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