xref: /openbmc/linux/block/elevator.c (revision 615c36f5)
1 /*
2  *  Block device elevator/IO-scheduler.
3  *
4  *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5  *
6  * 30042000 Jens Axboe <axboe@kernel.dk> :
7  *
8  * Split the elevator a bit so that it is possible to choose a different
9  * one or even write a new "plug in". There are three pieces:
10  * - elevator_fn, inserts a new request in the queue list
11  * - elevator_merge_fn, decides whether a new buffer can be merged with
12  *   an existing request
13  * - elevator_dequeue_fn, called when a request is taken off the active list
14  *
15  * 20082000 Dave Jones <davej@suse.de> :
16  * Removed tests for max-bomb-segments, which was breaking elvtune
17  *  when run without -bN
18  *
19  * Jens:
20  * - Rework again to work with bio instead of buffer_heads
21  * - loose bi_dev comparisons, partition handling is right now
22  * - completely modularize elevator setup and teardown
23  *
24  */
25 #include <linux/kernel.h>
26 #include <linux/fs.h>
27 #include <linux/blkdev.h>
28 #include <linux/elevator.h>
29 #include <linux/bio.h>
30 #include <linux/module.h>
31 #include <linux/slab.h>
32 #include <linux/init.h>
33 #include <linux/compiler.h>
34 #include <linux/blktrace_api.h>
35 #include <linux/hash.h>
36 #include <linux/uaccess.h>
37 
38 #include <trace/events/block.h>
39 
40 #include "blk.h"
41 
42 static DEFINE_SPINLOCK(elv_list_lock);
43 static LIST_HEAD(elv_list);
44 
45 /*
46  * Merge hash stuff.
47  */
48 static const int elv_hash_shift = 6;
49 #define ELV_HASH_BLOCK(sec)	((sec) >> 3)
50 #define ELV_HASH_FN(sec)	\
51 		(hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
52 #define ELV_HASH_ENTRIES	(1 << elv_hash_shift)
53 #define rq_hash_key(rq)		(blk_rq_pos(rq) + blk_rq_sectors(rq))
54 
55 /*
56  * Query io scheduler to see if the current process issuing bio may be
57  * merged with rq.
58  */
59 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
60 {
61 	struct request_queue *q = rq->q;
62 	struct elevator_queue *e = q->elevator;
63 
64 	if (e->ops->elevator_allow_merge_fn)
65 		return e->ops->elevator_allow_merge_fn(q, rq, bio);
66 
67 	return 1;
68 }
69 
70 /*
71  * can we safely merge with this request?
72  */
73 int elv_rq_merge_ok(struct request *rq, struct bio *bio)
74 {
75 	if (!rq_mergeable(rq))
76 		return 0;
77 
78 	/*
79 	 * Don't merge file system requests and discard requests
80 	 */
81 	if ((bio->bi_rw & REQ_DISCARD) != (rq->bio->bi_rw & REQ_DISCARD))
82 		return 0;
83 
84 	/*
85 	 * Don't merge discard requests and secure discard requests
86 	 */
87 	if ((bio->bi_rw & REQ_SECURE) != (rq->bio->bi_rw & REQ_SECURE))
88 		return 0;
89 
90 	/*
91 	 * different data direction or already started, don't merge
92 	 */
93 	if (bio_data_dir(bio) != rq_data_dir(rq))
94 		return 0;
95 
96 	/*
97 	 * must be same device and not a special request
98 	 */
99 	if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
100 		return 0;
101 
102 	/*
103 	 * only merge integrity protected bio into ditto rq
104 	 */
105 	if (bio_integrity(bio) != blk_integrity_rq(rq))
106 		return 0;
107 
108 	if (!elv_iosched_allow_merge(rq, bio))
109 		return 0;
110 
111 	return 1;
112 }
113 EXPORT_SYMBOL(elv_rq_merge_ok);
114 
115 int elv_try_merge(struct request *__rq, struct bio *bio)
116 {
117 	int ret = ELEVATOR_NO_MERGE;
118 
119 	/*
120 	 * we can merge and sequence is ok, check if it's possible
121 	 */
122 	if (elv_rq_merge_ok(__rq, bio)) {
123 		if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
124 			ret = ELEVATOR_BACK_MERGE;
125 		else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
126 			ret = ELEVATOR_FRONT_MERGE;
127 	}
128 
129 	return ret;
130 }
131 
132 static struct elevator_type *elevator_find(const char *name)
133 {
134 	struct elevator_type *e;
135 
136 	list_for_each_entry(e, &elv_list, list) {
137 		if (!strcmp(e->elevator_name, name))
138 			return e;
139 	}
140 
141 	return NULL;
142 }
143 
144 static void elevator_put(struct elevator_type *e)
145 {
146 	module_put(e->elevator_owner);
147 }
148 
149 static struct elevator_type *elevator_get(const char *name)
150 {
151 	struct elevator_type *e;
152 
153 	spin_lock(&elv_list_lock);
154 
155 	e = elevator_find(name);
156 	if (!e) {
157 		spin_unlock(&elv_list_lock);
158 		request_module("%s-iosched", name);
159 		spin_lock(&elv_list_lock);
160 		e = elevator_find(name);
161 	}
162 
163 	if (e && !try_module_get(e->elevator_owner))
164 		e = NULL;
165 
166 	spin_unlock(&elv_list_lock);
167 
168 	return e;
169 }
170 
171 static void *elevator_init_queue(struct request_queue *q,
172 				 struct elevator_queue *eq)
173 {
174 	return eq->ops->elevator_init_fn(q);
175 }
176 
177 static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
178 			   void *data)
179 {
180 	q->elevator = eq;
181 	eq->elevator_data = data;
182 }
183 
184 static char chosen_elevator[ELV_NAME_MAX];
185 
186 static int __init elevator_setup(char *str)
187 {
188 	/*
189 	 * Be backwards-compatible with previous kernels, so users
190 	 * won't get the wrong elevator.
191 	 */
192 	strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
193 	return 1;
194 }
195 
196 __setup("elevator=", elevator_setup);
197 
198 static struct kobj_type elv_ktype;
199 
200 static struct elevator_queue *elevator_alloc(struct request_queue *q,
201 				  struct elevator_type *e)
202 {
203 	struct elevator_queue *eq;
204 	int i;
205 
206 	eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
207 	if (unlikely(!eq))
208 		goto err;
209 
210 	eq->ops = &e->ops;
211 	eq->elevator_type = e;
212 	kobject_init(&eq->kobj, &elv_ktype);
213 	mutex_init(&eq->sysfs_lock);
214 
215 	eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
216 					GFP_KERNEL, q->node);
217 	if (!eq->hash)
218 		goto err;
219 
220 	for (i = 0; i < ELV_HASH_ENTRIES; i++)
221 		INIT_HLIST_HEAD(&eq->hash[i]);
222 
223 	return eq;
224 err:
225 	kfree(eq);
226 	elevator_put(e);
227 	return NULL;
228 }
229 
230 static void elevator_release(struct kobject *kobj)
231 {
232 	struct elevator_queue *e;
233 
234 	e = container_of(kobj, struct elevator_queue, kobj);
235 	elevator_put(e->elevator_type);
236 	kfree(e->hash);
237 	kfree(e);
238 }
239 
240 int elevator_init(struct request_queue *q, char *name)
241 {
242 	struct elevator_type *e = NULL;
243 	struct elevator_queue *eq;
244 	void *data;
245 
246 	if (unlikely(q->elevator))
247 		return 0;
248 
249 	INIT_LIST_HEAD(&q->queue_head);
250 	q->last_merge = NULL;
251 	q->end_sector = 0;
252 	q->boundary_rq = NULL;
253 
254 	if (name) {
255 		e = elevator_get(name);
256 		if (!e)
257 			return -EINVAL;
258 	}
259 
260 	if (!e && *chosen_elevator) {
261 		e = elevator_get(chosen_elevator);
262 		if (!e)
263 			printk(KERN_ERR "I/O scheduler %s not found\n",
264 							chosen_elevator);
265 	}
266 
267 	if (!e) {
268 		e = elevator_get(CONFIG_DEFAULT_IOSCHED);
269 		if (!e) {
270 			printk(KERN_ERR
271 				"Default I/O scheduler not found. " \
272 				"Using noop.\n");
273 			e = elevator_get("noop");
274 		}
275 	}
276 
277 	eq = elevator_alloc(q, e);
278 	if (!eq)
279 		return -ENOMEM;
280 
281 	data = elevator_init_queue(q, eq);
282 	if (!data) {
283 		kobject_put(&eq->kobj);
284 		return -ENOMEM;
285 	}
286 
287 	elevator_attach(q, eq, data);
288 	return 0;
289 }
290 EXPORT_SYMBOL(elevator_init);
291 
292 void elevator_exit(struct elevator_queue *e)
293 {
294 	mutex_lock(&e->sysfs_lock);
295 	if (e->ops->elevator_exit_fn)
296 		e->ops->elevator_exit_fn(e);
297 	e->ops = NULL;
298 	mutex_unlock(&e->sysfs_lock);
299 
300 	kobject_put(&e->kobj);
301 }
302 EXPORT_SYMBOL(elevator_exit);
303 
304 static inline void __elv_rqhash_del(struct request *rq)
305 {
306 	hlist_del_init(&rq->hash);
307 }
308 
309 static void elv_rqhash_del(struct request_queue *q, struct request *rq)
310 {
311 	if (ELV_ON_HASH(rq))
312 		__elv_rqhash_del(rq);
313 }
314 
315 static void elv_rqhash_add(struct request_queue *q, struct request *rq)
316 {
317 	struct elevator_queue *e = q->elevator;
318 
319 	BUG_ON(ELV_ON_HASH(rq));
320 	hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
321 }
322 
323 static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
324 {
325 	__elv_rqhash_del(rq);
326 	elv_rqhash_add(q, rq);
327 }
328 
329 static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
330 {
331 	struct elevator_queue *e = q->elevator;
332 	struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
333 	struct hlist_node *entry, *next;
334 	struct request *rq;
335 
336 	hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
337 		BUG_ON(!ELV_ON_HASH(rq));
338 
339 		if (unlikely(!rq_mergeable(rq))) {
340 			__elv_rqhash_del(rq);
341 			continue;
342 		}
343 
344 		if (rq_hash_key(rq) == offset)
345 			return rq;
346 	}
347 
348 	return NULL;
349 }
350 
351 /*
352  * RB-tree support functions for inserting/lookup/removal of requests
353  * in a sorted RB tree.
354  */
355 void elv_rb_add(struct rb_root *root, struct request *rq)
356 {
357 	struct rb_node **p = &root->rb_node;
358 	struct rb_node *parent = NULL;
359 	struct request *__rq;
360 
361 	while (*p) {
362 		parent = *p;
363 		__rq = rb_entry(parent, struct request, rb_node);
364 
365 		if (blk_rq_pos(rq) < blk_rq_pos(__rq))
366 			p = &(*p)->rb_left;
367 		else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
368 			p = &(*p)->rb_right;
369 	}
370 
371 	rb_link_node(&rq->rb_node, parent, p);
372 	rb_insert_color(&rq->rb_node, root);
373 }
374 EXPORT_SYMBOL(elv_rb_add);
375 
376 void elv_rb_del(struct rb_root *root, struct request *rq)
377 {
378 	BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
379 	rb_erase(&rq->rb_node, root);
380 	RB_CLEAR_NODE(&rq->rb_node);
381 }
382 EXPORT_SYMBOL(elv_rb_del);
383 
384 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
385 {
386 	struct rb_node *n = root->rb_node;
387 	struct request *rq;
388 
389 	while (n) {
390 		rq = rb_entry(n, struct request, rb_node);
391 
392 		if (sector < blk_rq_pos(rq))
393 			n = n->rb_left;
394 		else if (sector > blk_rq_pos(rq))
395 			n = n->rb_right;
396 		else
397 			return rq;
398 	}
399 
400 	return NULL;
401 }
402 EXPORT_SYMBOL(elv_rb_find);
403 
404 /*
405  * Insert rq into dispatch queue of q.  Queue lock must be held on
406  * entry.  rq is sort instead into the dispatch queue. To be used by
407  * specific elevators.
408  */
409 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
410 {
411 	sector_t boundary;
412 	struct list_head *entry;
413 	int stop_flags;
414 
415 	if (q->last_merge == rq)
416 		q->last_merge = NULL;
417 
418 	elv_rqhash_del(q, rq);
419 
420 	q->nr_sorted--;
421 
422 	boundary = q->end_sector;
423 	stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
424 	list_for_each_prev(entry, &q->queue_head) {
425 		struct request *pos = list_entry_rq(entry);
426 
427 		if ((rq->cmd_flags & REQ_DISCARD) !=
428 		    (pos->cmd_flags & REQ_DISCARD))
429 			break;
430 		if (rq_data_dir(rq) != rq_data_dir(pos))
431 			break;
432 		if (pos->cmd_flags & stop_flags)
433 			break;
434 		if (blk_rq_pos(rq) >= boundary) {
435 			if (blk_rq_pos(pos) < boundary)
436 				continue;
437 		} else {
438 			if (blk_rq_pos(pos) >= boundary)
439 				break;
440 		}
441 		if (blk_rq_pos(rq) >= blk_rq_pos(pos))
442 			break;
443 	}
444 
445 	list_add(&rq->queuelist, entry);
446 }
447 EXPORT_SYMBOL(elv_dispatch_sort);
448 
449 /*
450  * Insert rq into dispatch queue of q.  Queue lock must be held on
451  * entry.  rq is added to the back of the dispatch queue. To be used by
452  * specific elevators.
453  */
454 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
455 {
456 	if (q->last_merge == rq)
457 		q->last_merge = NULL;
458 
459 	elv_rqhash_del(q, rq);
460 
461 	q->nr_sorted--;
462 
463 	q->end_sector = rq_end_sector(rq);
464 	q->boundary_rq = rq;
465 	list_add_tail(&rq->queuelist, &q->queue_head);
466 }
467 EXPORT_SYMBOL(elv_dispatch_add_tail);
468 
469 int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
470 {
471 	struct elevator_queue *e = q->elevator;
472 	struct request *__rq;
473 	int ret;
474 
475 	/*
476 	 * Levels of merges:
477 	 * 	nomerges:  No merges at all attempted
478 	 * 	noxmerges: Only simple one-hit cache try
479 	 * 	merges:	   All merge tries attempted
480 	 */
481 	if (blk_queue_nomerges(q))
482 		return ELEVATOR_NO_MERGE;
483 
484 	/*
485 	 * First try one-hit cache.
486 	 */
487 	if (q->last_merge) {
488 		ret = elv_try_merge(q->last_merge, bio);
489 		if (ret != ELEVATOR_NO_MERGE) {
490 			*req = q->last_merge;
491 			return ret;
492 		}
493 	}
494 
495 	if (blk_queue_noxmerges(q))
496 		return ELEVATOR_NO_MERGE;
497 
498 	/*
499 	 * See if our hash lookup can find a potential backmerge.
500 	 */
501 	__rq = elv_rqhash_find(q, bio->bi_sector);
502 	if (__rq && elv_rq_merge_ok(__rq, bio)) {
503 		*req = __rq;
504 		return ELEVATOR_BACK_MERGE;
505 	}
506 
507 	if (e->ops->elevator_merge_fn)
508 		return e->ops->elevator_merge_fn(q, req, bio);
509 
510 	return ELEVATOR_NO_MERGE;
511 }
512 
513 /*
514  * Attempt to do an insertion back merge. Only check for the case where
515  * we can append 'rq' to an existing request, so we can throw 'rq' away
516  * afterwards.
517  *
518  * Returns true if we merged, false otherwise
519  */
520 static bool elv_attempt_insert_merge(struct request_queue *q,
521 				     struct request *rq)
522 {
523 	struct request *__rq;
524 
525 	if (blk_queue_nomerges(q))
526 		return false;
527 
528 	/*
529 	 * First try one-hit cache.
530 	 */
531 	if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
532 		return true;
533 
534 	if (blk_queue_noxmerges(q))
535 		return false;
536 
537 	/*
538 	 * See if our hash lookup can find a potential backmerge.
539 	 */
540 	__rq = elv_rqhash_find(q, blk_rq_pos(rq));
541 	if (__rq && blk_attempt_req_merge(q, __rq, rq))
542 		return true;
543 
544 	return false;
545 }
546 
547 void elv_merged_request(struct request_queue *q, struct request *rq, int type)
548 {
549 	struct elevator_queue *e = q->elevator;
550 
551 	if (e->ops->elevator_merged_fn)
552 		e->ops->elevator_merged_fn(q, rq, type);
553 
554 	if (type == ELEVATOR_BACK_MERGE)
555 		elv_rqhash_reposition(q, rq);
556 
557 	q->last_merge = rq;
558 }
559 
560 void elv_merge_requests(struct request_queue *q, struct request *rq,
561 			     struct request *next)
562 {
563 	struct elevator_queue *e = q->elevator;
564 	const int next_sorted = next->cmd_flags & REQ_SORTED;
565 
566 	if (next_sorted && e->ops->elevator_merge_req_fn)
567 		e->ops->elevator_merge_req_fn(q, rq, next);
568 
569 	elv_rqhash_reposition(q, rq);
570 
571 	if (next_sorted) {
572 		elv_rqhash_del(q, next);
573 		q->nr_sorted--;
574 	}
575 
576 	q->last_merge = rq;
577 }
578 
579 void elv_bio_merged(struct request_queue *q, struct request *rq,
580 			struct bio *bio)
581 {
582 	struct elevator_queue *e = q->elevator;
583 
584 	if (e->ops->elevator_bio_merged_fn)
585 		e->ops->elevator_bio_merged_fn(q, rq, bio);
586 }
587 
588 void elv_requeue_request(struct request_queue *q, struct request *rq)
589 {
590 	/*
591 	 * it already went through dequeue, we need to decrement the
592 	 * in_flight count again
593 	 */
594 	if (blk_account_rq(rq)) {
595 		q->in_flight[rq_is_sync(rq)]--;
596 		if (rq->cmd_flags & REQ_SORTED)
597 			elv_deactivate_rq(q, rq);
598 	}
599 
600 	rq->cmd_flags &= ~REQ_STARTED;
601 
602 	__elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
603 }
604 
605 void elv_drain_elevator(struct request_queue *q)
606 {
607 	static int printed;
608 
609 	lockdep_assert_held(q->queue_lock);
610 
611 	while (q->elevator->ops->elevator_dispatch_fn(q, 1))
612 		;
613 	if (q->nr_sorted && printed++ < 10) {
614 		printk(KERN_ERR "%s: forced dispatching is broken "
615 		       "(nr_sorted=%u), please report this\n",
616 		       q->elevator->elevator_type->elevator_name, q->nr_sorted);
617 	}
618 }
619 
620 void elv_quiesce_start(struct request_queue *q)
621 {
622 	if (!q->elevator)
623 		return;
624 
625 	spin_lock_irq(q->queue_lock);
626 	queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
627 	spin_unlock_irq(q->queue_lock);
628 
629 	blk_drain_queue(q, false);
630 }
631 
632 void elv_quiesce_end(struct request_queue *q)
633 {
634 	spin_lock_irq(q->queue_lock);
635 	queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
636 	spin_unlock_irq(q->queue_lock);
637 }
638 
639 void __elv_add_request(struct request_queue *q, struct request *rq, int where)
640 {
641 	trace_block_rq_insert(q, rq);
642 
643 	rq->q = q;
644 
645 	if (rq->cmd_flags & REQ_SOFTBARRIER) {
646 		/* barriers are scheduling boundary, update end_sector */
647 		if (rq->cmd_type == REQ_TYPE_FS ||
648 		    (rq->cmd_flags & REQ_DISCARD)) {
649 			q->end_sector = rq_end_sector(rq);
650 			q->boundary_rq = rq;
651 		}
652 	} else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
653 		    (where == ELEVATOR_INSERT_SORT ||
654 		     where == ELEVATOR_INSERT_SORT_MERGE))
655 		where = ELEVATOR_INSERT_BACK;
656 
657 	switch (where) {
658 	case ELEVATOR_INSERT_REQUEUE:
659 	case ELEVATOR_INSERT_FRONT:
660 		rq->cmd_flags |= REQ_SOFTBARRIER;
661 		list_add(&rq->queuelist, &q->queue_head);
662 		break;
663 
664 	case ELEVATOR_INSERT_BACK:
665 		rq->cmd_flags |= REQ_SOFTBARRIER;
666 		elv_drain_elevator(q);
667 		list_add_tail(&rq->queuelist, &q->queue_head);
668 		/*
669 		 * We kick the queue here for the following reasons.
670 		 * - The elevator might have returned NULL previously
671 		 *   to delay requests and returned them now.  As the
672 		 *   queue wasn't empty before this request, ll_rw_blk
673 		 *   won't run the queue on return, resulting in hang.
674 		 * - Usually, back inserted requests won't be merged
675 		 *   with anything.  There's no point in delaying queue
676 		 *   processing.
677 		 */
678 		__blk_run_queue(q);
679 		break;
680 
681 	case ELEVATOR_INSERT_SORT_MERGE:
682 		/*
683 		 * If we succeed in merging this request with one in the
684 		 * queue already, we are done - rq has now been freed,
685 		 * so no need to do anything further.
686 		 */
687 		if (elv_attempt_insert_merge(q, rq))
688 			break;
689 	case ELEVATOR_INSERT_SORT:
690 		BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
691 		       !(rq->cmd_flags & REQ_DISCARD));
692 		rq->cmd_flags |= REQ_SORTED;
693 		q->nr_sorted++;
694 		if (rq_mergeable(rq)) {
695 			elv_rqhash_add(q, rq);
696 			if (!q->last_merge)
697 				q->last_merge = rq;
698 		}
699 
700 		/*
701 		 * Some ioscheds (cfq) run q->request_fn directly, so
702 		 * rq cannot be accessed after calling
703 		 * elevator_add_req_fn.
704 		 */
705 		q->elevator->ops->elevator_add_req_fn(q, rq);
706 		break;
707 
708 	case ELEVATOR_INSERT_FLUSH:
709 		rq->cmd_flags |= REQ_SOFTBARRIER;
710 		blk_insert_flush(rq);
711 		break;
712 	default:
713 		printk(KERN_ERR "%s: bad insertion point %d\n",
714 		       __func__, where);
715 		BUG();
716 	}
717 }
718 EXPORT_SYMBOL(__elv_add_request);
719 
720 void elv_add_request(struct request_queue *q, struct request *rq, int where)
721 {
722 	unsigned long flags;
723 
724 	spin_lock_irqsave(q->queue_lock, flags);
725 	__elv_add_request(q, rq, where);
726 	spin_unlock_irqrestore(q->queue_lock, flags);
727 }
728 EXPORT_SYMBOL(elv_add_request);
729 
730 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
731 {
732 	struct elevator_queue *e = q->elevator;
733 
734 	if (e->ops->elevator_latter_req_fn)
735 		return e->ops->elevator_latter_req_fn(q, rq);
736 	return NULL;
737 }
738 
739 struct request *elv_former_request(struct request_queue *q, struct request *rq)
740 {
741 	struct elevator_queue *e = q->elevator;
742 
743 	if (e->ops->elevator_former_req_fn)
744 		return e->ops->elevator_former_req_fn(q, rq);
745 	return NULL;
746 }
747 
748 int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
749 {
750 	struct elevator_queue *e = q->elevator;
751 
752 	if (e->ops->elevator_set_req_fn)
753 		return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
754 
755 	rq->elevator_private[0] = NULL;
756 	return 0;
757 }
758 
759 void elv_put_request(struct request_queue *q, struct request *rq)
760 {
761 	struct elevator_queue *e = q->elevator;
762 
763 	if (e->ops->elevator_put_req_fn)
764 		e->ops->elevator_put_req_fn(rq);
765 }
766 
767 int elv_may_queue(struct request_queue *q, int rw)
768 {
769 	struct elevator_queue *e = q->elevator;
770 
771 	if (e->ops->elevator_may_queue_fn)
772 		return e->ops->elevator_may_queue_fn(q, rw);
773 
774 	return ELV_MQUEUE_MAY;
775 }
776 
777 void elv_abort_queue(struct request_queue *q)
778 {
779 	struct request *rq;
780 
781 	blk_abort_flushes(q);
782 
783 	while (!list_empty(&q->queue_head)) {
784 		rq = list_entry_rq(q->queue_head.next);
785 		rq->cmd_flags |= REQ_QUIET;
786 		trace_block_rq_abort(q, rq);
787 		/*
788 		 * Mark this request as started so we don't trigger
789 		 * any debug logic in the end I/O path.
790 		 */
791 		blk_start_request(rq);
792 		__blk_end_request_all(rq, -EIO);
793 	}
794 }
795 EXPORT_SYMBOL(elv_abort_queue);
796 
797 void elv_completed_request(struct request_queue *q, struct request *rq)
798 {
799 	struct elevator_queue *e = q->elevator;
800 
801 	/*
802 	 * request is released from the driver, io must be done
803 	 */
804 	if (blk_account_rq(rq)) {
805 		q->in_flight[rq_is_sync(rq)]--;
806 		if ((rq->cmd_flags & REQ_SORTED) &&
807 		    e->ops->elevator_completed_req_fn)
808 			e->ops->elevator_completed_req_fn(q, rq);
809 	}
810 }
811 
812 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
813 
814 static ssize_t
815 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
816 {
817 	struct elv_fs_entry *entry = to_elv(attr);
818 	struct elevator_queue *e;
819 	ssize_t error;
820 
821 	if (!entry->show)
822 		return -EIO;
823 
824 	e = container_of(kobj, struct elevator_queue, kobj);
825 	mutex_lock(&e->sysfs_lock);
826 	error = e->ops ? entry->show(e, page) : -ENOENT;
827 	mutex_unlock(&e->sysfs_lock);
828 	return error;
829 }
830 
831 static ssize_t
832 elv_attr_store(struct kobject *kobj, struct attribute *attr,
833 	       const char *page, size_t length)
834 {
835 	struct elv_fs_entry *entry = to_elv(attr);
836 	struct elevator_queue *e;
837 	ssize_t error;
838 
839 	if (!entry->store)
840 		return -EIO;
841 
842 	e = container_of(kobj, struct elevator_queue, kobj);
843 	mutex_lock(&e->sysfs_lock);
844 	error = e->ops ? entry->store(e, page, length) : -ENOENT;
845 	mutex_unlock(&e->sysfs_lock);
846 	return error;
847 }
848 
849 static const struct sysfs_ops elv_sysfs_ops = {
850 	.show	= elv_attr_show,
851 	.store	= elv_attr_store,
852 };
853 
854 static struct kobj_type elv_ktype = {
855 	.sysfs_ops	= &elv_sysfs_ops,
856 	.release	= elevator_release,
857 };
858 
859 int elv_register_queue(struct request_queue *q)
860 {
861 	struct elevator_queue *e = q->elevator;
862 	int error;
863 
864 	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
865 	if (!error) {
866 		struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
867 		if (attr) {
868 			while (attr->attr.name) {
869 				if (sysfs_create_file(&e->kobj, &attr->attr))
870 					break;
871 				attr++;
872 			}
873 		}
874 		kobject_uevent(&e->kobj, KOBJ_ADD);
875 		e->registered = 1;
876 	}
877 	return error;
878 }
879 EXPORT_SYMBOL(elv_register_queue);
880 
881 static void __elv_unregister_queue(struct elevator_queue *e)
882 {
883 	kobject_uevent(&e->kobj, KOBJ_REMOVE);
884 	kobject_del(&e->kobj);
885 	e->registered = 0;
886 }
887 
888 void elv_unregister_queue(struct request_queue *q)
889 {
890 	if (q)
891 		__elv_unregister_queue(q->elevator);
892 }
893 EXPORT_SYMBOL(elv_unregister_queue);
894 
895 void elv_register(struct elevator_type *e)
896 {
897 	char *def = "";
898 
899 	spin_lock(&elv_list_lock);
900 	BUG_ON(elevator_find(e->elevator_name));
901 	list_add_tail(&e->list, &elv_list);
902 	spin_unlock(&elv_list_lock);
903 
904 	if (!strcmp(e->elevator_name, chosen_elevator) ||
905 			(!*chosen_elevator &&
906 			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
907 				def = " (default)";
908 
909 	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
910 								def);
911 }
912 EXPORT_SYMBOL_GPL(elv_register);
913 
914 void elv_unregister(struct elevator_type *e)
915 {
916 	struct task_struct *g, *p;
917 
918 	/*
919 	 * Iterate every thread in the process to remove the io contexts.
920 	 */
921 	if (e->ops.trim) {
922 		read_lock(&tasklist_lock);
923 		do_each_thread(g, p) {
924 			task_lock(p);
925 			if (p->io_context)
926 				e->ops.trim(p->io_context);
927 			task_unlock(p);
928 		} while_each_thread(g, p);
929 		read_unlock(&tasklist_lock);
930 	}
931 
932 	spin_lock(&elv_list_lock);
933 	list_del_init(&e->list);
934 	spin_unlock(&elv_list_lock);
935 }
936 EXPORT_SYMBOL_GPL(elv_unregister);
937 
938 /*
939  * switch to new_e io scheduler. be careful not to introduce deadlocks -
940  * we don't free the old io scheduler, before we have allocated what we
941  * need for the new one. this way we have a chance of going back to the old
942  * one, if the new one fails init for some reason.
943  */
944 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
945 {
946 	struct elevator_queue *old_elevator, *e;
947 	void *data;
948 	int err;
949 
950 	/*
951 	 * Allocate new elevator
952 	 */
953 	e = elevator_alloc(q, new_e);
954 	if (!e)
955 		return -ENOMEM;
956 
957 	data = elevator_init_queue(q, e);
958 	if (!data) {
959 		kobject_put(&e->kobj);
960 		return -ENOMEM;
961 	}
962 
963 	/*
964 	 * Turn on BYPASS and drain all requests w/ elevator private data
965 	 */
966 	elv_quiesce_start(q);
967 
968 	/*
969 	 * Remember old elevator.
970 	 */
971 	old_elevator = q->elevator;
972 
973 	/*
974 	 * attach and start new elevator
975 	 */
976 	spin_lock_irq(q->queue_lock);
977 	elevator_attach(q, e, data);
978 	spin_unlock_irq(q->queue_lock);
979 
980 	if (old_elevator->registered) {
981 		__elv_unregister_queue(old_elevator);
982 
983 		err = elv_register_queue(q);
984 		if (err)
985 			goto fail_register;
986 	}
987 
988 	/*
989 	 * finally exit old elevator and turn off BYPASS.
990 	 */
991 	elevator_exit(old_elevator);
992 	elv_quiesce_end(q);
993 
994 	blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
995 
996 	return 0;
997 
998 fail_register:
999 	/*
1000 	 * switch failed, exit the new io scheduler and reattach the old
1001 	 * one again (along with re-adding the sysfs dir)
1002 	 */
1003 	elevator_exit(e);
1004 	q->elevator = old_elevator;
1005 	elv_register_queue(q);
1006 	elv_quiesce_end(q);
1007 
1008 	return err;
1009 }
1010 
1011 /*
1012  * Switch this queue to the given IO scheduler.
1013  */
1014 int elevator_change(struct request_queue *q, const char *name)
1015 {
1016 	char elevator_name[ELV_NAME_MAX];
1017 	struct elevator_type *e;
1018 
1019 	if (!q->elevator)
1020 		return -ENXIO;
1021 
1022 	strlcpy(elevator_name, name, sizeof(elevator_name));
1023 	e = elevator_get(strstrip(elevator_name));
1024 	if (!e) {
1025 		printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1026 		return -EINVAL;
1027 	}
1028 
1029 	if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1030 		elevator_put(e);
1031 		return 0;
1032 	}
1033 
1034 	return elevator_switch(q, e);
1035 }
1036 EXPORT_SYMBOL(elevator_change);
1037 
1038 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1039 			  size_t count)
1040 {
1041 	int ret;
1042 
1043 	if (!q->elevator)
1044 		return count;
1045 
1046 	ret = elevator_change(q, name);
1047 	if (!ret)
1048 		return count;
1049 
1050 	printk(KERN_ERR "elevator: switch to %s failed\n", name);
1051 	return ret;
1052 }
1053 
1054 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1055 {
1056 	struct elevator_queue *e = q->elevator;
1057 	struct elevator_type *elv;
1058 	struct elevator_type *__e;
1059 	int len = 0;
1060 
1061 	if (!q->elevator || !blk_queue_stackable(q))
1062 		return sprintf(name, "none\n");
1063 
1064 	elv = e->elevator_type;
1065 
1066 	spin_lock(&elv_list_lock);
1067 	list_for_each_entry(__e, &elv_list, list) {
1068 		if (!strcmp(elv->elevator_name, __e->elevator_name))
1069 			len += sprintf(name+len, "[%s] ", elv->elevator_name);
1070 		else
1071 			len += sprintf(name+len, "%s ", __e->elevator_name);
1072 	}
1073 	spin_unlock(&elv_list_lock);
1074 
1075 	len += sprintf(len+name, "\n");
1076 	return len;
1077 }
1078 
1079 struct request *elv_rb_former_request(struct request_queue *q,
1080 				      struct request *rq)
1081 {
1082 	struct rb_node *rbprev = rb_prev(&rq->rb_node);
1083 
1084 	if (rbprev)
1085 		return rb_entry_rq(rbprev);
1086 
1087 	return NULL;
1088 }
1089 EXPORT_SYMBOL(elv_rb_former_request);
1090 
1091 struct request *elv_rb_latter_request(struct request_queue *q,
1092 				      struct request *rq)
1093 {
1094 	struct rb_node *rbnext = rb_next(&rq->rb_node);
1095 
1096 	if (rbnext)
1097 		return rb_entry_rq(rbnext);
1098 
1099 	return NULL;
1100 }
1101 EXPORT_SYMBOL(elv_rb_latter_request);
1102