blob: 6f6abc08bb565e61d63294e1a14f66ac6cda6a3d [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) {
158 char elv[ELV_NAME_MAX + strlen("-iosched")];
159
160 spin_unlock(&elv_list_lock);
161
wzt.wzt@gmail.coma506aed2010-04-02 08:41:14 +0200162 snprintf(elv, sizeof(elv), "%s-iosched", name);
Jens Axboee1640942008-02-19 10:20:37 +0100163
maximilian attemse180f592008-07-01 09:42:47 +0200164 request_module("%s", elv);
Jens Axboee1640942008-02-19 10:20:37 +0100165 spin_lock(&elv_list_lock);
166 e = elevator_find(name);
167 }
168
Tejun Heo2824bc932005-10-20 10:56:41 +0200169 if (e && !try_module_get(e->elevator_owner))
170 e = NULL;
171
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200172 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 return e;
175}
176
Jens Axboe165125e2007-07-24 09:28:11 +0200177static void *elevator_init_queue(struct request_queue *q,
178 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Jens Axboebb37b942006-12-01 10:42:33 +0100180 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200181}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Jens Axboe165125e2007-07-24 09:28:11 +0200183static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200184 void *data)
185{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200187 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188}
189
190static char chosen_elevator[16];
191
Nate Diller5f003972006-01-24 10:07:58 +0100192static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100194 /*
195 * Be backwards-compatible with previous kernels, so users
196 * won't get the wrong elevator.
197 */
Jens Axboe492af632009-10-03 09:37:51 +0200198 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800199 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
202__setup("elevator=", elevator_setup);
203
Al Viro3d1ab402006-03-18 18:35:43 -0500204static struct kobj_type elv_ktype;
205
Jens Axboeb374d182008-10-31 10:05:07 +0100206static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200207 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500208{
Jens Axboeb374d182008-10-31 10:05:07 +0100209 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200210 int i;
211
Jens Axboeb374d182008-10-31 10:05:07 +0100212 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200213 if (unlikely(!eq))
214 goto err;
215
Jens Axboe98170642006-07-28 09:23:08 +0200216 eq->ops = &e->ops;
217 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700218 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200219 mutex_init(&eq->sysfs_lock);
220
Jens Axboeb5deef92006-07-19 23:39:40 +0200221 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
222 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200223 if (!eq->hash)
224 goto err;
225
226 for (i = 0; i < ELV_HASH_ENTRIES; i++)
227 INIT_HLIST_HEAD(&eq->hash[i]);
228
Al Viro3d1ab402006-03-18 18:35:43 -0500229 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200230err:
231 kfree(eq);
232 elevator_put(e);
233 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500234}
235
236static void elevator_release(struct kobject *kobj)
237{
Jens Axboeb374d182008-10-31 10:05:07 +0100238 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200239
Jens Axboeb374d182008-10-31 10:05:07 +0100240 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500241 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200242 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500243 kfree(e);
244}
245
Jens Axboe165125e2007-07-24 09:28:11 +0200246int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 struct elevator_type *e = NULL;
249 struct elevator_queue *eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200250 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400252 if (unlikely(q->elevator))
253 return 0;
254
Tejun Heocb98fc82005-10-28 08:29:39 +0200255 INIT_LIST_HEAD(&q->queue_head);
256 q->last_merge = NULL;
257 q->end_sector = 0;
258 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200259
Jens Axboe4eb166d2008-02-01 00:37:27 +0100260 if (name) {
261 e = elevator_get(name);
262 if (!e)
263 return -EINVAL;
264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Jens Axboe4eb166d2008-02-01 00:37:27 +0100266 if (!e && *chosen_elevator) {
267 e = elevator_get(chosen_elevator);
268 if (!e)
269 printk(KERN_ERR "I/O scheduler %s not found\n",
270 chosen_elevator);
271 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100272
Jens Axboe4eb166d2008-02-01 00:37:27 +0100273 if (!e) {
274 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
275 if (!e) {
276 printk(KERN_ERR
277 "Default I/O scheduler not found. " \
278 "Using noop.\n");
279 e = elevator_get("noop");
280 }
Nate Diller5f003972006-01-24 10:07:58 +0100281 }
282
Jens Axboeb5deef92006-07-19 23:39:40 +0200283 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500284 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Jens Axboebc1c1162006-06-08 08:49:06 +0200287 data = elevator_init_queue(q, eq);
288 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500289 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200290 return -ENOMEM;
291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Jens Axboebc1c1162006-06-08 08:49:06 +0200293 elevator_attach(q, eq, data);
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400294 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
Jens Axboe2e662b62006-07-13 11:55:04 +0200296EXPORT_SYMBOL(elevator_init);
297
Jens Axboeb374d182008-10-31 10:05:07 +0100298void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Al Viro3d1ab402006-03-18 18:35:43 -0500300 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (e->ops->elevator_exit_fn)
302 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500303 e->ops = NULL;
304 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Al Viro3d1ab402006-03-18 18:35:43 -0500306 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
Jens Axboe2e662b62006-07-13 11:55:04 +0200308EXPORT_SYMBOL(elevator_exit);
309
Jens Axboe98170642006-07-28 09:23:08 +0200310static inline void __elv_rqhash_del(struct request *rq)
311{
312 hlist_del_init(&rq->hash);
313}
314
Jens Axboe165125e2007-07-24 09:28:11 +0200315static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200316{
317 if (ELV_ON_HASH(rq))
318 __elv_rqhash_del(rq);
319}
320
Jens Axboe165125e2007-07-24 09:28:11 +0200321static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200322{
Jens Axboeb374d182008-10-31 10:05:07 +0100323 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200324
325 BUG_ON(ELV_ON_HASH(rq));
326 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
327}
328
Jens Axboe165125e2007-07-24 09:28:11 +0200329static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200330{
331 __elv_rqhash_del(rq);
332 elv_rqhash_add(q, rq);
333}
334
Jens Axboe165125e2007-07-24 09:28:11 +0200335static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200336{
Jens Axboeb374d182008-10-31 10:05:07 +0100337 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200338 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
339 struct hlist_node *entry, *next;
340 struct request *rq;
341
342 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
343 BUG_ON(!ELV_ON_HASH(rq));
344
345 if (unlikely(!rq_mergeable(rq))) {
346 __elv_rqhash_del(rq);
347 continue;
348 }
349
350 if (rq_hash_key(rq) == offset)
351 return rq;
352 }
353
354 return NULL;
355}
356
Tejun Heo8922e162005-10-20 16:23:44 +0200357/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200358 * RB-tree support functions for inserting/lookup/removal of requests
359 * in a sorted RB tree.
360 */
361struct request *elv_rb_add(struct rb_root *root, struct request *rq)
362{
363 struct rb_node **p = &root->rb_node;
364 struct rb_node *parent = NULL;
365 struct request *__rq;
366
367 while (*p) {
368 parent = *p;
369 __rq = rb_entry(parent, struct request, rb_node);
370
Tejun Heo83096eb2009-05-07 22:24:39 +0900371 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200372 p = &(*p)->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900373 else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200374 p = &(*p)->rb_right;
375 else
376 return __rq;
377 }
378
379 rb_link_node(&rq->rb_node, parent, p);
380 rb_insert_color(&rq->rb_node, root);
381 return NULL;
382}
Jens Axboe2e662b62006-07-13 11:55:04 +0200383EXPORT_SYMBOL(elv_rb_add);
384
385void elv_rb_del(struct rb_root *root, struct request *rq)
386{
387 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
388 rb_erase(&rq->rb_node, root);
389 RB_CLEAR_NODE(&rq->rb_node);
390}
Jens Axboe2e662b62006-07-13 11:55:04 +0200391EXPORT_SYMBOL(elv_rb_del);
392
393struct request *elv_rb_find(struct rb_root *root, sector_t sector)
394{
395 struct rb_node *n = root->rb_node;
396 struct request *rq;
397
398 while (n) {
399 rq = rb_entry(n, struct request, rb_node);
400
Tejun Heo83096eb2009-05-07 22:24:39 +0900401 if (sector < blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200402 n = n->rb_left;
Tejun Heo83096eb2009-05-07 22:24:39 +0900403 else if (sector > blk_rq_pos(rq))
Jens Axboe2e662b62006-07-13 11:55:04 +0200404 n = n->rb_right;
405 else
406 return rq;
407 }
408
409 return NULL;
410}
Jens Axboe2e662b62006-07-13 11:55:04 +0200411EXPORT_SYMBOL(elv_rb_find);
412
413/*
Tejun Heo8922e162005-10-20 16:23:44 +0200414 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200415 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200416 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200417 */
Jens Axboe165125e2007-07-24 09:28:11 +0200418void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200419{
420 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200421 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100422 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200423
Jens Axboe73c10102011-03-08 13:19:51 +0100424 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
425
Tejun Heo06b86242005-10-20 16:46:23 +0200426 if (q->last_merge == rq)
427 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200428
429 elv_rqhash_del(q, rq);
430
Tejun Heo15853af2005-11-10 08:52:05 +0100431 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200432
Jens Axboe1b47f532005-10-20 16:37:00 +0200433 boundary = q->end_sector;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100434 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200435 list_for_each_prev(entry, &q->queue_head) {
436 struct request *pos = list_entry_rq(entry);
437
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200438 if ((rq->cmd_flags & REQ_DISCARD) !=
439 (pos->cmd_flags & REQ_DISCARD))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100440 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100441 if (rq_data_dir(rq) != rq_data_dir(pos))
442 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100443 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200444 break;
Tejun Heo83096eb2009-05-07 22:24:39 +0900445 if (blk_rq_pos(rq) >= boundary) {
446 if (blk_rq_pos(pos) < boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200447 continue;
448 } else {
Tejun Heo83096eb2009-05-07 22:24:39 +0900449 if (blk_rq_pos(pos) >= boundary)
Tejun Heo8922e162005-10-20 16:23:44 +0200450 break;
451 }
Tejun Heo83096eb2009-05-07 22:24:39 +0900452 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
Tejun Heo8922e162005-10-20 16:23:44 +0200453 break;
454 }
455
456 list_add(&rq->queuelist, entry);
457}
Jens Axboe2e662b62006-07-13 11:55:04 +0200458EXPORT_SYMBOL(elv_dispatch_sort);
459
Jens Axboe98170642006-07-28 09:23:08 +0200460/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200461 * Insert rq into dispatch queue of q. Queue lock must be held on
462 * entry. rq is added to the back of the dispatch queue. To be used by
463 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200464 */
465void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
466{
467 if (q->last_merge == rq)
468 q->last_merge = NULL;
469
470 elv_rqhash_del(q, rq);
471
472 q->nr_sorted--;
473
474 q->end_sector = rq_end_sector(rq);
475 q->boundary_rq = rq;
476 list_add_tail(&rq->queuelist, &q->queue_head);
477}
Jens Axboe2e662b62006-07-13 11:55:04 +0200478EXPORT_SYMBOL(elv_dispatch_add_tail);
479
Jens Axboe165125e2007-07-24 09:28:11 +0200480int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Jens Axboeb374d182008-10-31 10:05:07 +0100482 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200483 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200484 int ret;
485
Jens Axboe98170642006-07-28 09:23:08 +0200486 /*
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100487 * Levels of merges:
488 * nomerges: No merges at all attempted
489 * noxmerges: Only simple one-hit cache try
490 * merges: All merge tries attempted
491 */
492 if (blk_queue_nomerges(q))
493 return ELEVATOR_NO_MERGE;
494
495 /*
Jens Axboe98170642006-07-28 09:23:08 +0200496 * First try one-hit cache.
497 */
Tejun Heo06b86242005-10-20 16:46:23 +0200498 if (q->last_merge) {
499 ret = elv_try_merge(q->last_merge, bio);
500 if (ret != ELEVATOR_NO_MERGE) {
501 *req = q->last_merge;
502 return ret;
503 }
504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Alan D. Brunelle488991e2010-01-29 09:04:08 +0100506 if (blk_queue_noxmerges(q))
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200507 return ELEVATOR_NO_MERGE;
508
Jens Axboe98170642006-07-28 09:23:08 +0200509 /*
510 * See if our hash lookup can find a potential backmerge.
511 */
512 __rq = elv_rqhash_find(q, bio->bi_sector);
513 if (__rq && elv_rq_merge_ok(__rq, bio)) {
514 *req = __rq;
515 return ELEVATOR_BACK_MERGE;
516 }
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 if (e->ops->elevator_merge_fn)
519 return e->ops->elevator_merge_fn(q, req, bio);
520
521 return ELEVATOR_NO_MERGE;
522}
523
Jens Axboe5e84ea32011-03-21 10:14:27 +0100524/*
525 * Attempt to do an insertion back merge. Only check for the case where
526 * we can append 'rq' to an existing request, so we can throw 'rq' away
527 * afterwards.
528 *
529 * Returns true if we merged, false otherwise
530 */
531static bool elv_attempt_insert_merge(struct request_queue *q,
532 struct request *rq)
533{
534 struct request *__rq;
535
536 if (blk_queue_nomerges(q))
537 return false;
538
539 /*
540 * First try one-hit cache.
541 */
542 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
543 return true;
544
545 if (blk_queue_noxmerges(q))
546 return false;
547
548 /*
549 * See if our hash lookup can find a potential backmerge.
550 */
551 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
552 if (__rq && blk_attempt_req_merge(q, __rq, rq))
553 return true;
554
555 return false;
556}
557
Jens Axboe165125e2007-07-24 09:28:11 +0200558void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Jens Axboeb374d182008-10-31 10:05:07 +0100560 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200563 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200564
Jens Axboe2e662b62006-07-13 11:55:04 +0200565 if (type == ELEVATOR_BACK_MERGE)
566 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200567
Tejun Heo06b86242005-10-20 16:46:23 +0200568 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569}
570
Jens Axboe165125e2007-07-24 09:28:11 +0200571void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 struct request *next)
573{
Jens Axboeb374d182008-10-31 10:05:07 +0100574 struct elevator_queue *e = q->elevator;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100575 const int next_sorted = next->cmd_flags & REQ_SORTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Jens Axboe5e84ea32011-03-21 10:14:27 +0100577 if (next_sorted && e->ops->elevator_merge_req_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200579
Jens Axboe98170642006-07-28 09:23:08 +0200580 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200581
Jens Axboe5e84ea32011-03-21 10:14:27 +0100582 if (next_sorted) {
583 elv_rqhash_del(q, next);
584 q->nr_sorted--;
585 }
586
Tejun Heo06b86242005-10-20 16:46:23 +0200587 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
589
Divyesh Shah812d4022010-04-08 21:14:23 -0700590void elv_bio_merged(struct request_queue *q, struct request *rq,
591 struct bio *bio)
592{
593 struct elevator_queue *e = q->elevator;
594
595 if (e->ops->elevator_bio_merged_fn)
596 e->ops->elevator_bio_merged_fn(q, rq, bio);
597}
598
Jens Axboe165125e2007-07-24 09:28:11 +0200599void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /*
602 * it already went through dequeue, we need to decrement the
603 * in_flight count again
604 */
Tejun Heo8922e162005-10-20 16:23:44 +0200605 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200606 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200607 if (rq->cmd_flags & REQ_SORTED)
Jens Axboecad97512007-01-14 22:26:09 +1100608 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Jens Axboe4aff5e22006-08-10 08:44:47 +0200611 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Jens Axboeb710a482011-03-30 09:52:30 +0200613 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614}
615
Jerome Marchand26308ea2009-03-27 10:31:51 +0100616void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100617{
618 static int printed;
619 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
620 ;
621 if (q->nr_sorted == 0)
622 return;
623 if (printed++ < 10) {
624 printk(KERN_ERR "%s: forced dispatching is broken "
625 "(nr_sorted=%u), please report this\n",
626 q->elevator->elevator_type->elevator_name, q->nr_sorted);
627 }
628}
629
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100630/*
631 * Call with queue lock held, interrupts disabled
632 */
Jens Axboef600abe2009-04-08 14:22:01 +0200633void elv_quiesce_start(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100634{
Martin K. Petersencd43e262009-05-22 17:17:52 -0400635 if (!q->elevator)
636 return;
637
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100638 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
639
640 /*
641 * make sure we don't have any requests in flight
642 */
643 elv_drain_elevator(q);
644 while (q->rq.elvpriv) {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200645 __blk_run_queue(q);
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100646 spin_unlock_irq(q->queue_lock);
647 msleep(10);
648 spin_lock_irq(q->queue_lock);
649 elv_drain_elevator(q);
650 }
651}
652
Jens Axboef600abe2009-04-08 14:22:01 +0200653void elv_quiesce_end(struct request_queue *q)
Jens Axboe6c7e8ce2009-03-27 10:30:47 +0100654{
655 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
656}
657
Jens Axboeb710a482011-03-30 09:52:30 +0200658void __elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100660 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 rq->q = q;
663
Jens Axboeb710a482011-03-30 09:52:30 +0200664 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
665
666 if (rq->cmd_flags & REQ_SOFTBARRIER) {
667 /* barriers are scheduling boundary, update end_sector */
668 if (rq->cmd_type == REQ_TYPE_FS ||
669 (rq->cmd_flags & REQ_DISCARD)) {
670 q->end_sector = rq_end_sector(rq);
671 q->boundary_rq = rq;
672 }
673 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
674 where == ELEVATOR_INSERT_SORT)
675 where = ELEVATOR_INSERT_BACK;
676
Tejun Heo8922e162005-10-20 16:23:44 +0200677 switch (where) {
Tejun Heo28e7d182010-09-03 11:56:16 +0200678 case ELEVATOR_INSERT_REQUEUE:
Tejun Heo8922e162005-10-20 16:23:44 +0200679 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200680 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200681 list_add(&rq->queuelist, &q->queue_head);
682 break;
683
684 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200685 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100686 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200687 list_add_tail(&rq->queuelist, &q->queue_head);
688 /*
689 * We kick the queue here for the following reasons.
690 * - The elevator might have returned NULL previously
691 * to delay requests and returned them now. As the
692 * queue wasn't empty before this request, ll_rw_blk
693 * won't run the queue on return, resulting in hang.
694 * - Usually, back inserted requests won't be merged
695 * with anything. There's no point in delaying queue
696 * processing.
697 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200698 __blk_run_queue(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200699 break;
700
Jens Axboe5e84ea32011-03-21 10:14:27 +0100701 case ELEVATOR_INSERT_SORT_MERGE:
702 /*
703 * If we succeed in merging this request with one in the
704 * queue already, we are done - rq has now been freed,
705 * so no need to do anything further.
706 */
707 if (elv_attempt_insert_merge(q, rq))
708 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200709 case ELEVATOR_INSERT_SORT:
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200710 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
711 !(rq->cmd_flags & REQ_DISCARD));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200712 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100713 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200714 if (rq_mergeable(rq)) {
715 elv_rqhash_add(q, rq);
716 if (!q->last_merge)
717 q->last_merge = rq;
718 }
719
Tejun Heoca235092005-11-01 17:23:49 +0900720 /*
721 * Some ioscheds (cfq) run q->request_fn directly, so
722 * rq cannot be accessed after calling
723 * elevator_add_req_fn.
724 */
725 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200726 break;
727
Tejun Heoae1b1532011-01-25 12:43:54 +0100728 case ELEVATOR_INSERT_FLUSH:
729 rq->cmd_flags |= REQ_SOFTBARRIER;
730 blk_insert_flush(rq);
731 break;
Tejun Heo8922e162005-10-20 16:23:44 +0200732 default:
733 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700734 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200735 BUG();
736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
Jens Axboe2e662b62006-07-13 11:55:04 +0200738EXPORT_SYMBOL(__elv_add_request);
739
Jens Axboe7eaceac2011-03-10 08:52:07 +0100740void elv_add_request(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 unsigned long flags;
743
744 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100745 __elv_add_request(q, rq, where);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 spin_unlock_irqrestore(q->queue_lock, flags);
747}
Jens Axboe2e662b62006-07-13 11:55:04 +0200748EXPORT_SYMBOL(elv_add_request);
749
Jens Axboe165125e2007-07-24 09:28:11 +0200750struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
Jens Axboeb374d182008-10-31 10:05:07 +0100752 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
754 if (e->ops->elevator_latter_req_fn)
755 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return NULL;
757}
758
Jens Axboe165125e2007-07-24 09:28:11 +0200759struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
Jens Axboeb374d182008-10-31 10:05:07 +0100761 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763 if (e->ops->elevator_former_req_fn)
764 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return NULL;
766}
767
Jens Axboe165125e2007-07-24 09:28:11 +0200768int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
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_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200773 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Mike Snitzerc1867942011-02-11 11:08:00 +0100775 rq->elevator_private[0] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 return 0;
777}
778
Jens Axboe165125e2007-07-24 09:28:11 +0200779void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Jens Axboeb374d182008-10-31 10:05:07 +0100781 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100784 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Jens Axboe165125e2007-07-24 09:28:11 +0200787int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Jens Axboeb374d182008-10-31 10:05:07 +0100789 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200792 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794 return ELV_MQUEUE_MAY;
795}
796
Mike Anderson11914a52008-09-13 20:31:27 +0200797void elv_abort_queue(struct request_queue *q)
798{
799 struct request *rq;
800
Tejun Heoae1b1532011-01-25 12:43:54 +0100801 blk_abort_flushes(q);
802
Mike Anderson11914a52008-09-13 20:31:27 +0200803 while (!list_empty(&q->queue_head)) {
804 rq = list_entry_rq(q->queue_head.next);
805 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100806 trace_block_rq_abort(q, rq);
Kiyoshi Ueda53c663c2009-06-02 08:44:01 +0200807 /*
808 * Mark this request as started so we don't trigger
809 * any debug logic in the end I/O path.
810 */
811 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +0900812 __blk_end_request_all(rq, -EIO);
Mike Anderson11914a52008-09-13 20:31:27 +0200813 }
814}
815EXPORT_SYMBOL(elv_abort_queue);
816
Jens Axboe165125e2007-07-24 09:28:11 +0200817void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
Jens Axboeb374d182008-10-31 10:05:07 +0100819 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 /*
822 * request is released from the driver, io must be done
823 */
Tejun Heo8922e162005-10-20 16:23:44 +0200824 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +0200825 q->in_flight[rq_is_sync(rq)]--;
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200826 if ((rq->cmd_flags & REQ_SORTED) &&
827 e->ops->elevator_completed_req_fn)
Tejun Heo1bc691d2006-01-12 15:39:26 +0100828 e->ops->elevator_completed_req_fn(q, rq);
829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830}
831
Al Viro3d1ab402006-03-18 18:35:43 -0500832#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
833
834static ssize_t
835elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
836{
Al Viro3d1ab402006-03-18 18:35:43 -0500837 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100838 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500839 ssize_t error;
840
841 if (!entry->show)
842 return -EIO;
843
Jens Axboeb374d182008-10-31 10:05:07 +0100844 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500845 mutex_lock(&e->sysfs_lock);
846 error = e->ops ? entry->show(e, page) : -ENOENT;
847 mutex_unlock(&e->sysfs_lock);
848 return error;
849}
850
851static ssize_t
852elv_attr_store(struct kobject *kobj, struct attribute *attr,
853 const char *page, size_t length)
854{
Al Viro3d1ab402006-03-18 18:35:43 -0500855 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100856 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500857 ssize_t error;
858
859 if (!entry->store)
860 return -EIO;
861
Jens Axboeb374d182008-10-31 10:05:07 +0100862 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500863 mutex_lock(&e->sysfs_lock);
864 error = e->ops ? entry->store(e, page, length) : -ENOENT;
865 mutex_unlock(&e->sysfs_lock);
866 return error;
867}
868
Emese Revfy52cf25d2010-01-19 02:58:23 +0100869static const struct sysfs_ops elv_sysfs_ops = {
Al Viro3d1ab402006-03-18 18:35:43 -0500870 .show = elv_attr_show,
871 .store = elv_attr_store,
872};
873
874static struct kobj_type elv_ktype = {
875 .sysfs_ops = &elv_sysfs_ops,
876 .release = elevator_release,
877};
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879int elv_register_queue(struct request_queue *q)
880{
Jens Axboeb374d182008-10-31 10:05:07 +0100881 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500882 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -0700884 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -0500885 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500886 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500887 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500888 while (attr->attr.name) {
889 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500890 break;
Al Viroe572ec72006-03-18 22:27:18 -0500891 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500892 }
893 }
894 kobject_uevent(&e->kobj, KOBJ_ADD);
Jens Axboe430c62f2010-10-07 09:35:16 +0200895 e->registered = 1;
Al Viro3d1ab402006-03-18 18:35:43 -0500896 }
897 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
Mike Snitzer01effb02010-05-11 08:57:42 +0200899EXPORT_SYMBOL(elv_register_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Jens Axboeb374d182008-10-31 10:05:07 +0100901static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +0200902{
903 kobject_uevent(&e->kobj, KOBJ_REMOVE);
904 kobject_del(&e->kobj);
Jens Axboe430c62f2010-10-07 09:35:16 +0200905 e->registered = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200906}
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908void elv_unregister_queue(struct request_queue *q)
909{
Jens Axboebc1c1162006-06-08 08:49:06 +0200910 if (q)
911 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
Mike Snitzer01effb02010-05-11 08:57:42 +0200913EXPORT_SYMBOL(elv_unregister_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Adrian Bunk2fdd82b2007-12-12 18:51:56 +0100915void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100917 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200918
919 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100920 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200922 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Nate Diller5f003972006-01-24 10:07:58 +0100924 if (!strcmp(e->elevator_name, chosen_elevator) ||
925 (!*chosen_elevator &&
926 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +0100927 def = " (default)";
928
Jens Axboe4eb166d2008-02-01 00:37:27 +0100929 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
930 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931}
932EXPORT_SYMBOL_GPL(elv_register);
933
934void elv_unregister(struct elevator_type *e)
935{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800936 struct task_struct *g, *p;
937
938 /*
939 * Iterate every thread in the process to remove the io contexts.
940 */
Al Viroe17a9482006-03-18 13:21:20 -0500941 if (e->ops.trim) {
942 read_lock(&tasklist_lock);
943 do_each_thread(g, p) {
944 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +0400945 if (p->io_context)
946 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -0500947 task_unlock(p);
948 } while_each_thread(g, p);
949 read_unlock(&tasklist_lock);
950 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800951
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200952 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200954 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956EXPORT_SYMBOL_GPL(elv_unregister);
957
958/*
959 * switch to new_e io scheduler. be careful not to introduce deadlocks -
960 * we don't free the old io scheduler, before we have allocated what we
961 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200962 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 */
Jens Axboe165125e2007-07-24 09:28:11 +0200964static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
Jens Axboeb374d182008-10-31 10:05:07 +0100966 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200967 void *data;
Jens Axboe5dd531a2010-08-23 13:52:19 +0200968 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Tejun Heocb98fc82005-10-28 08:29:39 +0200970 /*
971 * Allocate new elevator
972 */
Jens Axboeb5deef92006-07-19 23:39:40 +0200973 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (!e)
Jens Axboe5dd531a2010-08-23 13:52:19 +0200975 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Jens Axboebc1c1162006-06-08 08:49:06 +0200977 data = elevator_init_queue(q, e);
978 if (!data) {
979 kobject_put(&e->kobj);
Jens Axboe5dd531a2010-08-23 13:52:19 +0200980 return -ENOMEM;
Jens Axboebc1c1162006-06-08 08:49:06 +0200981 }
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200984 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 */
Tejun Heocb98fc82005-10-28 08:29:39 +0200986 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +0200987 elv_quiesce_start(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 /*
Jens Axboebc1c1162006-06-08 08:49:06 +0200990 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 old_elevator = q->elevator;
993
994 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 * attach and start new elevator
996 */
Jens Axboebc1c1162006-06-08 08:49:06 +0200997 elevator_attach(q, e, data);
998
999 spin_unlock_irq(q->queue_lock);
1000
Jens Axboe430c62f2010-10-07 09:35:16 +02001001 if (old_elevator->registered) {
1002 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Jens Axboe430c62f2010-10-07 09:35:16 +02001004 err = elv_register_queue(q);
1005 if (err)
1006 goto fail_register;
1007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001010 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 */
1012 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001013 spin_lock_irq(q->queue_lock);
Jens Axboef600abe2009-04-08 14:22:01 +02001014 elv_quiesce_end(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001015 spin_unlock_irq(q->queue_lock);
1016
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001017 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1018
Jens Axboe5dd531a2010-08-23 13:52:19 +02001019 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021fail_register:
1022 /*
1023 * switch failed, exit the new io scheduler and reattach the old
1024 * one again (along with re-adding the sysfs dir)
1025 */
1026 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 q->elevator = old_elevator;
1028 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001029
1030 spin_lock_irq(q->queue_lock);
1031 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1032 spin_unlock_irq(q->queue_lock);
1033
Jens Axboe5dd531a2010-08-23 13:52:19 +02001034 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035}
1036
Jens Axboe5dd531a2010-08-23 13:52:19 +02001037/*
1038 * Switch this queue to the given IO scheduler.
1039 */
1040int elevator_change(struct request_queue *q, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 char elevator_name[ELV_NAME_MAX];
1043 struct elevator_type *e;
1044
Martin K. Petersencd43e262009-05-22 17:17:52 -04001045 if (!q->elevator)
Jens Axboe5dd531a2010-08-23 13:52:19 +02001046 return -ENXIO;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001047
Li Zefanee2e9922008-10-14 08:49:56 +02001048 strlcpy(elevator_name, name, sizeof(elevator_name));
KOSAKI Motohiro8c279592009-10-09 08:48:08 +02001049 e = elevator_get(strstrip(elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (!e) {
1051 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1052 return -EINVAL;
1053 }
1054
Nate Diller2ca7d932005-10-30 15:02:24 -08001055 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1056 elevator_put(e);
Jens Axboe5dd531a2010-08-23 13:52:19 +02001057 return 0;
Nate Diller2ca7d932005-10-30 15:02:24 -08001058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Jens Axboe5dd531a2010-08-23 13:52:19 +02001060 return elevator_switch(q, e);
1061}
1062EXPORT_SYMBOL(elevator_change);
1063
1064ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1065 size_t count)
1066{
1067 int ret;
1068
1069 if (!q->elevator)
1070 return count;
1071
1072 ret = elevator_change(q, name);
1073 if (!ret)
1074 return count;
1075
1076 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1077 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Jens Axboe165125e2007-07-24 09:28:11 +02001080ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
Jens Axboeb374d182008-10-31 10:05:07 +01001082 struct elevator_queue *e = q->elevator;
Martin K. Petersencd43e262009-05-22 17:17:52 -04001083 struct elevator_type *elv;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001084 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 int len = 0;
1086
Mike Snitzere36f7242010-05-24 09:07:32 +02001087 if (!q->elevator || !blk_queue_stackable(q))
Martin K. Petersencd43e262009-05-22 17:17:52 -04001088 return sprintf(name, "none\n");
1089
1090 elv = e->elevator_type;
1091
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001092 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001093 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (!strcmp(elv->elevator_name, __e->elevator_name))
1095 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1096 else
1097 len += sprintf(name+len, "%s ", __e->elevator_name);
1098 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001099 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 len += sprintf(len+name, "\n");
1102 return len;
1103}
1104
Jens Axboe165125e2007-07-24 09:28:11 +02001105struct request *elv_rb_former_request(struct request_queue *q,
1106 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001107{
1108 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1109
1110 if (rbprev)
1111 return rb_entry_rq(rbprev);
1112
1113 return NULL;
1114}
Jens Axboe2e662b62006-07-13 11:55:04 +02001115EXPORT_SYMBOL(elv_rb_former_request);
1116
Jens Axboe165125e2007-07-24 09:28:11 +02001117struct request *elv_rb_latter_request(struct request_queue *q,
1118 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001119{
1120 struct rb_node *rbnext = rb_next(&rq->rb_node);
1121
1122 if (rbnext)
1123 return rb_entry_rq(rbnext);
1124
1125 return NULL;
1126}
Jens Axboe2e662b62006-07-13 11:55:04 +02001127EXPORT_SYMBOL(elv_rb_latter_request);