blob: 98259eda0ef66d4051cc5da958c0191f353b8158 [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>
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010036#include <trace/block.h>
Jens Axboe98170642006-07-28 09:23:08 +020037#include <linux/hash.h>
Jens Axboe0835da62008-08-26 09:15:47 +020038#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Jens Axboe242f9dc2008-09-14 05:55:09 -070040#include "blk.h"
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042static DEFINE_SPINLOCK(elv_list_lock);
43static LIST_HEAD(elv_list);
44
Ingo Molnar0bfc2452008-11-26 11:59:56 +010045DEFINE_TRACE(block_rq_abort);
46
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
Jens Axboe98170642006-07-28 09:23:08 +020048 * Merge hash stuff.
49 */
50static const int elv_hash_shift = 6;
51#define ELV_HASH_BLOCK(sec) ((sec) >> 3)
Jens Axboe4eb166d2008-02-01 00:37:27 +010052#define ELV_HASH_FN(sec) \
53 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
Jens Axboe98170642006-07-28 09:23:08 +020054#define ELV_HASH_ENTRIES (1 << elv_hash_shift)
55#define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors)
56#define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash))
57
Ingo Molnar0bfc2452008-11-26 11:59:56 +010058DEFINE_TRACE(block_rq_insert);
59DEFINE_TRACE(block_rq_issue);
60
Jens Axboe98170642006-07-28 09:23:08 +020061/*
Jens Axboeda775262006-12-20 11:04:12 +010062 * Query io scheduler to see if the current process issuing bio may be
63 * merged with rq.
64 */
65static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
66{
Jens Axboe165125e2007-07-24 09:28:11 +020067 struct request_queue *q = rq->q;
Jens Axboeb374d182008-10-31 10:05:07 +010068 struct elevator_queue *e = q->elevator;
Jens Axboeda775262006-12-20 11:04:12 +010069
70 if (e->ops->elevator_allow_merge_fn)
71 return e->ops->elevator_allow_merge_fn(q, rq, bio);
72
73 return 1;
74}
75
76/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 * can we safely merge with this request?
78 */
Adrian Bunk72ed0bf2008-04-29 09:49:05 +020079int elv_rq_merge_ok(struct request *rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
81 if (!rq_mergeable(rq))
82 return 0;
83
84 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +010085 * Don't merge file system requests and discard requests
86 */
87 if (bio_discard(bio) != bio_discard(rq->bio))
88 return 0;
89
90 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 * different data direction or already started, don't merge
92 */
93 if (bio_data_dir(bio) != rq_data_dir(rq))
94 return 0;
95
96 /*
Jens Axboeda775262006-12-20 11:04:12 +010097 * must be same device and not a special request
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 */
Jens Axboebb4067e2006-12-21 21:20:01 +010099 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
Jens Axboeda775262006-12-20 11:04:12 +0100100 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200102 /*
103 * only merge integrity protected bio into ditto rq
104 */
105 if (bio_integrity(bio) != blk_integrity_rq(rq))
106 return 0;
107
Jens Axboeda775262006-12-20 11:04:12 +0100108 if (!elv_iosched_allow_merge(rq, bio))
109 return 0;
110
111 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113EXPORT_SYMBOL(elv_rq_merge_ok);
114
Coywolf Qi Hunt769db452005-12-28 10:55:49 +0100115static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
117 int ret = ELEVATOR_NO_MERGE;
118
119 /*
120 * we can merge and sequence is ok, check if it's possible
121 */
122 if (elv_rq_merge_ok(__rq, bio)) {
123 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
124 ret = ELEVATOR_BACK_MERGE;
125 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
126 ret = ELEVATOR_FRONT_MERGE;
127 }
128
129 return ret;
130}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132static struct elevator_type *elevator_find(const char *name)
133{
Vasily Tarasova22b1692006-10-11 09:24:27 +0200134 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
Matthias Kaehlcke70cee262007-07-10 12:26:24 +0200136 list_for_each_entry(e, &elv_list, list) {
Vasily Tarasova22b1692006-10-11 09:24:27 +0200137 if (!strcmp(e->elevator_name, name))
138 return e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Vasily Tarasova22b1692006-10-11 09:24:27 +0200141 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
144static void elevator_put(struct elevator_type *e)
145{
146 module_put(e->elevator_owner);
147}
148
149static struct elevator_type *elevator_get(const char *name)
150{
Tejun Heo2824bc932005-10-20 10:56:41 +0200151 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200153 spin_lock(&elv_list_lock);
Tejun Heo2824bc932005-10-20 10:56:41 +0200154
155 e = elevator_find(name);
Jens Axboee1640942008-02-19 10:20:37 +0100156 if (!e) {
157 char elv[ELV_NAME_MAX + strlen("-iosched")];
158
159 spin_unlock(&elv_list_lock);
160
161 if (!strcmp(name, "anticipatory"))
162 sprintf(elv, "as-iosched");
163 else
164 sprintf(elv, "%s-iosched", name);
165
maximilian attemse180f592008-07-01 09:42:47 +0200166 request_module("%s", elv);
Jens Axboee1640942008-02-19 10:20:37 +0100167 spin_lock(&elv_list_lock);
168 e = elevator_find(name);
169 }
170
Tejun Heo2824bc932005-10-20 10:56:41 +0200171 if (e && !try_module_get(e->elevator_owner))
172 e = NULL;
173
Jens Axboe2a12dcd2007-04-26 14:41:53 +0200174 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 return e;
177}
178
Jens Axboe165125e2007-07-24 09:28:11 +0200179static void *elevator_init_queue(struct request_queue *q,
180 struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
Jens Axboebb37b942006-12-01 10:42:33 +0100182 return eq->ops->elevator_init_fn(q);
Jens Axboebc1c1162006-06-08 08:49:06 +0200183}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Jens Axboe165125e2007-07-24 09:28:11 +0200185static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
Jens Axboebc1c1162006-06-08 08:49:06 +0200186 void *data)
187{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200189 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
192static char chosen_elevator[16];
193
Nate Diller5f003972006-01-24 10:07:58 +0100194static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100196 /*
197 * Be backwards-compatible with previous kernels, so users
198 * won't get the wrong elevator.
199 */
Nate Diller5f003972006-01-24 10:07:58 +0100200 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100201 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100202 else
Nate Diller5f003972006-01-24 10:07:58 +0100203 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800204 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
207__setup("elevator=", elevator_setup);
208
Al Viro3d1ab402006-03-18 18:35:43 -0500209static struct kobj_type elv_ktype;
210
Jens Axboeb374d182008-10-31 10:05:07 +0100211static struct elevator_queue *elevator_alloc(struct request_queue *q,
Jens Axboe165125e2007-07-24 09:28:11 +0200212 struct elevator_type *e)
Al Viro3d1ab402006-03-18 18:35:43 -0500213{
Jens Axboeb374d182008-10-31 10:05:07 +0100214 struct elevator_queue *eq;
Jens Axboe98170642006-07-28 09:23:08 +0200215 int i;
216
Jens Axboeb374d182008-10-31 10:05:07 +0100217 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200218 if (unlikely(!eq))
219 goto err;
220
Jens Axboe98170642006-07-28 09:23:08 +0200221 eq->ops = &e->ops;
222 eq->elevator_type = e;
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700223 kobject_init(&eq->kobj, &elv_ktype);
Jens Axboe98170642006-07-28 09:23:08 +0200224 mutex_init(&eq->sysfs_lock);
225
Jens Axboeb5deef92006-07-19 23:39:40 +0200226 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
227 GFP_KERNEL, q->node);
Jens Axboe98170642006-07-28 09:23:08 +0200228 if (!eq->hash)
229 goto err;
230
231 for (i = 0; i < ELV_HASH_ENTRIES; i++)
232 INIT_HLIST_HEAD(&eq->hash[i]);
233
Al Viro3d1ab402006-03-18 18:35:43 -0500234 return eq;
Jens Axboe98170642006-07-28 09:23:08 +0200235err:
236 kfree(eq);
237 elevator_put(e);
238 return NULL;
Al Viro3d1ab402006-03-18 18:35:43 -0500239}
240
241static void elevator_release(struct kobject *kobj)
242{
Jens Axboeb374d182008-10-31 10:05:07 +0100243 struct elevator_queue *e;
Jens Axboe98170642006-07-28 09:23:08 +0200244
Jens Axboeb374d182008-10-31 10:05:07 +0100245 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500246 elevator_put(e->elevator_type);
Jens Axboe98170642006-07-28 09:23:08 +0200247 kfree(e->hash);
Al Viro3d1ab402006-03-18 18:35:43 -0500248 kfree(e);
249}
250
Jens Axboe165125e2007-07-24 09:28:11 +0200251int elevator_init(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
253 struct elevator_type *e = NULL;
254 struct elevator_queue *eq;
255 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200256 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Tejun Heocb98fc82005-10-28 08:29:39 +0200258 INIT_LIST_HEAD(&q->queue_head);
259 q->last_merge = NULL;
260 q->end_sector = 0;
261 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200262
Jens Axboe4eb166d2008-02-01 00:37:27 +0100263 if (name) {
264 e = elevator_get(name);
265 if (!e)
266 return -EINVAL;
267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Jens Axboe4eb166d2008-02-01 00:37:27 +0100269 if (!e && *chosen_elevator) {
270 e = elevator_get(chosen_elevator);
271 if (!e)
272 printk(KERN_ERR "I/O scheduler %s not found\n",
273 chosen_elevator);
274 }
Nate Diller248d5ca2006-01-24 10:09:14 +0100275
Jens Axboe4eb166d2008-02-01 00:37:27 +0100276 if (!e) {
277 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
278 if (!e) {
279 printk(KERN_ERR
280 "Default I/O scheduler not found. " \
281 "Using noop.\n");
282 e = elevator_get("noop");
283 }
Nate Diller5f003972006-01-24 10:07:58 +0100284 }
285
Jens Axboeb5deef92006-07-19 23:39:40 +0200286 eq = elevator_alloc(q, e);
Al Viro3d1ab402006-03-18 18:35:43 -0500287 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Jens Axboebc1c1162006-06-08 08:49:06 +0200290 data = elevator_init_queue(q, eq);
291 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500292 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200293 return -ENOMEM;
294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Jens Axboebc1c1162006-06-08 08:49:06 +0200296 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return ret;
298}
Jens Axboe2e662b62006-07-13 11:55:04 +0200299EXPORT_SYMBOL(elevator_init);
300
Jens Axboeb374d182008-10-31 10:05:07 +0100301void elevator_exit(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Al Viro3d1ab402006-03-18 18:35:43 -0500303 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 if (e->ops->elevator_exit_fn)
305 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500306 e->ops = NULL;
307 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Al Viro3d1ab402006-03-18 18:35:43 -0500309 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
Jens Axboe2e662b62006-07-13 11:55:04 +0200311EXPORT_SYMBOL(elevator_exit);
312
Jens Axboe165125e2007-07-24 09:28:11 +0200313static void elv_activate_rq(struct request_queue *q, struct request *rq)
Jens Axboecad97512007-01-14 22:26:09 +1100314{
Jens Axboeb374d182008-10-31 10:05:07 +0100315 struct elevator_queue *e = q->elevator;
Jens Axboecad97512007-01-14 22:26:09 +1100316
317 if (e->ops->elevator_activate_req_fn)
318 e->ops->elevator_activate_req_fn(q, rq);
319}
320
Jens Axboe165125e2007-07-24 09:28:11 +0200321static void elv_deactivate_rq(struct request_queue *q, struct request *rq)
Jens Axboecad97512007-01-14 22:26:09 +1100322{
Jens Axboeb374d182008-10-31 10:05:07 +0100323 struct elevator_queue *e = q->elevator;
Jens Axboecad97512007-01-14 22:26:09 +1100324
325 if (e->ops->elevator_deactivate_req_fn)
326 e->ops->elevator_deactivate_req_fn(q, rq);
327}
328
Jens Axboe98170642006-07-28 09:23:08 +0200329static inline void __elv_rqhash_del(struct request *rq)
330{
331 hlist_del_init(&rq->hash);
332}
333
Jens Axboe165125e2007-07-24 09:28:11 +0200334static void elv_rqhash_del(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200335{
336 if (ELV_ON_HASH(rq))
337 __elv_rqhash_del(rq);
338}
339
Jens Axboe165125e2007-07-24 09:28:11 +0200340static void elv_rqhash_add(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200341{
Jens Axboeb374d182008-10-31 10:05:07 +0100342 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200343
344 BUG_ON(ELV_ON_HASH(rq));
345 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
346}
347
Jens Axboe165125e2007-07-24 09:28:11 +0200348static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
Jens Axboe98170642006-07-28 09:23:08 +0200349{
350 __elv_rqhash_del(rq);
351 elv_rqhash_add(q, rq);
352}
353
Jens Axboe165125e2007-07-24 09:28:11 +0200354static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
Jens Axboe98170642006-07-28 09:23:08 +0200355{
Jens Axboeb374d182008-10-31 10:05:07 +0100356 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200357 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
358 struct hlist_node *entry, *next;
359 struct request *rq;
360
361 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
362 BUG_ON(!ELV_ON_HASH(rq));
363
364 if (unlikely(!rq_mergeable(rq))) {
365 __elv_rqhash_del(rq);
366 continue;
367 }
368
369 if (rq_hash_key(rq) == offset)
370 return rq;
371 }
372
373 return NULL;
374}
375
Tejun Heo8922e162005-10-20 16:23:44 +0200376/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200377 * RB-tree support functions for inserting/lookup/removal of requests
378 * in a sorted RB tree.
379 */
380struct request *elv_rb_add(struct rb_root *root, struct request *rq)
381{
382 struct rb_node **p = &root->rb_node;
383 struct rb_node *parent = NULL;
384 struct request *__rq;
385
386 while (*p) {
387 parent = *p;
388 __rq = rb_entry(parent, struct request, rb_node);
389
390 if (rq->sector < __rq->sector)
391 p = &(*p)->rb_left;
392 else if (rq->sector > __rq->sector)
393 p = &(*p)->rb_right;
394 else
395 return __rq;
396 }
397
398 rb_link_node(&rq->rb_node, parent, p);
399 rb_insert_color(&rq->rb_node, root);
400 return NULL;
401}
Jens Axboe2e662b62006-07-13 11:55:04 +0200402EXPORT_SYMBOL(elv_rb_add);
403
404void elv_rb_del(struct rb_root *root, struct request *rq)
405{
406 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
407 rb_erase(&rq->rb_node, root);
408 RB_CLEAR_NODE(&rq->rb_node);
409}
Jens Axboe2e662b62006-07-13 11:55:04 +0200410EXPORT_SYMBOL(elv_rb_del);
411
412struct request *elv_rb_find(struct rb_root *root, sector_t sector)
413{
414 struct rb_node *n = root->rb_node;
415 struct request *rq;
416
417 while (n) {
418 rq = rb_entry(n, struct request, rb_node);
419
420 if (sector < rq->sector)
421 n = n->rb_left;
422 else if (sector > rq->sector)
423 n = n->rb_right;
424 else
425 return rq;
426 }
427
428 return NULL;
429}
Jens Axboe2e662b62006-07-13 11:55:04 +0200430EXPORT_SYMBOL(elv_rb_find);
431
432/*
Tejun Heo8922e162005-10-20 16:23:44 +0200433 * Insert rq into dispatch queue of q. Queue lock must be held on
Uwe Kleine-Königdbe7f762007-10-20 01:55:04 +0200434 * entry. rq is sort instead into the dispatch queue. To be used by
Jens Axboe2e662b62006-07-13 11:55:04 +0200435 * specific elevators.
Tejun Heo8922e162005-10-20 16:23:44 +0200436 */
Jens Axboe165125e2007-07-24 09:28:11 +0200437void elv_dispatch_sort(struct request_queue *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200438{
439 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200440 struct list_head *entry;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100441 int stop_flags;
Tejun Heo8922e162005-10-20 16:23:44 +0200442
Tejun Heo06b86242005-10-20 16:46:23 +0200443 if (q->last_merge == rq)
444 q->last_merge = NULL;
Jens Axboe98170642006-07-28 09:23:08 +0200445
446 elv_rqhash_del(q, rq);
447
Tejun Heo15853af2005-11-10 08:52:05 +0100448 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200449
Jens Axboe1b47f532005-10-20 16:37:00 +0200450 boundary = q->end_sector;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100451 stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
Tejun Heo8922e162005-10-20 16:23:44 +0200452 list_for_each_prev(entry, &q->queue_head) {
453 struct request *pos = list_entry_rq(entry);
454
David Woodhousee17fc0a2008-08-09 16:42:20 +0100455 if (blk_discard_rq(rq) != blk_discard_rq(pos))
456 break;
Jens Axboe783660b2007-01-19 11:27:47 +1100457 if (rq_data_dir(rq) != rq_data_dir(pos))
458 break;
Jens Axboe4eb166d2008-02-01 00:37:27 +0100459 if (pos->cmd_flags & stop_flags)
Tejun Heo8922e162005-10-20 16:23:44 +0200460 break;
461 if (rq->sector >= boundary) {
462 if (pos->sector < boundary)
463 continue;
464 } else {
465 if (pos->sector >= boundary)
466 break;
467 }
468 if (rq->sector >= pos->sector)
469 break;
470 }
471
472 list_add(&rq->queuelist, entry);
473}
Jens Axboe2e662b62006-07-13 11:55:04 +0200474EXPORT_SYMBOL(elv_dispatch_sort);
475
Jens Axboe98170642006-07-28 09:23:08 +0200476/*
Jens Axboe2e662b62006-07-13 11:55:04 +0200477 * Insert rq into dispatch queue of q. Queue lock must be held on
478 * entry. rq is added to the back of the dispatch queue. To be used by
479 * specific elevators.
Jens Axboe98170642006-07-28 09:23:08 +0200480 */
481void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
482{
483 if (q->last_merge == rq)
484 q->last_merge = NULL;
485
486 elv_rqhash_del(q, rq);
487
488 q->nr_sorted--;
489
490 q->end_sector = rq_end_sector(rq);
491 q->boundary_rq = rq;
492 list_add_tail(&rq->queuelist, &q->queue_head);
493}
Jens Axboe2e662b62006-07-13 11:55:04 +0200494EXPORT_SYMBOL(elv_dispatch_add_tail);
495
Jens Axboe165125e2007-07-24 09:28:11 +0200496int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Jens Axboeb374d182008-10-31 10:05:07 +0100498 struct elevator_queue *e = q->elevator;
Jens Axboe98170642006-07-28 09:23:08 +0200499 struct request *__rq;
Tejun Heo06b86242005-10-20 16:46:23 +0200500 int ret;
501
Jens Axboe98170642006-07-28 09:23:08 +0200502 /*
503 * First try one-hit cache.
504 */
Tejun Heo06b86242005-10-20 16:46:23 +0200505 if (q->last_merge) {
506 ret = elv_try_merge(q->last_merge, bio);
507 if (ret != ELEVATOR_NO_MERGE) {
508 *req = q->last_merge;
509 return ret;
510 }
511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Alan D. Brunelleac9fafa2008-04-29 14:44:19 +0200513 if (blk_queue_nomerges(q))
514 return ELEVATOR_NO_MERGE;
515
Jens Axboe98170642006-07-28 09:23:08 +0200516 /*
517 * See if our hash lookup can find a potential backmerge.
518 */
519 __rq = elv_rqhash_find(q, bio->bi_sector);
520 if (__rq && elv_rq_merge_ok(__rq, bio)) {
521 *req = __rq;
522 return ELEVATOR_BACK_MERGE;
523 }
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (e->ops->elevator_merge_fn)
526 return e->ops->elevator_merge_fn(q, req, bio);
527
528 return ELEVATOR_NO_MERGE;
529}
530
Jens Axboe165125e2007-07-24 09:28:11 +0200531void elv_merged_request(struct request_queue *q, struct request *rq, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532{
Jens Axboeb374d182008-10-31 10:05:07 +0100533 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 if (e->ops->elevator_merged_fn)
Jens Axboe2e662b62006-07-13 11:55:04 +0200536 e->ops->elevator_merged_fn(q, rq, type);
Tejun Heo06b86242005-10-20 16:46:23 +0200537
Jens Axboe2e662b62006-07-13 11:55:04 +0200538 if (type == ELEVATOR_BACK_MERGE)
539 elv_rqhash_reposition(q, rq);
Jens Axboe98170642006-07-28 09:23:08 +0200540
Tejun Heo06b86242005-10-20 16:46:23 +0200541 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
Jens Axboe165125e2007-07-24 09:28:11 +0200544void elv_merge_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 struct request *next)
546{
Jens Axboeb374d182008-10-31 10:05:07 +0100547 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (e->ops->elevator_merge_req_fn)
550 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo06b86242005-10-20 16:46:23 +0200551
Jens Axboe98170642006-07-28 09:23:08 +0200552 elv_rqhash_reposition(q, rq);
553 elv_rqhash_del(q, next);
554
555 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200556 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Jens Axboe165125e2007-07-24 09:28:11 +0200559void elv_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 /*
562 * it already went through dequeue, we need to decrement the
563 * in_flight count again
564 */
Tejun Heo8922e162005-10-20 16:23:44 +0200565 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 q->in_flight--;
Jens Axboecad97512007-01-14 22:26:09 +1100567 if (blk_sorted_rq(rq))
568 elv_deactivate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jens Axboe4aff5e22006-08-10 08:44:47 +0200571 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Tejun Heo30e96562006-02-08 01:01:31 -0800573 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Jens Axboe165125e2007-07-24 09:28:11 +0200576static void elv_drain_elevator(struct request_queue *q)
Tejun Heo15853af2005-11-10 08:52:05 +0100577{
578 static int printed;
579 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
580 ;
581 if (q->nr_sorted == 0)
582 return;
583 if (printed++ < 10) {
584 printk(KERN_ERR "%s: forced dispatching is broken "
585 "(nr_sorted=%u), please report this\n",
586 q->elevator->elevator_type->elevator_name, q->nr_sorted);
587 }
588}
589
Jens Axboe165125e2007-07-24 09:28:11 +0200590void elv_insert(struct request_queue *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Tejun Heo797e7db2006-01-06 09:51:03 +0100592 struct list_head *pos;
593 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200594 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100595
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100596 trace_block_rq_insert(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 rq->q = q;
599
Tejun Heo8922e162005-10-20 16:23:44 +0200600 switch (where) {
601 case ELEVATOR_INSERT_FRONT:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200602 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo8922e162005-10-20 16:23:44 +0200603
604 list_add(&rq->queuelist, &q->queue_head);
605 break;
606
607 case ELEVATOR_INSERT_BACK:
Jens Axboe4aff5e22006-08-10 08:44:47 +0200608 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100609 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200610 list_add_tail(&rq->queuelist, &q->queue_head);
611 /*
612 * We kick the queue here for the following reasons.
613 * - The elevator might have returned NULL previously
614 * to delay requests and returned them now. As the
615 * queue wasn't empty before this request, ll_rw_blk
616 * won't run the queue on return, resulting in hang.
617 * - Usually, back inserted requests won't be merged
618 * with anything. There's no point in delaying queue
619 * processing.
620 */
621 blk_remove_plug(q);
Jens Axboe80a4b582008-10-14 09:51:06 +0200622 blk_start_queueing(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200623 break;
624
625 case ELEVATOR_INSERT_SORT:
David Woodhousee17fc0a2008-08-09 16:42:20 +0100626 BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
Jens Axboe4aff5e22006-08-10 08:44:47 +0200627 rq->cmd_flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100628 q->nr_sorted++;
Jens Axboe98170642006-07-28 09:23:08 +0200629 if (rq_mergeable(rq)) {
630 elv_rqhash_add(q, rq);
631 if (!q->last_merge)
632 q->last_merge = rq;
633 }
634
Tejun Heoca235092005-11-01 17:23:49 +0900635 /*
636 * Some ioscheds (cfq) run q->request_fn directly, so
637 * rq cannot be accessed after calling
638 * elevator_add_req_fn.
639 */
640 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200641 break;
642
Tejun Heo797e7db2006-01-06 09:51:03 +0100643 case ELEVATOR_INSERT_REQUEUE:
644 /*
645 * If ordered flush isn't in progress, we do front
646 * insertion; otherwise, requests should be requeued
647 * in ordseq order.
648 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200649 rq->cmd_flags |= REQ_SOFTBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100650
Linas Vepstas95543172007-01-23 19:40:54 +0100651 /*
652 * Most requeues happen because of a busy condition,
653 * don't force unplug of the queue for that case.
654 */
655 unplug_it = 0;
656
Tejun Heo797e7db2006-01-06 09:51:03 +0100657 if (q->ordseq == 0) {
658 list_add(&rq->queuelist, &q->queue_head);
659 break;
660 }
661
662 ordseq = blk_ordered_req_seq(rq);
663
664 list_for_each(pos, &q->queue_head) {
665 struct request *pos_rq = list_entry_rq(pos);
666 if (ordseq <= blk_ordered_req_seq(pos_rq))
667 break;
668 }
669
670 list_add_tail(&rq->queuelist, pos);
671 break;
672
Tejun Heo8922e162005-10-20 16:23:44 +0200673 default:
674 printk(KERN_ERR "%s: bad insertion point %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700675 __func__, where);
Tejun Heo8922e162005-10-20 16:23:44 +0200676 BUG();
677 }
678
Jens Axboedac07ec2006-05-11 08:20:16 +0200679 if (unplug_it && blk_queue_plugged(q)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200680 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
681 - q->in_flight;
682
683 if (nrq >= q->unplug_thresh)
684 __generic_unplug_device(q);
685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686}
687
Jens Axboe165125e2007-07-24 09:28:11 +0200688void __elv_add_request(struct request_queue *q, struct request *rq, int where,
Tejun Heo30e96562006-02-08 01:01:31 -0800689 int plug)
690{
691 if (q->ordcolor)
Jens Axboe4aff5e22006-08-10 08:44:47 +0200692 rq->cmd_flags |= REQ_ORDERED_COLOR;
Tejun Heo30e96562006-02-08 01:01:31 -0800693
Jens Axboe4aff5e22006-08-10 08:44:47 +0200694 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800695 /*
696 * toggle ordered color
697 */
698 if (blk_barrier_rq(rq))
699 q->ordcolor ^= 1;
700
701 /*
702 * barriers implicitly indicate back insertion
703 */
704 if (where == ELEVATOR_INSERT_SORT)
705 where = ELEVATOR_INSERT_BACK;
706
707 /*
708 * this request is scheduling boundary, update
709 * end_sector
710 */
David Woodhousee17fc0a2008-08-09 16:42:20 +0100711 if (blk_fs_request(rq) || blk_discard_rq(rq)) {
Tejun Heo30e96562006-02-08 01:01:31 -0800712 q->end_sector = rq_end_sector(rq);
713 q->boundary_rq = rq;
714 }
Jens Axboe4eb166d2008-02-01 00:37:27 +0100715 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
716 where == ELEVATOR_INSERT_SORT)
Tejun Heo30e96562006-02-08 01:01:31 -0800717 where = ELEVATOR_INSERT_BACK;
718
719 if (plug)
720 blk_plug_device(q);
721
722 elv_insert(q, rq, where);
723}
Jens Axboe2e662b62006-07-13 11:55:04 +0200724EXPORT_SYMBOL(__elv_add_request);
725
Jens Axboe165125e2007-07-24 09:28:11 +0200726void elv_add_request(struct request_queue *q, struct request *rq, int where,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 int plug)
728{
729 unsigned long flags;
730
731 spin_lock_irqsave(q->queue_lock, flags);
732 __elv_add_request(q, rq, where, plug);
733 spin_unlock_irqrestore(q->queue_lock, flags);
734}
Jens Axboe2e662b62006-07-13 11:55:04 +0200735EXPORT_SYMBOL(elv_add_request);
736
Jens Axboe165125e2007-07-24 09:28:11 +0200737static inline struct request *__elv_next_request(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
Tejun Heo8922e162005-10-20 16:23:44 +0200739 struct request *rq;
740
Tejun Heo797e7db2006-01-06 09:51:03 +0100741 while (1) {
742 while (!list_empty(&q->queue_head)) {
743 rq = list_entry_rq(q->queue_head.next);
744 if (blk_do_ordered(q, &rq))
745 return rq;
746 }
Tejun Heo8922e162005-10-20 16:23:44 +0200747
Tejun Heo797e7db2006-01-06 09:51:03 +0100748 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
749 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
Jens Axboe165125e2007-07-24 09:28:11 +0200753struct request *elv_next_request(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 struct request *rq;
756 int ret;
757
758 while ((rq = __elv_next_request(q)) != NULL) {
Jens Axboe4aff5e22006-08-10 08:44:47 +0200759 if (!(rq->cmd_flags & REQ_STARTED)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200760 /*
761 * This is the first time the device driver
762 * sees this request (possibly after
763 * requeueing). Notify IO scheduler.
764 */
Jens Axboecad97512007-01-14 22:26:09 +1100765 if (blk_sorted_rq(rq))
766 elv_activate_rq(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200767
768 /*
769 * just mark as started even if we don't start
770 * it, a request that has been delayed should
771 * not be passed by new incoming requests
772 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200773 rq->cmd_flags |= REQ_STARTED;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100774 trace_block_rq_issue(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Tejun Heo8922e162005-10-20 16:23:44 +0200777 if (!q->boundary_rq || q->boundary_rq == rq) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200778 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200779 q->boundary_rq = NULL;
780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
James Bottomleyfa0ccd82008-01-10 11:30:36 -0600782 if (rq->cmd_flags & REQ_DONTPREP)
783 break;
784
785 if (q->dma_drain_size && rq->data_len) {
786 /*
787 * make sure space for the drain appears we
788 * know we can do this because max_hw_segments
789 * has been adjusted to be one fewer than the
790 * device can handle
791 */
792 rq->nr_phys_segments++;
James Bottomleyfa0ccd82008-01-10 11:30:36 -0600793 }
794
795 if (!q->prep_rq_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 break;
797
798 ret = q->prep_rq_fn(q, rq);
799 if (ret == BLKPREP_OK) {
800 break;
801 } else if (ret == BLKPREP_DEFER) {
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500802 /*
803 * the request may have been (partially) prepped.
804 * we need to keep this request in the front to
Tejun Heo8922e162005-10-20 16:23:44 +0200805 * avoid resource deadlock. REQ_STARTED will
806 * prevent other fs requests from passing this one.
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500807 */
James Bottomleyfa0ccd82008-01-10 11:30:36 -0600808 if (q->dma_drain_size && rq->data_len &&
809 !(rq->cmd_flags & REQ_DONTPREP)) {
810 /*
811 * remove the space for the drain we added
812 * so that we don't add it again
813 */
814 --rq->nr_phys_segments;
James Bottomleyfa0ccd82008-01-10 11:30:36 -0600815 }
816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 rq = NULL;
818 break;
819 } else if (ret == BLKPREP_KILL) {
Jens Axboe4aff5e22006-08-10 08:44:47 +0200820 rq->cmd_flags |= REQ_QUIET;
Kiyoshi Ueda99cd3382008-10-01 10:13:44 -0400821 __blk_end_request(rq, -EIO, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 } else {
Harvey Harrison24c03d42008-05-01 04:35:17 -0700823 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 break;
825 }
826 }
827
828 return rq;
829}
Jens Axboe2e662b62006-07-13 11:55:04 +0200830EXPORT_SYMBOL(elv_next_request);
831
Jens Axboe165125e2007-07-24 09:28:11 +0200832void elv_dequeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
Tejun Heo8922e162005-10-20 16:23:44 +0200834 BUG_ON(list_empty(&rq->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +0200835 BUG_ON(ELV_ON_HASH(rq));
Tejun Heo8922e162005-10-20 16:23:44 +0200836
837 list_del_init(&rq->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839 /*
840 * the time frame between a request being removed from the lists
841 * and to it is freed is accounted as io that is in progress at
Tejun Heo8922e162005-10-20 16:23:44 +0200842 * the driver side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
844 if (blk_account_rq(rq))
845 q->in_flight++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
Jens Axboe2e662b62006-07-13 11:55:04 +0200847
Jens Axboe165125e2007-07-24 09:28:11 +0200848int elv_queue_empty(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
Jens Axboeb374d182008-10-31 10:05:07 +0100850 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Tejun Heo8922e162005-10-20 16:23:44 +0200852 if (!list_empty(&q->queue_head))
853 return 0;
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (e->ops->elevator_queue_empty_fn)
856 return e->ops->elevator_queue_empty_fn(q);
857
Tejun Heo8922e162005-10-20 16:23:44 +0200858 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859}
Jens Axboe2e662b62006-07-13 11:55:04 +0200860EXPORT_SYMBOL(elv_queue_empty);
861
Jens Axboe165125e2007-07-24 09:28:11 +0200862struct request *elv_latter_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Jens Axboeb374d182008-10-31 10:05:07 +0100864 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
866 if (e->ops->elevator_latter_req_fn)
867 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 return NULL;
869}
870
Jens Axboe165125e2007-07-24 09:28:11 +0200871struct request *elv_former_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872{
Jens Axboeb374d182008-10-31 10:05:07 +0100873 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 if (e->ops->elevator_former_req_fn)
876 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return NULL;
878}
879
Jens Axboe165125e2007-07-24 09:28:11 +0200880int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
Jens Axboeb374d182008-10-31 10:05:07 +0100882 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 if (e->ops->elevator_set_req_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200885 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 rq->elevator_private = NULL;
888 return 0;
889}
890
Jens Axboe165125e2007-07-24 09:28:11 +0200891void elv_put_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Jens Axboeb374d182008-10-31 10:05:07 +0100893 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 if (e->ops->elevator_put_req_fn)
Jens Axboebb37b942006-12-01 10:42:33 +0100896 e->ops->elevator_put_req_fn(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Jens Axboe165125e2007-07-24 09:28:11 +0200899int elv_may_queue(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
Jens Axboeb374d182008-10-31 10:05:07 +0100901 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 if (e->ops->elevator_may_queue_fn)
Jens Axboecb78b282006-07-28 09:32:57 +0200904 return e->ops->elevator_may_queue_fn(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 return ELV_MQUEUE_MAY;
907}
908
Mike Anderson11914a52008-09-13 20:31:27 +0200909void elv_abort_queue(struct request_queue *q)
910{
911 struct request *rq;
912
913 while (!list_empty(&q->queue_head)) {
914 rq = list_entry_rq(q->queue_head.next);
915 rq->cmd_flags |= REQ_QUIET;
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100916 trace_block_rq_abort(q, rq);
Kiyoshi Ueda99cd3382008-10-01 10:13:44 -0400917 __blk_end_request(rq, -EIO, blk_rq_bytes(rq));
Mike Anderson11914a52008-09-13 20:31:27 +0200918 }
919}
920EXPORT_SYMBOL(elv_abort_queue);
921
Jens Axboe165125e2007-07-24 09:28:11 +0200922void elv_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923{
Jens Axboeb374d182008-10-31 10:05:07 +0100924 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 /*
927 * request is released from the driver, io must be done
928 */
Tejun Heo8922e162005-10-20 16:23:44 +0200929 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 q->in_flight--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100931 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
932 e->ops->elevator_completed_req_fn(q, rq);
933 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100934
Tejun Heo1bc691d2006-01-12 15:39:26 +0100935 /*
936 * Check if the queue is waiting for fs requests to be
937 * drained for flush sequence.
938 */
939 if (unlikely(q->ordseq)) {
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900940 struct request *next = NULL;
941
942 if (!list_empty(&q->queue_head))
943 next = list_entry_rq(q->queue_head.next);
944
945 if (!q->in_flight &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100946 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
Tejun Heo8f11b3e2008-11-28 13:32:05 +0900947 (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
Tejun Heo797e7db2006-01-06 09:51:03 +0100948 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
Jens Axboe80a4b582008-10-14 09:51:06 +0200949 blk_start_queueing(q);
Tejun Heo797e7db2006-01-06 09:51:03 +0100950 }
Tejun Heo8922e162005-10-20 16:23:44 +0200951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952}
953
Al Viro3d1ab402006-03-18 18:35:43 -0500954#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
955
956static ssize_t
957elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
958{
Al Viro3d1ab402006-03-18 18:35:43 -0500959 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100960 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500961 ssize_t error;
962
963 if (!entry->show)
964 return -EIO;
965
Jens Axboeb374d182008-10-31 10:05:07 +0100966 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500967 mutex_lock(&e->sysfs_lock);
968 error = e->ops ? entry->show(e, page) : -ENOENT;
969 mutex_unlock(&e->sysfs_lock);
970 return error;
971}
972
973static ssize_t
974elv_attr_store(struct kobject *kobj, struct attribute *attr,
975 const char *page, size_t length)
976{
Al Viro3d1ab402006-03-18 18:35:43 -0500977 struct elv_fs_entry *entry = to_elv(attr);
Jens Axboeb374d182008-10-31 10:05:07 +0100978 struct elevator_queue *e;
Al Viro3d1ab402006-03-18 18:35:43 -0500979 ssize_t error;
980
981 if (!entry->store)
982 return -EIO;
983
Jens Axboeb374d182008-10-31 10:05:07 +0100984 e = container_of(kobj, struct elevator_queue, kobj);
Al Viro3d1ab402006-03-18 18:35:43 -0500985 mutex_lock(&e->sysfs_lock);
986 error = e->ops ? entry->store(e, page, length) : -ENOENT;
987 mutex_unlock(&e->sysfs_lock);
988 return error;
989}
990
991static struct sysfs_ops elv_sysfs_ops = {
992 .show = elv_attr_show,
993 .store = elv_attr_store,
994};
995
996static struct kobj_type elv_ktype = {
997 .sysfs_ops = &elv_sysfs_ops,
998 .release = elevator_release,
999};
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001int elv_register_queue(struct request_queue *q)
1002{
Jens Axboeb374d182008-10-31 10:05:07 +01001003 struct elevator_queue *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -05001004 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07001006 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
Al Viro3d1ab402006-03-18 18:35:43 -05001007 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -05001008 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -05001009 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -05001010 while (attr->attr.name) {
1011 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -05001012 break;
Al Viroe572ec72006-03-18 22:27:18 -05001013 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -05001014 }
1015 }
1016 kobject_uevent(&e->kobj, KOBJ_ADD);
1017 }
1018 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019}
1020
Jens Axboeb374d182008-10-31 10:05:07 +01001021static void __elv_unregister_queue(struct elevator_queue *e)
Jens Axboebc1c1162006-06-08 08:49:06 +02001022{
1023 kobject_uevent(&e->kobj, KOBJ_REMOVE);
1024 kobject_del(&e->kobj);
1025}
1026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027void elv_unregister_queue(struct request_queue *q)
1028{
Jens Axboebc1c1162006-06-08 08:49:06 +02001029 if (q)
1030 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
Adrian Bunk2fdd82b2007-12-12 18:51:56 +01001033void elv_register(struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +01001035 char *def = "";
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001036
1037 spin_lock(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +01001038 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 list_add_tail(&e->list, &elv_list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001040 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Nate Diller5f003972006-01-24 10:07:58 +01001042 if (!strcmp(e->elevator_name, chosen_elevator) ||
1043 (!*chosen_elevator &&
1044 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
Thibaut VARENE1ffb96c2007-03-15 12:59:19 +01001045 def = " (default)";
1046
Jens Axboe4eb166d2008-02-01 00:37:27 +01001047 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
1048 def);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049}
1050EXPORT_SYMBOL_GPL(elv_register);
1051
1052void elv_unregister(struct elevator_type *e)
1053{
Christoph Hellwig83521d32005-10-30 15:01:39 -08001054 struct task_struct *g, *p;
1055
1056 /*
1057 * Iterate every thread in the process to remove the io contexts.
1058 */
Al Viroe17a9482006-03-18 13:21:20 -05001059 if (e->ops.trim) {
1060 read_lock(&tasklist_lock);
1061 do_each_thread(g, p) {
1062 task_lock(p);
Oleg Nesterov2d8f6132006-08-22 21:22:13 +04001063 if (p->io_context)
1064 e->ops.trim(p->io_context);
Al Viroe17a9482006-03-18 13:21:20 -05001065 task_unlock(p);
1066 } while_each_thread(g, p);
1067 read_unlock(&tasklist_lock);
1068 }
Christoph Hellwig83521d32005-10-30 15:01:39 -08001069
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001070 spin_lock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 list_del_init(&e->list);
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001072 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074EXPORT_SYMBOL_GPL(elv_unregister);
1075
1076/*
1077 * switch to new_e io scheduler. be careful not to introduce deadlocks -
1078 * we don't free the old io scheduler, before we have allocated what we
1079 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +02001080 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 */
Jens Axboe165125e2007-07-24 09:28:11 +02001082static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
Jens Axboeb374d182008-10-31 10:05:07 +01001084 struct elevator_queue *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +02001085 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Tejun Heocb98fc82005-10-28 08:29:39 +02001087 /*
1088 * Allocate new elevator
1089 */
Jens Axboeb5deef92006-07-19 23:39:40 +02001090 e = elevator_alloc(q, new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -05001092 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Jens Axboebc1c1162006-06-08 08:49:06 +02001094 data = elevator_init_queue(q, e);
1095 if (!data) {
1096 kobject_put(&e->kobj);
1097 return 0;
1098 }
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001101 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001103 spin_lock_irq(q->queue_lock);
1104
Nick Piggin75ad23b2008-04-29 14:48:33 +02001105 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001106
Tejun Heo15853af2005-11-10 08:52:05 +01001107 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001108
1109 while (q->rq.elvpriv) {
Jens Axboe80a4b582008-10-14 09:51:06 +02001110 blk_start_queueing(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001111 spin_unlock_irq(q->queue_lock);
Jens Axboe64521d12005-10-28 08:30:39 +02001112 msleep(10);
Tejun Heocb98fc82005-10-28 08:29:39 +02001113 spin_lock_irq(q->queue_lock);
Tejun Heo15853af2005-11-10 08:52:05 +01001114 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +02001115 }
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /*
Jens Axboebc1c1162006-06-08 08:49:06 +02001118 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 old_elevator = q->elevator;
1121
1122 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 * attach and start new elevator
1124 */
Jens Axboebc1c1162006-06-08 08:49:06 +02001125 elevator_attach(q, e, data);
1126
1127 spin_unlock_irq(q->queue_lock);
1128
1129 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
1131 if (elv_register_queue(q))
1132 goto fail_register;
1133
1134 /*
Tejun Heocb98fc82005-10-28 08:29:39 +02001135 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 */
1137 elevator_exit(old_elevator);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001138 spin_lock_irq(q->queue_lock);
1139 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1140 spin_unlock_irq(q->queue_lock);
1141
Alan D. Brunelle4722dc52008-05-27 14:55:00 +02001142 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1143
Al Viro3d1ab402006-03-18 18:35:43 -05001144 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146fail_register:
1147 /*
1148 * switch failed, exit the new io scheduler and reattach the old
1149 * one again (along with re-adding the sysfs dir)
1150 */
1151 elevator_exit(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 q->elevator = old_elevator;
1153 elv_register_queue(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +02001154
1155 spin_lock_irq(q->queue_lock);
1156 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1157 spin_unlock_irq(q->queue_lock);
1158
Al Viro3d1ab402006-03-18 18:35:43 -05001159 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160}
1161
Jens Axboe165125e2007-07-24 09:28:11 +02001162ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1163 size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164{
1165 char elevator_name[ELV_NAME_MAX];
1166 struct elevator_type *e;
1167
Li Zefanee2e9922008-10-14 08:49:56 +02001168 strlcpy(elevator_name, name, sizeof(elevator_name));
1169 strstrip(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 e = elevator_get(elevator_name);
1172 if (!e) {
1173 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1174 return -EINVAL;
1175 }
1176
Nate Diller2ca7d932005-10-30 15:02:24 -08001177 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1178 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -08001180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Al Viro3d1ab402006-03-18 18:35:43 -05001182 if (!elevator_switch(q, e))
Jens Axboe4eb166d2008-02-01 00:37:27 +01001183 printk(KERN_ERR "elevator: switch to %s failed\n",
1184 elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return count;
1186}
1187
Jens Axboe165125e2007-07-24 09:28:11 +02001188ssize_t elv_iosched_show(struct request_queue *q, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Jens Axboeb374d182008-10-31 10:05:07 +01001190 struct elevator_queue *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 struct elevator_type *elv = e->elevator_type;
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001192 struct elevator_type *__e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 int len = 0;
1194
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001195 spin_lock(&elv_list_lock);
Matthias Kaehlcke70cee262007-07-10 12:26:24 +02001196 list_for_each_entry(__e, &elv_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 if (!strcmp(elv->elevator_name, __e->elevator_name))
1198 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1199 else
1200 len += sprintf(name+len, "%s ", __e->elevator_name);
1201 }
Jens Axboe2a12dcd2007-04-26 14:41:53 +02001202 spin_unlock(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 len += sprintf(len+name, "\n");
1205 return len;
1206}
1207
Jens Axboe165125e2007-07-24 09:28:11 +02001208struct request *elv_rb_former_request(struct request_queue *q,
1209 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001210{
1211 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1212
1213 if (rbprev)
1214 return rb_entry_rq(rbprev);
1215
1216 return NULL;
1217}
Jens Axboe2e662b62006-07-13 11:55:04 +02001218EXPORT_SYMBOL(elv_rb_former_request);
1219
Jens Axboe165125e2007-07-24 09:28:11 +02001220struct request *elv_rb_latter_request(struct request_queue *q,
1221 struct request *rq)
Jens Axboe2e662b62006-07-13 11:55:04 +02001222{
1223 struct rb_node *rbnext = rb_next(&rq->rb_node);
1224
1225 if (rbnext)
1226 return rb_entry_rq(rbnext);
1227
1228 return NULL;
1229}
Jens Axboe2e662b62006-07-13 11:55:04 +02001230EXPORT_SYMBOL(elv_rb_latter_request);