv_register);

void elv_unregister(struct elevator_type *e)
{
851
	/* unregister */
852
	spin_lock(&elv_list_lock);
L
Linus Torvalds 已提交
853
	list_del_init(&e->list);
854
	spin_unlock(&elv_list_lock);
855 856 857 858 859 860 861 862 863 864

	/*
	 * Destroy icq_cache if it exists.  icq's are RCU managed.  Make
	 * sure all RCU operations are complete before proceeding.
	 */
	if (e->icq_cache) {
		rcu_barrier();
		kmem_cache_destroy(e->icq_cache);
		e->icq_cache = NULL;
	}
L
Linus Torvalds 已提交
865 866 867 868 869 870 871
}
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
T
Tejun Heo 已提交
872
 * one, if the new one fails init for some reason.
L
Linus Torvalds 已提交
873
 */
874
static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
L
Linus Torvalds 已提交
875
{
876 877
	struct elevator_queue *old = q->elevator;
	bool registered = old->registered;
878
	int err;
L
Linus Torvalds 已提交
879

880 881 882 883 884 885 886
	/*
	 * Turn on BYPASS and drain all requests w/ elevator private data.
	 * Block layer doesn't call into a quiesced elevator - all requests
	 * are directly put on the dispatch list without elevator data
	 * using INSERT_BACK.  All requests have SOFTBARRIER set and no
	 * merge happens either.
	 */
887
	blk_queue_bypass_start(q);
T
Tejun Heo 已提交
888

889 890
	/* unregister and clear all auxiliary data of the old elevator */
	if (registered)
891
		elv_unregister_queue(q);
L
Linus Torvalds 已提交
892

893
	spin_lock_irq(q->queue_lock);
894
	ioc_clear_queue(q);
895 896
	spin_unlock_irq(q->queue_lock);

897 898 899 900 901 902
	/* allocate, init and register new elevator */
	err = -ENOMEM;
	q->elevator = elevator_alloc(q, new_e);
	if (!q->elevator)
		goto fail_init;

903
	err = new_e->ops.elevator_init_fn(q);
904 905 906 907 908 909 910 911 912 913 914 915 916
	if (err) {
		kobject_put(&q->elevator->kobj);
		goto fail_init;
	}

	if (registered) {
		err = elv_register_queue(q);
		if (err)
			goto fail_register;
	}

	/* done, kill the old one and finish */
	elevator_exit(old);
917
	blk_queue_bypass_end(q);
N
Nick Piggin 已提交
918

919
	blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
920

921
	return 0;
L
Linus Torvalds 已提交
922 923

fail_register:
924 925 926 927
	elevator_exit(q->elevator);
fail_init:
	/* switch failed, restore and re-register old elevator */
	q->elevator = old;
L
Linus Torvalds 已提交
928
	elv_register_queue(q);
929
	blk_queue_bypass_end(q);
N
Nick Piggin 已提交
930

931
	return err;
L
Linus Torvalds 已提交
932 933
}

934 935 936 937
/*
 * Switch this queue to the given IO scheduler.
 */
int elevator_change(struct request_queue *q, const char *name)
L
Linus Torvalds 已提交
938 939 940 941
{
	char elevator_name[ELV_NAME_MAX];
	struct elevator_type *e;

942
	if (!q->elevator)
943
		return -ENXIO;
944

945
	strlcpy(elevator_name, name, sizeof(elevator_name));
946
	e = elevator_get(strstrip(elevator_name));
L
Linus Torvalds 已提交
947 948 949 950 951
	if (!e) {
		printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
		return -EINVAL;
	}

T
Tejun Heo 已提交
952
	if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
953
		elevator_put(e);
954
		return 0;
955
	}
L
Linus Torvalds 已提交
956

957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974
	return elevator_switch(q, e);
}
EXPORT_SYMBOL(elevator_change);

ssize_t elv_iosched_store(struct request_queue *q, const char *name,
			  size_t count)
{
	int ret;

	if (!q->elevator)
		return count;

	ret = elevator_change(q, name);
	if (!ret)
		return count;

	printk(KERN_ERR "elevator: switch to %s failed\n", name);
	return ret;
L
Linus Torvalds 已提交
975 976
}

977
ssize_t elv_iosched_show(struct request_queue *q, char *name)
L
Linus Torvalds 已提交
978
{
J
Jens Axboe 已提交
979
	struct elevator_queue *e = q->elevator;
980
	struct elevator_type *elv;
981
	struct elevator_type *__e;
L
Linus Torvalds 已提交
982 983
	int len = 0;

984
	if (!q->elevator || !blk_queue_stackable(q))
985 986
		return sprintf(name, "none\n");

T
Tejun Heo 已提交
987
	elv = e->type;
988

989
	spin_lock(&elv_list_lock);
990
	list_for_each_entry(__e, &elv_list, list) {
L
Linus Torvalds 已提交
991 992 993 994 995
		if (!strcmp(elv->elevator_name, __e->elevator_name))
			len += sprintf(name+len, "[%s] ", elv->elevator_name);
		else
			len += sprintf(name+len, "%s ", __e->elevator_name);
	}
996
	spin_unlock(&elv_list_lock);
L
Linus Torvalds 已提交
997 998 999 1000 1001

	len += sprintf(len+name, "\n");
	return len;
}

1002 1003
struct request *elv_rb_former_request(struct request_queue *q,
				      struct request *rq)
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
{
	struct rb_node *rbprev = rb_prev(&rq->rb_node);

	if (rbprev)
		return rb_entry_rq(rbprev);

	return NULL;
}
EXPORT_SYMBOL(elv_rb_former_request);

1014 1015
struct request *elv_rb_latter_request(struct request_queue *q,
				      struct request *rq)
1016 1017 1018 1019 1020 1021 1022 1023 1024
{
	struct rb_node *rbnext = rb_next(&rq->rb_node);

	if (rbnext)
		return rb_entry_rq(rbnext);

	return NULL;
}
EXPORT_SYMBOL(elv_rb_latter_request);