blob: a3b64bc71d8821fcdf4fa15dbbd0704ac34b843a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Block device elevator/IO-scheduler.
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 *
Jens Axboe0fe23472006-09-04 15:41:16 +02006 * 30042000 Jens Axboe <axboe@kernel.dk> :
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/module.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/compiler.h>
Tejun Heocb98fc82005-10-28 08:29:39 +020034#include <linux/delay.h>
Jens Axboe2056a782006-03-23 20:00:26 +010035#include <linux/blktrace_api.h>
Jens Axboe98170642006-07-28 09:23:08 +020036#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020037#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Li Zefan55782132009-06-09 13:43:05 +080039#include <trace/events/block.h>
40
Jens Axboe242f9dc2008-09-14 05:55:09 -070041#include "blk.h"
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static DEFINE_SPINLOCK(elv_list_lock);
44static LIST_HEAD(elv_list);
45
46/*
Jens Axboe98170642006-07-28 09:23:08 +020047 * Merge hash stuff.
48 */
49static const int elv_hash_shift = 6;
50#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010051#define ELV_HASH_FN(sec) \
52 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020053#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
Tejun Heo83096eb2009-05-07 22:24:39 +090054#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
Jens Axboe98170642006-07-28 09:23:08 +020055
56/*
Jens Axboeda775262006-12-20 11:04:12 +010057 * Query io scheduler to see if the current process issuing bio may be
58 * merged with rq.
59 */
60static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
61{
Jens Axboe165125e2007-07-24 09:28:11 +020062 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010063 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010064
65 if (e->ops->elevator_allow_merge_fn)
66 return e->ops->elevator_allow_merge_fn(q, rq, bio);
67
68 return 1;
69}
70
71/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * can we safely merge with this request?
73 */
Adrian Bunk72ed0bf2008-04-29 09:49:05 +020074int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
76 if (!rq_mergeable(rq))
77 return 0;
78
79 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +010080 * Don't merge file system requests and discard requests
81 */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +020082 if ((bio->bi_rw & REQ_DISCARD) != (rq->bio->bi_rw & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +010083 return 0;
84
85 /*
Adrian Hunter8d57a982010-08-11 14:17:49 -070086 * Don't merge discard requests and secure discard requests
87 */
88 if ((bio->bi_rw & REQ_SECURE) != (rq->bio->bi_rw & REQ_SECURE))
89 return 0;
90
91 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * different data direction or already started, don't merge
93 */
94 if (bio_data_dir(bio) != rq_data_dir(rq))
95 return 0;
96
97 /*
Jens Axboeda775262006-12-20 11:04:12 +010098 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 */
Jens Axboebb4067e2006-12-21 21:20:01 +0100100 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +0100101 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200103 /*
104 * only merge integrity protected bio into ditto rq
105 */
106 if (bio_integrity(bio) != blk_integrity_rq(rq))
107 return 0;
108
Jens Axboeda775262006-12-20 11:04:12 +0100109 if (!elv_iosched_allow_merge(rq, bio))
110 return 0;
111
112 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114EXPORT_SYMBOL(elv_rq_merge_ok);
115
Jens Axboe73c10102011-03-08 13:19:51 +0100116int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
118 int ret = ELEVATOR_NO_MERGE;
119
120 /*
121 * we can merge and sequence is ok, check if it's possible
122 */
123 if (elv_rq_merge_ok(__rq, bio)) {
Tejun Heo83096eb2009-05-07 22:24:39 +0900124 if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 ret = ELEVATOR_BACK_MERGE;
Tejun Heo83096eb2009-05-07 22:24:39 +0900126 else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 ret = ELEVATOR_FRONT_MERGE;
128 }
129
130 return ret;
131}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133static struct elevator_type *elevator_find(const char *name)
134{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200135 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200137 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +0200138 if (!strcmp(e->elevator_name, name))
139 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Vasily Tarasova22b1692006-10-11 09:24:27 +0200142 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
145static void elevator_put(struct elevator_type *e)
146{
147 module_put(e->elevator_owner);
148}
149
150static struct elevator_type *elevator_get(const char *name)
151{
Tejun Heo2824bc932005-10-20 10:56:41 +0200152 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200154 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200155
156 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100157 if (!e) {
Jens Axboee1640942008-02-19 10:20:37 +0100158 spin_unlock(&elv_list_lock);
Kees Cook490b94b2011-05-05 18:02:12 -0600159 request_module("%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100160 spin_lock(&elv_list_lock);
161 e = elevator_find(name);
162 }
163
Tejun Heo2824bc932005-10-20 10:56:41 +0200164 if (e && !try_module_get(e->elevator_owner))
165 e = NULL;
166
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200167 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
169 return e;
170}
171
Jens Axboe165125e2007-07-24 09:28:11 +0200172static void *elevator_init_queue(struct request_queue *q,
173 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
Jens Axboebb37b942006-12-01 10:42:33 +0100175 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Jens Axboe165125e2007-07-24 09:28:11 +0200178static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200179 void *data)
180{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200182 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185static char chosen_elevator[16];
186
Nate Diller5f003972006-01-24 10:07:58 +0100187static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100189 /*
190 * Be backwards-compatible with previous kernels, so users
191 * won't get the wrong elevator.
192 */
Jens Axboe492af632009-10-03 09:37:51 +0200193 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800194 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195}
196
197__setup("elevator=", elevator_setup);
198
Al Viro3d1ab402006-03-18 18:35:43 -0500199static struct kobj_type elv_ktype;
200
Jens Axboeb374d182008-10-31 10:05:07 +0100201static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200202 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500203{
Jens Axboeb374d182008-10-31 10:05:07 +0100204 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200205 int i;
206
Jens Axboeb374d182008-10-31 10:05:07 +0100207 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200208 if (unlikely(!eq))
209 goto err;
210
Jens Axboe98170642006-07-28 09:23:08 +0200211 eq->ops = &e->ops;
212 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700213 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200214 mutex_init(&eq->sysfs_lock);
215
Jens Axboeb5deef92006-07-19 23:39:40 +0200216 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
217 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200218 if (!eq->hash)
219 goto err;
220
221 for (i = 0; i < ELV_HASH_ENTRIES; i++)
222 INIT_HLIST_HEAD(&eq->hash[i]);
223
Al Viro3d1ab402006-03-18 18:35:43 -0500224 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200225err:
226 kfree(eq);
227 elevator_put(e);
228 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500229}
230
231static void elevator_release(struct kobject *kobj)
232{
Jens Axboeb374d182008-10-31 10:05:07 +0100233 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200234
Jens Axboeb374d182008-10-31 10:05:07 +0100235 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500236 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200237 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500238 kfree(e);
239}
240
Jens Axboe165125e2007-07-24 09:28:11 +0200241int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
243 struct elevator_type *e = NULL;
244 struct elevator_queue *eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200245 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400247 if (unlikely(q->elevator))
248 return 0;
249
Tejun Heocb98fc82005-10-28 08:29:39 +0200250 INIT_LIST_HEAD(&q->queue_head);
251 q->last_merge = NULL;
252 q->end_sector = 0;
253 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200254
Jens Axboe4eb166d2008-02-01 00:37:27 +0100255 if (name) {
256 e = elevator_get(name);
257 if (!e)
258 return -EINVAL;
259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Jens Axboe4eb166d2008-02-01 00:37:27 +0100261 if (!e && *chosen_elevator) {
262 e = elevator_get(chosen_elevator);
263 if (!e)
264 printk(KERN_ERR "I/O scheduler %s not found\n",
265 chosen_elevator);
266 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100267
Jens Axboe4eb166d2008-02-01 00:37:27 +0100268 if (!e) {
269 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
270 if (!e) {
271 printk(KERN_ERR
272 "Default I/O scheduler not found. " \
273 "Using noop.\n");
274 e = elevator_get("noop");
275 }
Nate Diller5f003972006-01-24 10:07:58 +0100276 }
277
Jens Axboeb5deef92006-07-19 23:39:40 +0200278 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500279 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Jens Axboebc1c1162006-06-08 08:49:06 +0200282 data = elevator_init_queue(q, eq);
283 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500284 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200285 return -ENOMEM;
286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Jens Axboebc1c1162006-06-08 08:49:06 +0200288 elevator_attach(q, eq, data);
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400289 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
Jens Axboe2e662b62006-07-13 11:55:04 +0200291EXPORT_SYMBOL(elevator_init);
292
Jens Axboeb374d182008-10-31 10:05:07 +0100293void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
Al Viro3d1ab402006-03-18 18:35:43 -0500295 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (e->ops->elevator_exit_fn)
297 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500298 e->ops = NULL;
299 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Al Viro3d1ab402006-03-18 18:35:43 -0500301 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
Jens Axboe2e662b62006-07-13 11:55:04 +0200303EXPORT_SYMBOL(elevator_exit);
304
Jens Axboe98170642006-07-28 09:23:08 +0200305static inline void __elv_rqhash_del(struct request *rq)
306{
307 hlist_del_init(&rq->hash);
308}
309
Jens Axboe165125e2007-07-24 09:28:11 +0200310static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200311{
312 if (ELV_ON_HASH(rq))
313 __elv_rqhash_del(rq);
314}
315
Jens Axboe165125e2007-07-24 09:28:11 +0200316static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200317{
Jens Axboeb374d182008-10-31 10:05:07 +0100318 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200319
320 BUG_ON(ELV_ON_HASH(rq));
321 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
322}
323
Jens Axboe165125e2007-07-24 09:28:11 +0200324static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200325{
326 __elv_rqhash_del(rq);
327 elv_rqhash_add(q, rq);
328}
329
Jens Axboe165125e2007-07-24 09:28:11 +0200330static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200331{
Jens Axboeb374d182008-10-31 10:05:07 +0100332 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200333 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
334 struct hlist_node *entry, *next;
335 struct request *rq;
336
337 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
338 BUG_ON(!ELV_ON_HASH(rq));
339
340 if (unlikely(!rq_mergeable(rq))) {
341 __elv_rqhash_del(rq);
342 continue;
343 }
344
345 if (rq_hash_key(rq) == offset)
346 return rq;
347 }
348
349 return NULL;
350}
351
Tejun Heo8922e162005-10-20 16:23:44 +0200352/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200353 * RB-tree support functions for inserting/lookup/removal of requests
354 * in a sorted RB tree.
355 */
Jeff Moyer796d5112011-06-02 21:19:05 +0200356void elv_rb_add(struct rb_root *root, struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +0200357{
358 struct rb_node **p = &root->rb_node;
359 struct rb_node *parent = NULL;
360 struct request *__rq;
361
362 while (*p) {
363 parent = *p;
364 __rq = rb_entry(parent, struct request, rb_node);
365
Tejun Heo83096eb2009-05-07 22:24:39 +0900366 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200367 p = &(*p)->rb_left;
Jeff Moyer796d5112011-06-02 21:19:05 +0200368 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200369 p = &(*p)->rb_right;
Jens Axboe2e662b62006-07-13 11:55:04 +0200370 }
371
372 rb_link_node(&rq->rb_node, parent, p);
373 rb_insert_color(&rq->rb_node, root);
Jens Axboe2e662b62006-07-13 11:55:04 +0200374}
Jens Axboe2e662b62006-07-13 11:55:04 +0200375EXPORT_SYMBOL(elv_rb_add);
376
377void elv_rb_del(struct rb_root *root, struct request *rq)
378{
379 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
380 rb_erase(&rq->rb_node, root);
381 RB_CLEAR_NODE(&rq->rb_node);
382}
Jens Axboe2e662b62006-07-13 11:55:04 +0200383EXPORT_SYMBOL(elv_rb_del);
384
385struct request *elv_rb_find(struct rb_root *root, sector_t sector)
386{
387 struct rb_node *n = root->rb_node;
388 struct request *rq;
389
390 while (n) {
391 rq = rb_entry(n, struct request, rb_node);
392
Tejun Heo83096eb2009-05-07 22:24:39 +0900393 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200394 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900395 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200396 n = n->rb_right;
397 else
398 return rq;
399 }
400
401 return NULL;
402}
Jens Axboe2e662b62006-07-13 11:55:04 +0200403EXPORT_SYMBOL(elv_rb_find);
404
405/*
Tejun Heo8922e162005-10-20 16:23:44 +0200406 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200407 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200408 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200409 */
Jens Axboe165125e2007-07-24 09:28:11 +0200410void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200411{
412 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200413 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100414 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200415
Tejun Heo06b86242005-10-20 16:46:23 +0200416 if (q->last_merge == rq)
417 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200418
419 elv_rqhash_del(q, rq);
420
Tejun Heo15853af2005-11-10 08:52:05 +0100421 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200422
Jens Axboe1b47f532005-10-20 16:37:00 +0200423 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100424 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200425 list_for_each_prev(entry, &q->queue_head) {
426 struct request *pos = list_entry_rq(entry);
427
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200428 if ((rq->cmd_flags & REQ_DISCARD) !=
429 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100430 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100431 if (rq_data_dir(rq) != rq_data_dir(pos))
432 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100433 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200434 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900435 if (blk_rq_pos(rq) >= boundary) {
436 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200437 continue;
438 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900439 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200440 break;
441 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900442 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200443 break;
444 }
445
446 list_add(&rq->queuelist, entry);
447}
Jens Axboe2e662b62006-07-13 11:55:04 +0200448EXPORT_SYMBOL(elv_dispatch_sort);
449
Jens Axboe98170642006-07-28 09:23:08 +0200450/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200451 * Insert rq into dispatch queue of q. Queue lock must be held on
452 * entry. rq is added to the back of the dispatch queue. To be used by
453 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200454 */
455void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
456{
457 if (q->last_merge == rq)
458 q->last_merge = NULL;
459
460 elv_rqhash_del(q, rq);
461
462 q->nr_sorted--;
463
464 q->end_sector = rq_end_sector(rq);
465 q->boundary_rq = rq;
466 list_add_tail(&rq->queuelist, &q->queue_head);
467}
Jens Axboe2e662b62006-07-13 11:55:04 +0200468EXPORT_SYMBOL(elv_dispatch_add_tail);
469
Jens Axboe165125e2007-07-24 09:28:11 +0200470int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Jens Axboeb374d182008-10-31 10:05:07 +0100472 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200473 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200474 int ret;
475
Jens Axboe98170642006-07-28 09:23:08 +0200476 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100477 * Levels of merges:
478 * nomerges: No merges at all attempted
479 * noxmerges: Only simple one-hit cache try
480 * merges: All merge tries attempted
481 */
482 if (blk_queue_nomerges(q))
483 return ELEVATOR_NO_MERGE;
484
485 /*
Jens Axboe98170642006-07-28 09:23:08 +0200486 * First try one-hit cache.
487 */
Tejun Heo06b86242005-10-20 16:46:23 +0200488 if (q->last_merge) {
489 ret = elv_try_merge(q->last_merge, bio);
490 if (ret != ELEVATOR_NO_MERGE) {
491 *req = q->last_merge;
492 return ret;
493 }
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100496 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200497 return ELEVATOR_NO_MERGE;
498
Jens Axboe98170642006-07-28 09:23:08 +0200499 /*
500 * See if our hash lookup can find a potential backmerge.
501 */
502 __rq = elv_rqhash_find(q, bio->bi_sector);
503 if (__rq && elv_rq_merge_ok(__rq, bio)) {
504 *req = __rq;
505 return ELEVATOR_BACK_MERGE;
506 }
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (e->ops->elevator_merge_fn)
509 return e->ops->elevator_merge_fn(q, req, bio);
510
511 return ELEVATOR_NO_MERGE;
512}
513
Jens Axboe5e84ea32011-03-21 10:14:27 +0100514/*
515 * Attempt to do an insertion back merge. Only check for the case where
516 * we can append 'rq' to an existing request, so we can throw 'rq' away
517 * afterwards.
518 *
519 * Returns true if we merged, false otherwise
520 */
521static bool elv_attempt_insert_merge(struct request_queue *q,
522 struct request *rq)
523{
524 struct request *__rq;
525
526 if (blk_queue_nomerges(q))
527 return false;
528
529 /*
530 * First try one-hit cache.
531 */
532 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
533 return true;
534
535 if (blk_queue_noxmerges(q))
536 return false;
537
538 /*
539 * See if our hash lookup can find a potential backmerge.
540 */
541 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
542 if (__rq && blk_attempt_req_merge(q, __rq, rq))
543 return true;
544
545 return false;
546}
547
Jens Axboe165125e2007-07-24 09:28:11 +0200548void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Jens Axboeb374d182008-10-31 10:05:07 +0100550 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200553 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200554
Jens Axboe2e662b62006-07-13 11:55:04 +0200555 if (type == ELEVATOR_BACK_MERGE)
556 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200557
Tejun Heo06b86242005-10-20 16:46:23 +0200558 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
Jens Axboe165125e2007-07-24 09:28:11 +0200561void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 struct request *next)
563{
Jens Axboeb374d182008-10-31 10:05:07 +0100564 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100565 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Jens Axboe5e84ea32011-03-21 10:14:27 +0100567 if (next_sorted && e->ops->elevator_merge_req_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200569
Jens Axboe98170642006-07-28 09:23:08 +0200570 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200571
Jens Axboe5e84ea32011-03-21 10:14:27 +0100572 if (next_sorted) {
573 elv_rqhash_del(q, next);
574 q->nr_sorted--;
575 }
576
Tejun Heo06b86242005-10-20 16:46:23 +0200577 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Divyesh Shah812d4022010-04-08 21:14:23 -0700580void elv_bio_merged(struct request_queue *q, struct request *rq,
581 struct bio *bio)
582{
583 struct elevator_queue *e = q->elevator;
584
585 if (e->ops->elevator_bio_merged_fn)
586 e->ops->elevator_bio_merged_fn(q, rq, bio);
587}
588
Jens Axboe165125e2007-07-24 09:28:11 +0200589void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 /*
592 * it already went through dequeue, we need to decrement the
593 * in_flight count again
594 */
Tejun Heo8922e162005-10-20 16:23:44 +0200595 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200596 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200597 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100598 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Jens Axboe4aff5e22006-08-10 08:44:47 +0200601 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Jens Axboeb710a482011-03-30 09:52:30 +0200603 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Jerome Marchand26308ea2009-03-27 10:31:51 +0100606void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100607{
608 static int printed;
609 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
610 ;
611 if (q->nr_sorted == 0)
612 return;
613 if (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
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100620/*
621 * Call with queue lock held, interrupts disabled
622 */
Jens Axboef600abe2009-04-08 14:22:01 +0200623void elv_quiesce_start(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100624{
Martin K. Petersencd43e262009-05-22 17:17:52 -0400625 if (!q->elevator)
626 return;
627
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100628 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
629
630 /*
631 * make sure we don't have any requests in flight
632 */
633 elv_drain_elevator(q);
634 while (q->rq.elvpriv) {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200635 __blk_run_queue(q);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100636 spin_unlock_irq(q->queue_lock);
637 msleep(10);
638 spin_lock_irq(q->queue_lock);
639 elv_drain_elevator(q);
640 }
641}
642
Jens Axboef600abe2009-04-08 14:22:01 +0200643void elv_quiesce_end(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100644{
645 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
646}
647
Jens Axboeb710a482011-03-30 09:52:30 +0200648void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100650 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 rq->q = q;
653
Jens Axboeb710a482011-03-30 09:52:30 +0200654 if (rq->cmd_flags & REQ_SOFTBARRIER) {
655 /* barriers are scheduling boundary, update end_sector */
656 if (rq->cmd_type == REQ_TYPE_FS ||
657 (rq->cmd_flags & REQ_DISCARD)) {
658 q->end_sector = rq_end_sector(rq);
659 q->boundary_rq = rq;
660 }
661 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
Jens Axboe3aa72872011-04-21 19:28:35 +0200662 (where == ELEVATOR_INSERT_SORT ||
663 where == ELEVATOR_INSERT_SORT_MERGE))
Jens Axboeb710a482011-03-30 09:52:30 +0200664 where = ELEVATOR_INSERT_BACK;
665
Tejun Heo8922e162005-10-20 16:23:44 +0200666 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200667 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200668 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200669 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200670 list_add(&rq->queuelist, &q->queue_head);
671 break;
672
673 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200674 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100675 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200676 list_add_tail(&rq->queuelist, &q->queue_head);
677 /*
678 * We kick the queue here for the following reasons.
679 * - The elevator might have returned NULL previously
680 * to delay requests and returned them now. As the
681 * queue wasn't empty before this request, ll_rw_blk
682 * won't run the queue on return, resulting in hang.
683 * - Usually, back inserted requests won't be merged
684 * with anything. There's no point in delaying queue
685 * processing.
686 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200687 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200688 break;
689
Jens Axboe5e84ea32011-03-21 10:14:27 +0100690 case ELEVATOR_INSERT_SORT_MERGE:
691 /*
692 * If we succeed in merging this request with one in the
693 * queue already, we are done - rq has now been freed,
694 * so no need to do anything further.
695 */
696 if (elv_attempt_insert_merge(q, rq))
697 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200698 case ELEVATOR_INSERT_SORT:
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200699 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
700 !(rq->cmd_flags & REQ_DISCARD));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200701 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100702 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200703 if (rq_mergeable(rq)) {
704 elv_rqhash_add(q, rq);
705 if (!q->last_merge)
706 q->last_merge = rq;
707 }
708
Tejun Heoca235092005-11-01 17:23:49 +0900709 /*
710 * Some ioscheds (cfq) run q->request_fn directly, so
711 * rq cannot be accessed after calling
712 * elevator_add_req_fn.
713 */
714 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200715 break;
716
Tejun Heoae1b1532011-01-25 12:43:54 +0100717 case ELEVATOR_INSERT_FLUSH:
718 rq->cmd_flags |= REQ_SOFTBARRIER;
719 blk_insert_flush(rq);
720 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200721 default:
722 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700723 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200724 BUG();
725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
Jens Axboe2e662b62006-07-13 11:55:04 +0200727EXPORT_SYMBOL(__elv_add_request);
728
Jens Axboe7eaceac2011-03-10 08:52:07 +0100729void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
731 unsigned long flags;
732
733 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100734 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 spin_unlock_irqrestore(q->queue_lock, flags);
736}
Jens Axboe2e662b62006-07-13 11:55:04 +0200737EXPORT_SYMBOL(elv_add_request);
738
Jens Axboe165125e2007-07-24 09:28:11 +0200739struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
Jens Axboeb374d182008-10-31 10:05:07 +0100741 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
743 if (e->ops->elevator_latter_req_fn)
744 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return NULL;
746}
747
Jens Axboe165125e2007-07-24 09:28:11 +0200748struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Jens Axboeb374d182008-10-31 10:05:07 +0100750 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 if (e->ops->elevator_former_req_fn)
753 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return NULL;
755}
756
Jens Axboe165125e2007-07-24 09:28:11 +0200757int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
Jens Axboeb374d182008-10-31 10:05:07 +0100759 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200762 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Mike Snitzerc1867942011-02-11 11:08:00 +0100764 rq->elevator_private[0] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return 0;
766}
767
Jens Axboe165125e2007-07-24 09:28:11 +0200768void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Jens Axboeb374d182008-10-31 10:05:07 +0100770 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100773 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Jens Axboe165125e2007-07-24 09:28:11 +0200776int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Jens Axboeb374d182008-10-31 10:05:07 +0100778 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200781 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 return ELV_MQUEUE_MAY;
784}
785
Mike Anderson11914a52008-09-13 20:31:27 +0200786void elv_abort_queue(struct request_queue *q)
787{
788 struct request *rq;
789
Tejun Heoae1b1532011-01-25 12:43:54 +0100790 blk_abort_flushes(q);
791
Mike Anderson11914a52008-09-13 20:31:27 +0200792 while (!list_empty(&q->queue_head)) {
793 rq = list_entry_rq(q->queue_head.next);
794 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100795 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200796 /*
797 * Mark this request as started so we don't trigger
798 * any debug logic in the end I/O path.
799 */
800 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900801 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200802 }
803}
804EXPORT_SYMBOL(elv_abort_queue);
805
Jens Axboe165125e2007-07-24 09:28:11 +0200806void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Jens Axboeb374d182008-10-31 10:05:07 +0100808 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
810 /*
811 * request is released from the driver, io must be done
812 */
Tejun Heo8922e162005-10-20 16:23:44 +0200813 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200814 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200815 if ((rq->cmd_flags & REQ_SORTED) &&
816 e->ops->elevator_completed_req_fn)
Tejun Heo1bc691d2006-01-12 15:39:26 +0100817 e->ops->elevator_completed_req_fn(q, rq);
818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
Al Viro3d1ab402006-03-18 18:35:43 -0500821#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
822
823static ssize_t
824elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
825{
Al Viro3d1ab402006-03-18 18:35:43 -0500826 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100827 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500828 ssize_t error;
829
830 if (!entry->show)
831 return -EIO;
832
Jens Axboeb374d182008-10-31 10:05:07 +0100833 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500834 mutex_lock(&e->sysfs_lock);
835 error = e->ops ? entry->show(e, page) : -ENOENT;
836 mutex_unlock(&e->sysfs_lock);
837 return error;
838}
839
840static ssize_t
841elv_attr_store(struct kobject *kobj, struct attribute *attr,
842 const char *page, size_t length)
843{
Al Viro3d1ab402006-03-18 18:35:43 -0500844 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100845 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500846 ssize_t error;
847
848 if (!entry->store)
849 return -EIO;
850
Jens Axboeb374d182008-10-31 10:05:07 +0100851 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500852 mutex_lock(&e->sysfs_lock);
853 error = e->ops ? entry->store(e, page, length) : -ENOENT;
854 mutex_unlock(&e->sysfs_lock);
855 return error;
856}
857
Emese Revfy52cf25d2010-01-19 02:58:23 +0100858static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500859 .show = elv_attr_show,
860 .store = elv_attr_store,
861};
862
863static struct kobj_type elv_ktype = {
864 .sysfs_ops = &elv_sysfs_ops,
865 .release = elevator_release,
866};
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868int elv_register_queue(struct request_queue *q)
869{
Jens Axboeb374d182008-10-31 10:05:07 +0100870 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500871 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700873 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500874 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500875 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500876 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500877 while (attr->attr.name) {
878 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500879 break;
Al Viroe572ec72006-03-18 22:27:18 -0500880 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500881 }
882 }
883 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200884 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500885 }
886 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
Mike Snitzer01effb02010-05-11 08:57:42 +0200888EXPORT_SYMBOL(elv_register_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Jens Axboeb374d182008-10-31 10:05:07 +0100890static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +0200891{
892 kobject_uevent(&e->kobj, KOBJ_REMOVE);
893 kobject_del(&e->kobj);
Jens Axboe430c62f2010-10-07 09:35:16 +0200894 e->registered = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200895}
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897void elv_unregister_queue(struct request_queue *q)
898{
Jens Axboebc1c1162006-06-08 08:49:06 +0200899 if (q)
900 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901}
Mike Snitzer01effb02010-05-11 08:57:42 +0200902EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Adrian Bunk2fdd82b2007-12-12 18:51:56 +0100904void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100906 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200907
908 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100909 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200911 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Nate Diller5f003972006-01-24 10:07:58 +0100913 if (!strcmp(e->elevator_name, chosen_elevator) ||
914 (!*chosen_elevator &&
915 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100916 def = " (default)";
917
Jens Axboe4eb166d2008-02-01 00:37:27 +0100918 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
919 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}
921EXPORT_SYMBOL_GPL(elv_register);
922
923void elv_unregister(struct elevator_type *e)
924{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800925 struct task_struct *g, *p;
926
927 /*
928 * Iterate every thread in the process to remove the io contexts.
929 */
Al Viroe17a9482006-03-18 13:21:20 -0500930 if (e->ops.trim) {
931 read_lock(&tasklist_lock);
932 do_each_thread(g, p) {
933 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400934 if (p->io_context)
935 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500936 task_unlock(p);
937 } while_each_thread(g, p);
938 read_unlock(&tasklist_lock);
939 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800940
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200941 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200943 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945EXPORT_SYMBOL_GPL(elv_unregister);
946
947/*
948 * switch to new_e io scheduler. be careful not to introduce deadlocks -
949 * we don't free the old io scheduler, before we have allocated what we
950 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200951 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 */
Jens Axboe165125e2007-07-24 09:28:11 +0200953static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Jens Axboeb374d182008-10-31 10:05:07 +0100955 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200956 void *data;
Jens Axboe5dd531a2010-08-23 13:52:19 +0200957 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Tejun Heocb98fc82005-10-28 08:29:39 +0200959 /*
960 * Allocate new elevator
961 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200962 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (!e)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200964 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Jens Axboebc1c1162006-06-08 08:49:06 +0200966 data = elevator_init_queue(q, e);
967 if (!data) {
968 kobject_put(&e->kobj);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200969 return -ENOMEM;
Jens Axboebc1c1162006-06-08 08:49:06 +0200970 }
971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200973 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 */
Tejun Heocb98fc82005-10-28 08:29:39 +0200975 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +0200976 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /*
Jens Axboebc1c1162006-06-08 08:49:06 +0200979 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 old_elevator = q->elevator;
982
983 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 * attach and start new elevator
985 */
Jens Axboebc1c1162006-06-08 08:49:06 +0200986 elevator_attach(q, e, data);
987
988 spin_unlock_irq(q->queue_lock);
989
Jens Axboe430c62f2010-10-07 09:35:16 +0200990 if (old_elevator->registered) {
991 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Jens Axboe430c62f2010-10-07 09:35:16 +0200993 err = elv_register_queue(q);
994 if (err)
995 goto fail_register;
996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200999 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 */
1001 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001002 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001003 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001004 spin_unlock_irq(q->queue_lock);
1005
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001006 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1007
Jens Axboe5dd531a2010-08-23 13:52:19 +02001008 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010fail_register:
1011 /*
1012 * switch failed, exit the new io scheduler and reattach the old
1013 * one again (along with re-adding the sysfs dir)
1014 */
1015 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 q->elevator = old_elevator;
1017 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001018
1019 spin_lock_irq(q->queue_lock);
1020 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1021 spin_unlock_irq(q->queue_lock);
1022
Jens Axboe5dd531a2010-08-23 13:52:19 +02001023 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024}
1025
Jens Axboe5dd531a2010-08-23 13:52:19 +02001026/*
1027 * Switch this queue to the given IO scheduler.
1028 */
1029int elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
1031 char elevator_name[ELV_NAME_MAX];
1032 struct elevator_type *e;
1033
Martin K. Petersencd43e262009-05-22 17:17:52 -04001034 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +02001035 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001036
Li Zefanee2e9922008-10-14 08:49:56 +02001037 strlcpy(elevator_name, name, sizeof(elevator_name));
KOSAKI Motohiro8c279592009-10-09 08:48:08 +02001038 e = elevator_get(strstrip(elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (!e) {
1040 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1041 return -EINVAL;
1042 }
1043
Nate Diller2ca7d932005-10-30 15:02:24 -08001044 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1045 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +02001046 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -08001047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Jens Axboe5dd531a2010-08-23 13:52:19 +02001049 return elevator_switch(q, e);
1050}
1051EXPORT_SYMBOL(elevator_change);
1052
1053ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1054 size_t count)
1055{
1056 int ret;
1057
1058 if (!q->elevator)
1059 return count;
1060
1061 ret = elevator_change(q, name);
1062 if (!ret)
1063 return count;
1064
1065 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1066 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067}
1068
Jens Axboe165125e2007-07-24 09:28:11 +02001069ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Jens Axboeb374d182008-10-31 10:05:07 +01001071 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001072 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001073 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 int len = 0;
1075
Mike Snitzere36f7242010-05-24 09:07:32 +02001076 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001077 return sprintf(name, "none\n");
1078
1079 elv = e->elevator_type;
1080
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001081 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001082 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (!strcmp(elv->elevator_name, __e->elevator_name))
1084 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1085 else
1086 len += sprintf(name+len, "%s ", __e->elevator_name);
1087 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001088 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
1090 len += sprintf(len+name, "\n");
1091 return len;
1092}
1093
Jens Axboe165125e2007-07-24 09:28:11 +02001094struct request *elv_rb_former_request(struct request_queue *q,
1095 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001096{
1097 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1098
1099 if (rbprev)
1100 return rb_entry_rq(rbprev);
1101
1102 return NULL;
1103}
Jens Axboe2e662b62006-07-13 11:55:04 +02001104EXPORT_SYMBOL(elv_rb_former_request);
1105
Jens Axboe165125e2007-07-24 09:28:11 +02001106struct request *elv_rb_latter_request(struct request_queue *q,
1107 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001108{
1109 struct rb_node *rbnext = rb_next(&rq->rb_node);
1110
1111 if (rbnext)
1112 return rb_entry_rq(rbnext);
1113
1114 return NULL;
1115}
Jens Axboe2e662b62006-07-13 11:55:04 +02001116EXPORT_SYMBOL(elv_rb_latter_request);