blob: a0afdd317ceffb3d59f230c775da6c03740a426f [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 *
6 * 30042000 Jens Axboe <axboe@suse.de> :
7 *
8 * Split the elevator a bit so that it is possible to choose a different
9 * one or even write a new "plug in". There are three pieces:
10 * - elevator_fn, inserts a new request in the queue list
11 * - elevator_merge_fn, decides whether a new buffer can be merged with
12 * an existing request
13 * - elevator_dequeue_fn, called when a request is taken off the active list
14 *
15 * 20082000 Dave Jones <davej@suse.de> :
16 * Removed tests for max-bomb-segments, which was breaking elvtune
17 * when run without -bN
18 *
19 * Jens:
20 * - Rework again to work with bio instead of buffer_heads
21 * - loose bi_dev comparisons, partition handling is right now
22 * - completely modularize elevator setup and teardown
23 *
24 */
25#include <linux/kernel.h>
26#include <linux/fs.h>
27#include <linux/blkdev.h>
28#include <linux/elevator.h>
29#include <linux/bio.h>
30#include <linux/config.h>
31#include <linux/module.h>
32#include <linux/slab.h>
33#include <linux/init.h>
34#include <linux/compiler.h>
Tejun Heocb98fc82005-10-28 08:29:39 +020035#include <linux/delay.h>
Jens Axboe2056a782006-03-23 20:00:26 +010036#include <linux/blktrace_api.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include <asm/uaccess.h>
39
40static DEFINE_SPINLOCK(elv_list_lock);
41static LIST_HEAD(elv_list);
42
43/*
44 * can we safely merge with this request?
45 */
46inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
47{
48 if (!rq_mergeable(rq))
49 return 0;
50
51 /*
52 * different data direction or already started, don't merge
53 */
54 if (bio_data_dir(bio) != rq_data_dir(rq))
55 return 0;
56
57 /*
58 * same device and no special stuff set, merge is ok
59 */
60 if (rq->rq_disk == bio->bi_bdev->bd_disk &&
61 !rq->waiting && !rq->special)
62 return 1;
63
64 return 0;
65}
66EXPORT_SYMBOL(elv_rq_merge_ok);
67
Coywolf Qi Hunt769db452005-12-28 10:55:49 +010068static inline int elv_try_merge(struct request *__rq, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
70 int ret = ELEVATOR_NO_MERGE;
71
72 /*
73 * we can merge and sequence is ok, check if it's possible
74 */
75 if (elv_rq_merge_ok(__rq, bio)) {
76 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
77 ret = ELEVATOR_BACK_MERGE;
78 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
79 ret = ELEVATOR_FRONT_MERGE;
80 }
81
82 return ret;
83}
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085static struct elevator_type *elevator_find(const char *name)
86{
87 struct elevator_type *e = NULL;
88 struct list_head *entry;
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 list_for_each(entry, &elv_list) {
91 struct elevator_type *__e;
92
93 __e = list_entry(entry, struct elevator_type, list);
94
95 if (!strcmp(__e->elevator_name, name)) {
96 e = __e;
97 break;
98 }
99 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
101 return e;
102}
103
104static void elevator_put(struct elevator_type *e)
105{
106 module_put(e->elevator_owner);
107}
108
109static struct elevator_type *elevator_get(const char *name)
110{
Tejun Heo2824bc932005-10-20 10:56:41 +0200111 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Tejun Heo2824bc932005-10-20 10:56:41 +0200113 spin_lock_irq(&elv_list_lock);
114
115 e = elevator_find(name);
116 if (e && !try_module_get(e->elevator_owner))
117 e = NULL;
118
119 spin_unlock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 return e;
122}
123
Jens Axboebc1c1162006-06-08 08:49:06 +0200124static void *elevator_init_queue(request_queue_t *q, struct elevator_queue *eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
Jens Axboebc1c1162006-06-08 08:49:06 +0200126 return eq->ops->elevator_init_fn(q, eq);
127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Jens Axboebc1c1162006-06-08 08:49:06 +0200129static void elevator_attach(request_queue_t *q, struct elevator_queue *eq,
130 void *data)
131{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 q->elevator = eq;
Jens Axboebc1c1162006-06-08 08:49:06 +0200133 eq->elevator_data = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
136static char chosen_elevator[16];
137
Nate Diller5f003972006-01-24 10:07:58 +0100138static int __init elevator_setup(char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100140 /*
141 * Be backwards-compatible with previous kernels, so users
142 * won't get the wrong elevator.
143 */
Nate Diller5f003972006-01-24 10:07:58 +0100144 if (!strcmp(str, "as"))
Chuck Ebbert752a3b72006-01-16 09:47:37 +0100145 strcpy(chosen_elevator, "anticipatory");
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100146 else
Nate Diller5f003972006-01-24 10:07:58 +0100147 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800148 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151__setup("elevator=", elevator_setup);
152
Al Viro3d1ab402006-03-18 18:35:43 -0500153static struct kobj_type elv_ktype;
154
155static elevator_t *elevator_alloc(struct elevator_type *e)
156{
157 elevator_t *eq = kmalloc(sizeof(elevator_t), GFP_KERNEL);
158 if (eq) {
159 memset(eq, 0, sizeof(*eq));
160 eq->ops = &e->ops;
161 eq->elevator_type = e;
162 kobject_init(&eq->kobj);
163 snprintf(eq->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
164 eq->kobj.ktype = &elv_ktype;
165 mutex_init(&eq->sysfs_lock);
166 } else {
167 elevator_put(e);
168 }
169 return eq;
170}
171
172static void elevator_release(struct kobject *kobj)
173{
174 elevator_t *e = container_of(kobj, elevator_t, kobj);
175 elevator_put(e->elevator_type);
176 kfree(e);
177}
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179int elevator_init(request_queue_t *q, char *name)
180{
181 struct elevator_type *e = NULL;
182 struct elevator_queue *eq;
183 int ret = 0;
Jens Axboebc1c1162006-06-08 08:49:06 +0200184 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Tejun Heocb98fc82005-10-28 08:29:39 +0200186 INIT_LIST_HEAD(&q->queue_head);
187 q->last_merge = NULL;
188 q->end_sector = 0;
189 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200190
Nate Diller5f003972006-01-24 10:07:58 +0100191 if (name && !(e = elevator_get(name)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return -EINVAL;
193
Nate Diller248d5ca2006-01-24 10:09:14 +0100194 if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator)))
195 printk("I/O scheduler %s not found\n", chosen_elevator);
196
197 if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) {
198 printk("Default I/O scheduler not found, using no-op\n");
199 e = elevator_get("noop");
Nate Diller5f003972006-01-24 10:07:58 +0100200 }
201
Al Viro3d1ab402006-03-18 18:35:43 -0500202 eq = elevator_alloc(e);
203 if (!eq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Jens Axboebc1c1162006-06-08 08:49:06 +0200206 data = elevator_init_queue(q, eq);
207 if (!data) {
Al Viro3d1ab402006-03-18 18:35:43 -0500208 kobject_put(&eq->kobj);
Jens Axboebc1c1162006-06-08 08:49:06 +0200209 return -ENOMEM;
210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Jens Axboebc1c1162006-06-08 08:49:06 +0200212 elevator_attach(q, eq, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 return ret;
214}
215
216void elevator_exit(elevator_t *e)
217{
Al Viro3d1ab402006-03-18 18:35:43 -0500218 mutex_lock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 if (e->ops->elevator_exit_fn)
220 e->ops->elevator_exit_fn(e);
Al Viro3d1ab402006-03-18 18:35:43 -0500221 e->ops = NULL;
222 mutex_unlock(&e->sysfs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Al Viro3d1ab402006-03-18 18:35:43 -0500224 kobject_put(&e->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Tejun Heo8922e162005-10-20 16:23:44 +0200227/*
228 * Insert rq into dispatch queue of q. Queue lock must be held on
229 * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be
230 * appended to the dispatch queue. To be used by specific elevators.
231 */
Jens Axboe1b47f532005-10-20 16:37:00 +0200232void elv_dispatch_sort(request_queue_t *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200233{
234 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200235 struct list_head *entry;
236
Tejun Heo06b86242005-10-20 16:46:23 +0200237 if (q->last_merge == rq)
238 q->last_merge = NULL;
Tejun Heo15853af2005-11-10 08:52:05 +0100239 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200240
Jens Axboe1b47f532005-10-20 16:37:00 +0200241 boundary = q->end_sector;
Tejun Heocb198332005-10-24 08:35:58 +0200242
Tejun Heo8922e162005-10-20 16:23:44 +0200243 list_for_each_prev(entry, &q->queue_head) {
244 struct request *pos = list_entry_rq(entry);
245
246 if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
247 break;
248 if (rq->sector >= boundary) {
249 if (pos->sector < boundary)
250 continue;
251 } else {
252 if (pos->sector >= boundary)
253 break;
254 }
255 if (rq->sector >= pos->sector)
256 break;
257 }
258
259 list_add(&rq->queuelist, entry);
260}
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
263{
264 elevator_t *e = q->elevator;
Tejun Heo06b86242005-10-20 16:46:23 +0200265 int ret;
266
267 if (q->last_merge) {
268 ret = elv_try_merge(q->last_merge, bio);
269 if (ret != ELEVATOR_NO_MERGE) {
270 *req = q->last_merge;
271 return ret;
272 }
273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 if (e->ops->elevator_merge_fn)
276 return e->ops->elevator_merge_fn(q, req, bio);
277
278 return ELEVATOR_NO_MERGE;
279}
280
281void elv_merged_request(request_queue_t *q, struct request *rq)
282{
283 elevator_t *e = q->elevator;
284
285 if (e->ops->elevator_merged_fn)
286 e->ops->elevator_merged_fn(q, rq);
Tejun Heo06b86242005-10-20 16:46:23 +0200287
288 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291void elv_merge_requests(request_queue_t *q, struct request *rq,
292 struct request *next)
293{
294 elevator_t *e = q->elevator;
295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (e->ops->elevator_merge_req_fn)
297 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo15853af2005-11-10 08:52:05 +0100298 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200299
300 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Tejun Heo8922e162005-10-20 16:23:44 +0200303void elv_requeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 elevator_t *e = q->elevator;
306
307 /*
308 * it already went through dequeue, we need to decrement the
309 * in_flight count again
310 */
Tejun Heo8922e162005-10-20 16:23:44 +0200311 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 q->in_flight--;
Tejun Heo8922e162005-10-20 16:23:44 +0200313 if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn)
314 e->ops->elevator_deactivate_req_fn(q, rq);
315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 rq->flags &= ~REQ_STARTED;
318
Tejun Heo30e96562006-02-08 01:01:31 -0800319 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321
Tejun Heo15853af2005-11-10 08:52:05 +0100322static void elv_drain_elevator(request_queue_t *q)
323{
324 static int printed;
325 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
326 ;
327 if (q->nr_sorted == 0)
328 return;
329 if (printed++ < 10) {
330 printk(KERN_ERR "%s: forced dispatching is broken "
331 "(nr_sorted=%u), please report this\n",
332 q->elevator->elevator_type->elevator_name, q->nr_sorted);
333 }
334}
335
Tejun Heo30e96562006-02-08 01:01:31 -0800336void elv_insert(request_queue_t *q, struct request *rq, int where)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Tejun Heo797e7db2006-01-06 09:51:03 +0100338 struct list_head *pos;
339 unsigned ordseq;
Jens Axboedac07ec2006-05-11 08:20:16 +0200340 int unplug_it = 1;
Tejun Heo797e7db2006-01-06 09:51:03 +0100341
Jens Axboe2056a782006-03-23 20:00:26 +0100342 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 rq->q = q;
345
Tejun Heo8922e162005-10-20 16:23:44 +0200346 switch (where) {
347 case ELEVATOR_INSERT_FRONT:
348 rq->flags |= REQ_SOFTBARRIER;
349
350 list_add(&rq->queuelist, &q->queue_head);
351 break;
352
353 case ELEVATOR_INSERT_BACK:
354 rq->flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100355 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200356 list_add_tail(&rq->queuelist, &q->queue_head);
357 /*
358 * We kick the queue here for the following reasons.
359 * - The elevator might have returned NULL previously
360 * to delay requests and returned them now. As the
361 * queue wasn't empty before this request, ll_rw_blk
362 * won't run the queue on return, resulting in hang.
363 * - Usually, back inserted requests won't be merged
364 * with anything. There's no point in delaying queue
365 * processing.
366 */
367 blk_remove_plug(q);
368 q->request_fn(q);
369 break;
370
371 case ELEVATOR_INSERT_SORT:
372 BUG_ON(!blk_fs_request(rq));
373 rq->flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100374 q->nr_sorted++;
Tejun Heo06b86242005-10-20 16:46:23 +0200375 if (q->last_merge == NULL && rq_mergeable(rq))
376 q->last_merge = rq;
Tejun Heoca235092005-11-01 17:23:49 +0900377 /*
378 * Some ioscheds (cfq) run q->request_fn directly, so
379 * rq cannot be accessed after calling
380 * elevator_add_req_fn.
381 */
382 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200383 break;
384
Tejun Heo797e7db2006-01-06 09:51:03 +0100385 case ELEVATOR_INSERT_REQUEUE:
386 /*
387 * If ordered flush isn't in progress, we do front
388 * insertion; otherwise, requests should be requeued
389 * in ordseq order.
390 */
391 rq->flags |= REQ_SOFTBARRIER;
392
393 if (q->ordseq == 0) {
394 list_add(&rq->queuelist, &q->queue_head);
395 break;
396 }
397
398 ordseq = blk_ordered_req_seq(rq);
399
400 list_for_each(pos, &q->queue_head) {
401 struct request *pos_rq = list_entry_rq(pos);
402 if (ordseq <= blk_ordered_req_seq(pos_rq))
403 break;
404 }
405
406 list_add_tail(&rq->queuelist, pos);
Jens Axboedac07ec2006-05-11 08:20:16 +0200407 /*
408 * most requeues happen because of a busy condition, don't
409 * force unplug of the queue for that case.
410 */
411 unplug_it = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100412 break;
413
Tejun Heo8922e162005-10-20 16:23:44 +0200414 default:
415 printk(KERN_ERR "%s: bad insertion point %d\n",
416 __FUNCTION__, where);
417 BUG();
418 }
419
Jens Axboedac07ec2006-05-11 08:20:16 +0200420 if (unplug_it && blk_queue_plugged(q)) {
Tejun Heo8922e162005-10-20 16:23:44 +0200421 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
422 - q->in_flight;
423
424 if (nrq >= q->unplug_thresh)
425 __generic_unplug_device(q);
426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
Tejun Heo30e96562006-02-08 01:01:31 -0800429void __elv_add_request(request_queue_t *q, struct request *rq, int where,
430 int plug)
431{
432 if (q->ordcolor)
433 rq->flags |= REQ_ORDERED_COLOR;
434
435 if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
436 /*
437 * toggle ordered color
438 */
439 if (blk_barrier_rq(rq))
440 q->ordcolor ^= 1;
441
442 /*
443 * barriers implicitly indicate back insertion
444 */
445 if (where == ELEVATOR_INSERT_SORT)
446 where = ELEVATOR_INSERT_BACK;
447
448 /*
449 * this request is scheduling boundary, update
450 * end_sector
451 */
452 if (blk_fs_request(rq)) {
453 q->end_sector = rq_end_sector(rq);
454 q->boundary_rq = rq;
455 }
456 } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
457 where = ELEVATOR_INSERT_BACK;
458
459 if (plug)
460 blk_plug_device(q);
461
462 elv_insert(q, rq, where);
463}
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465void elv_add_request(request_queue_t *q, struct request *rq, int where,
466 int plug)
467{
468 unsigned long flags;
469
470 spin_lock_irqsave(q->queue_lock, flags);
471 __elv_add_request(q, rq, where, plug);
472 spin_unlock_irqrestore(q->queue_lock, flags);
473}
474
475static inline struct request *__elv_next_request(request_queue_t *q)
476{
Tejun Heo8922e162005-10-20 16:23:44 +0200477 struct request *rq;
478
Tejun Heo797e7db2006-01-06 09:51:03 +0100479 while (1) {
480 while (!list_empty(&q->queue_head)) {
481 rq = list_entry_rq(q->queue_head.next);
482 if (blk_do_ordered(q, &rq))
483 return rq;
484 }
Tejun Heo8922e162005-10-20 16:23:44 +0200485
Tejun Heo797e7db2006-01-06 09:51:03 +0100486 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
487 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490
491struct request *elv_next_request(request_queue_t *q)
492{
493 struct request *rq;
494 int ret;
495
496 while ((rq = __elv_next_request(q)) != NULL) {
Tejun Heo8922e162005-10-20 16:23:44 +0200497 if (!(rq->flags & REQ_STARTED)) {
498 elevator_t *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Tejun Heo8922e162005-10-20 16:23:44 +0200500 /*
501 * This is the first time the device driver
502 * sees this request (possibly after
503 * requeueing). Notify IO scheduler.
504 */
505 if (blk_sorted_rq(rq) &&
506 e->ops->elevator_activate_req_fn)
507 e->ops->elevator_activate_req_fn(q, rq);
508
509 /*
510 * just mark as started even if we don't start
511 * it, a request that has been delayed should
512 * not be passed by new incoming requests
513 */
514 rq->flags |= REQ_STARTED;
Jens Axboe2056a782006-03-23 20:00:26 +0100515 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
Tejun Heo8922e162005-10-20 16:23:44 +0200516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Tejun Heo8922e162005-10-20 16:23:44 +0200518 if (!q->boundary_rq || q->boundary_rq == rq) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200519 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200520 q->boundary_rq = NULL;
521 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn)
524 break;
525
526 ret = q->prep_rq_fn(q, rq);
527 if (ret == BLKPREP_OK) {
528 break;
529 } else if (ret == BLKPREP_DEFER) {
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500530 /*
531 * the request may have been (partially) prepped.
532 * we need to keep this request in the front to
Tejun Heo8922e162005-10-20 16:23:44 +0200533 * avoid resource deadlock. REQ_STARTED will
534 * prevent other fs requests from passing this one.
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500535 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 rq = NULL;
537 break;
538 } else if (ret == BLKPREP_KILL) {
539 int nr_bytes = rq->hard_nr_sectors << 9;
540
541 if (!nr_bytes)
542 nr_bytes = rq->data_len;
543
544 blkdev_dequeue_request(rq);
545 rq->flags |= REQ_QUIET;
546 end_that_request_chunk(rq, 0, nr_bytes);
Tejun Heo8ffdc652006-01-06 09:49:03 +0100547 end_that_request_last(rq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 } else {
549 printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
550 ret);
551 break;
552 }
553 }
554
555 return rq;
556}
557
Tejun Heo8922e162005-10-20 16:23:44 +0200558void elv_dequeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Tejun Heo8922e162005-10-20 16:23:44 +0200560 BUG_ON(list_empty(&rq->queuelist));
561
562 list_del_init(&rq->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /*
565 * the time frame between a request being removed from the lists
566 * and to it is freed is accounted as io that is in progress at
Tejun Heo8922e162005-10-20 16:23:44 +0200567 * the driver side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
569 if (blk_account_rq(rq))
570 q->in_flight++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
573int elv_queue_empty(request_queue_t *q)
574{
575 elevator_t *e = q->elevator;
576
Tejun Heo8922e162005-10-20 16:23:44 +0200577 if (!list_empty(&q->queue_head))
578 return 0;
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (e->ops->elevator_queue_empty_fn)
581 return e->ops->elevator_queue_empty_fn(q);
582
Tejun Heo8922e162005-10-20 16:23:44 +0200583 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
586struct request *elv_latter_request(request_queue_t *q, struct request *rq)
587{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 elevator_t *e = q->elevator;
589
590 if (e->ops->elevator_latter_req_fn)
591 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 return NULL;
593}
594
595struct request *elv_former_request(request_queue_t *q, struct request *rq)
596{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 elevator_t *e = q->elevator;
598
599 if (e->ops->elevator_former_req_fn)
600 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return NULL;
602}
603
Jens Axboe22e2c502005-06-27 10:55:12 +0200604int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -0400605 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
607 elevator_t *e = q->elevator;
608
609 if (e->ops->elevator_set_req_fn)
Jens Axboe22e2c502005-06-27 10:55:12 +0200610 return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 rq->elevator_private = NULL;
613 return 0;
614}
615
616void elv_put_request(request_queue_t *q, struct request *rq)
617{
618 elevator_t *e = q->elevator;
619
620 if (e->ops->elevator_put_req_fn)
621 e->ops->elevator_put_req_fn(q, rq);
622}
623
Jens Axboe22e2c502005-06-27 10:55:12 +0200624int elv_may_queue(request_queue_t *q, int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
626 elevator_t *e = q->elevator;
627
628 if (e->ops->elevator_may_queue_fn)
Jens Axboe22e2c502005-06-27 10:55:12 +0200629 return e->ops->elevator_may_queue_fn(q, rw, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
631 return ELV_MQUEUE_MAY;
632}
633
634void elv_completed_request(request_queue_t *q, struct request *rq)
635{
636 elevator_t *e = q->elevator;
637
638 /*
639 * request is released from the driver, io must be done
640 */
Tejun Heo8922e162005-10-20 16:23:44 +0200641 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 q->in_flight--;
Tejun Heo1bc691d2006-01-12 15:39:26 +0100643 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
644 e->ops->elevator_completed_req_fn(q, rq);
645 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100646
Tejun Heo1bc691d2006-01-12 15:39:26 +0100647 /*
648 * Check if the queue is waiting for fs requests to be
649 * drained for flush sequence.
650 */
651 if (unlikely(q->ordseq)) {
652 struct request *first_rq = list_entry_rq(q->queue_head.next);
653 if (q->in_flight == 0 &&
Tejun Heo797e7db2006-01-06 09:51:03 +0100654 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
655 blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
656 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
657 q->request_fn(q);
658 }
Tejun Heo8922e162005-10-20 16:23:44 +0200659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
Al Viro3d1ab402006-03-18 18:35:43 -0500662#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
663
664static ssize_t
665elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
666{
667 elevator_t *e = container_of(kobj, elevator_t, kobj);
668 struct elv_fs_entry *entry = to_elv(attr);
669 ssize_t error;
670
671 if (!entry->show)
672 return -EIO;
673
674 mutex_lock(&e->sysfs_lock);
675 error = e->ops ? entry->show(e, page) : -ENOENT;
676 mutex_unlock(&e->sysfs_lock);
677 return error;
678}
679
680static ssize_t
681elv_attr_store(struct kobject *kobj, struct attribute *attr,
682 const char *page, size_t length)
683{
684 elevator_t *e = container_of(kobj, elevator_t, kobj);
685 struct elv_fs_entry *entry = to_elv(attr);
686 ssize_t error;
687
688 if (!entry->store)
689 return -EIO;
690
691 mutex_lock(&e->sysfs_lock);
692 error = e->ops ? entry->store(e, page, length) : -ENOENT;
693 mutex_unlock(&e->sysfs_lock);
694 return error;
695}
696
697static struct sysfs_ops elv_sysfs_ops = {
698 .show = elv_attr_show,
699 .store = elv_attr_store,
700};
701
702static struct kobj_type elv_ktype = {
703 .sysfs_ops = &elv_sysfs_ops,
704 .release = elevator_release,
705};
706
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707int elv_register_queue(struct request_queue *q)
708{
709 elevator_t *e = q->elevator;
Al Viro3d1ab402006-03-18 18:35:43 -0500710 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Al Viro3d1ab402006-03-18 18:35:43 -0500712 e->kobj.parent = &q->kobj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Al Viro3d1ab402006-03-18 18:35:43 -0500714 error = kobject_add(&e->kobj);
715 if (!error) {
Al Viroe572ec72006-03-18 22:27:18 -0500716 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
Al Viro3d1ab402006-03-18 18:35:43 -0500717 if (attr) {
Al Viroe572ec72006-03-18 22:27:18 -0500718 while (attr->attr.name) {
719 if (sysfs_create_file(&e->kobj, &attr->attr))
Al Viro3d1ab402006-03-18 18:35:43 -0500720 break;
Al Viroe572ec72006-03-18 22:27:18 -0500721 attr++;
Al Viro3d1ab402006-03-18 18:35:43 -0500722 }
723 }
724 kobject_uevent(&e->kobj, KOBJ_ADD);
725 }
726 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
728
Jens Axboebc1c1162006-06-08 08:49:06 +0200729static void __elv_unregister_queue(elevator_t *e)
730{
731 kobject_uevent(&e->kobj, KOBJ_REMOVE);
732 kobject_del(&e->kobj);
733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735void elv_unregister_queue(struct request_queue *q)
736{
Jens Axboebc1c1162006-06-08 08:49:06 +0200737 if (q)
738 __elv_unregister_queue(q->elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
741int elv_register(struct elevator_type *e)
742{
Tejun Heo2824bc932005-10-20 10:56:41 +0200743 spin_lock_irq(&elv_list_lock);
Eric Sesterhennce524492006-03-24 18:43:26 +0100744 BUG_ON(elevator_find(e->elevator_name));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 list_add_tail(&e->list, &elv_list);
746 spin_unlock_irq(&elv_list_lock);
747
748 printk(KERN_INFO "io scheduler %s registered", e->elevator_name);
Nate Diller5f003972006-01-24 10:07:58 +0100749 if (!strcmp(e->elevator_name, chosen_elevator) ||
750 (!*chosen_elevator &&
751 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
752 printk(" (default)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 printk("\n");
754 return 0;
755}
756EXPORT_SYMBOL_GPL(elv_register);
757
758void elv_unregister(struct elevator_type *e)
759{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800760 struct task_struct *g, *p;
761
762 /*
763 * Iterate every thread in the process to remove the io contexts.
764 */
Al Viroe17a9482006-03-18 13:21:20 -0500765 if (e->ops.trim) {
766 read_lock(&tasklist_lock);
767 do_each_thread(g, p) {
768 task_lock(p);
769 e->ops.trim(p->io_context);
770 task_unlock(p);
771 } while_each_thread(g, p);
772 read_unlock(&tasklist_lock);
773 }
Christoph Hellwig83521d32005-10-30 15:01:39 -0800774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 spin_lock_irq(&elv_list_lock);
776 list_del_init(&e->list);
777 spin_unlock_irq(&elv_list_lock);
778}
779EXPORT_SYMBOL_GPL(elv_unregister);
780
781/*
782 * switch to new_e io scheduler. be careful not to introduce deadlocks -
783 * we don't free the old io scheduler, before we have allocated what we
784 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200785 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 */
Al Viro3d1ab402006-03-18 18:35:43 -0500787static int elevator_switch(request_queue_t *q, struct elevator_type *new_e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Tejun Heocb98fc82005-10-28 08:29:39 +0200789 elevator_t *old_elevator, *e;
Jens Axboebc1c1162006-06-08 08:49:06 +0200790 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Tejun Heocb98fc82005-10-28 08:29:39 +0200792 /*
793 * Allocate new elevator
794 */
Al Viro3d1ab402006-03-18 18:35:43 -0500795 e = elevator_alloc(new_e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 if (!e)
Al Viro3d1ab402006-03-18 18:35:43 -0500797 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Jens Axboebc1c1162006-06-08 08:49:06 +0200799 data = elevator_init_queue(q, e);
800 if (!data) {
801 kobject_put(&e->kobj);
802 return 0;
803 }
804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200806 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 */
Tejun Heocb98fc82005-10-28 08:29:39 +0200808 spin_lock_irq(q->queue_lock);
809
Jens Axboe64521d12005-10-28 08:30:39 +0200810 set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200811
Tejun Heo15853af2005-11-10 08:52:05 +0100812 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200813
814 while (q->rq.elvpriv) {
Tejun Heo407df2a2005-11-10 08:48:21 +0100815 blk_remove_plug(q);
816 q->request_fn(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200817 spin_unlock_irq(q->queue_lock);
Jens Axboe64521d12005-10-28 08:30:39 +0200818 msleep(10);
Tejun Heocb98fc82005-10-28 08:29:39 +0200819 spin_lock_irq(q->queue_lock);
Tejun Heo15853af2005-11-10 08:52:05 +0100820 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200821 }
822
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 /*
Jens Axboebc1c1162006-06-08 08:49:06 +0200824 * Remember old elevator.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 old_elevator = q->elevator;
827
828 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 * attach and start new elevator
830 */
Jens Axboebc1c1162006-06-08 08:49:06 +0200831 elevator_attach(q, e, data);
832
833 spin_unlock_irq(q->queue_lock);
834
835 __elv_unregister_queue(old_elevator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
837 if (elv_register_queue(q))
838 goto fail_register;
839
840 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200841 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 */
843 elevator_exit(old_elevator);
Jens Axboe64521d12005-10-28 08:30:39 +0200844 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -0500845 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847fail_register:
848 /*
849 * switch failed, exit the new io scheduler and reattach the old
850 * one again (along with re-adding the sysfs dir)
851 */
852 elevator_exit(e);
Tejun Heocb98fc82005-10-28 08:29:39 +0200853 e = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 q->elevator = old_elevator;
855 elv_register_queue(q);
Jens Axboe64521d12005-10-28 08:30:39 +0200856 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Al Viro3d1ab402006-03-18 18:35:43 -0500857 if (e)
858 kobject_put(&e->kobj);
859 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
862ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
863{
864 char elevator_name[ELV_NAME_MAX];
Tejun Heobe561232005-11-10 08:55:01 +0100865 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 struct elevator_type *e;
867
Tejun Heobe561232005-11-10 08:55:01 +0100868 elevator_name[sizeof(elevator_name) - 1] = '\0';
869 strncpy(elevator_name, name, sizeof(elevator_name) - 1);
870 len = strlen(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Tejun Heobe561232005-11-10 08:55:01 +0100872 if (len && elevator_name[len - 1] == '\n')
873 elevator_name[len - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 e = elevator_get(elevator_name);
876 if (!e) {
877 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
878 return -EINVAL;
879 }
880
Nate Diller2ca7d932005-10-30 15:02:24 -0800881 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
882 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -0800884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
Al Viro3d1ab402006-03-18 18:35:43 -0500886 if (!elevator_switch(q, e))
887 printk(KERN_ERR "elevator: switch to %s failed\n",elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 return count;
889}
890
891ssize_t elv_iosched_show(request_queue_t *q, char *name)
892{
893 elevator_t *e = q->elevator;
894 struct elevator_type *elv = e->elevator_type;
895 struct list_head *entry;
896 int len = 0;
897
898 spin_lock_irq(q->queue_lock);
899 list_for_each(entry, &elv_list) {
900 struct elevator_type *__e;
901
902 __e = list_entry(entry, struct elevator_type, list);
903 if (!strcmp(elv->elevator_name, __e->elevator_name))
904 len += sprintf(name+len, "[%s] ", elv->elevator_name);
905 else
906 len += sprintf(name+len, "%s ", __e->elevator_name);
907 }
908 spin_unlock_irq(q->queue_lock);
909
910 len += sprintf(len+name, "\n");
911 return len;
912}
913
Jens Axboe1b47f532005-10-20 16:37:00 +0200914EXPORT_SYMBOL(elv_dispatch_sort);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915EXPORT_SYMBOL(elv_add_request);
916EXPORT_SYMBOL(__elv_add_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917EXPORT_SYMBOL(elv_next_request);
Tejun Heo8922e162005-10-20 16:23:44 +0200918EXPORT_SYMBOL(elv_dequeue_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919EXPORT_SYMBOL(elv_queue_empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920EXPORT_SYMBOL(elevator_exit);
921EXPORT_SYMBOL(elevator_init);