blob: a824e49c0d0ac0567ad8294f6289544fa40553b6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Jens Axboe2056a782006-03-23 20:00:26 +010029#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080030#include <linux/fault-inject.h>
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010031#include <trace/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Jens Axboe8324aa92008-01-29 14:51:59 +010033#include "blk.h"
34
Ingo Molnar0bfc2452008-11-26 11:59:56 +010035DEFINE_TRACE(block_plug);
36DEFINE_TRACE(block_unplug_io);
37DEFINE_TRACE(block_unplug_timer);
38DEFINE_TRACE(block_getrq);
39DEFINE_TRACE(block_sleeprq);
40DEFINE_TRACE(block_rq_requeue);
41DEFINE_TRACE(block_bio_backmerge);
42DEFINE_TRACE(block_bio_frontmerge);
43DEFINE_TRACE(block_bio_queue);
44DEFINE_TRACE(block_rq_complete);
45DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */
46EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
47
Jens Axboe165125e2007-07-24 09:28:11 +020048static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * For the allocated request tables
52 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010053static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * For queue allocation
57 */
Jens Axboe6728cb02008-01-31 13:03:55 +010058struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * Controlling structure to kblockd
62 */
Jens Axboeff856ba2006-01-09 16:02:34 +010063static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jens Axboe26b82562008-01-29 13:54:41 +010065static void drive_stat_acct(struct request *rq, int new_io)
66{
Jens Axboe28f13702008-05-07 10:15:46 +020067 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010068 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090069 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010070
71 if (!blk_fs_request(rq) || !rq->rq_disk)
72 return;
73
Tejun Heo074a7ac2008-08-25 19:56:14 +090074 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +020075 part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
Tejun Heoc9959052008-08-25 19:47:21 +090076
Jens Axboe28f13702008-05-07 10:15:46 +020077 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090078 part_stat_inc(cpu, part, merges[rw]);
Jens Axboe28f13702008-05-07 10:15:46 +020079 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090080 part_round_stats(cpu, part);
81 part_inc_in_flight(part);
Jens Axboe26b82562008-01-29 13:54:41 +010082 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020083
Tejun Heo074a7ac2008-08-25 19:56:14 +090084 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010085}
86
Jens Axboe8324aa92008-01-29 14:51:59 +010087void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 int nr;
90
91 nr = q->nr_requests - (q->nr_requests / 8) + 1;
92 if (nr > q->nr_requests)
93 nr = q->nr_requests;
94 q->nr_congestion_on = nr;
95
96 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
97 if (nr < 1)
98 nr = 1;
99 q->nr_congestion_off = nr;
100}
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/**
103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
104 * @bdev: device
105 *
106 * Locates the passed device's request queue and returns the address of its
107 * backing_dev_info
108 *
109 * Will return NULL if the request queue cannot be located.
110 */
111struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
112{
113 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200114 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 if (q)
117 ret = &q->backing_dev_info;
118 return ret;
119}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120EXPORT_SYMBOL(blk_get_backing_dev_info);
121
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200122void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200124 memset(rq, 0, sizeof(*rq));
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700127 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200128 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100129 rq->q = q;
130 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200133 rq->cmd = rq->__cmd;
Jens Axboe63a71382008-02-08 12:41:03 +0100134 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200137EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
NeilBrown5bb23a62007-09-27 12:46:13 +0200139static void req_bio_endio(struct request *rq, struct bio *bio,
140 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100141{
Jens Axboe165125e2007-07-24 09:28:11 +0200142 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100143
NeilBrown5bb23a62007-09-27 12:46:13 +0200144 if (&q->bar_rq != rq) {
145 if (error)
146 clear_bit(BIO_UPTODATE, &bio->bi_flags);
147 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
148 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100149
NeilBrown5bb23a62007-09-27 12:46:13 +0200150 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100151 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700152 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200153 nbytes = bio->bi_size;
154 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100155
Keith Mannthey08bafc02008-11-25 10:24:35 +0100156 if (unlikely(rq->cmd_flags & REQ_QUIET))
157 set_bit(BIO_QUIET, &bio->bi_flags);
158
NeilBrown5bb23a62007-09-27 12:46:13 +0200159 bio->bi_size -= nbytes;
160 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200161
162 if (bio_integrity(bio))
163 bio_integrity_advance(bio, nbytes);
164
NeilBrown5bb23a62007-09-27 12:46:13 +0200165 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200166 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200167 } else {
168
169 /*
170 * Okay, this is the barrier request in progress, just
171 * record the error;
172 */
173 if (error && !q->orderr)
174 q->orderr = error;
175 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
Jens Axboe6728cb02008-01-31 13:03:55 +0100182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Jens Axboe6728cb02008-01-31 13:03:55 +0100186 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
187 (unsigned long long)rq->sector,
188 rq->nr_sectors,
189 rq->current_nr_sectors);
190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
191 rq->bio, rq->biotail,
192 rq->buffer, rq->data,
193 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Jens Axboe4aff5e22006-08-10 08:44:47 +0200195 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100196 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200197 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 printk("%02x ", rq->cmd[bit]);
199 printk("\n");
200 }
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_dump_rq_flags);
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204/*
205 * "plug" the device if there are no outstanding requests: this will
206 * force the transfer to start only after we have put all the requests
207 * on the list.
208 *
209 * This is called with interrupts off and no requests on the queue and
210 * with the queue lock held.
211 */
Jens Axboe165125e2007-07-24 09:28:11 +0200212void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 WARN_ON(!irqs_disabled());
215
216 /*
217 * don't plug a stopped queue, it must be paired with blk_start_queue()
218 * which will restart the queueing
219 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200220 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 return;
222
Jens Axboee48ec692008-07-03 13:18:54 +0200223 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100225 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228EXPORT_SYMBOL(blk_plug_device);
229
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200230/**
231 * blk_plug_device_unlocked - plug a device without queue lock held
232 * @q: The &struct request_queue to plug
233 *
234 * Description:
235 * Like @blk_plug_device(), but grabs the queue lock and disables
236 * interrupts.
237 **/
238void blk_plug_device_unlocked(struct request_queue *q)
239{
240 unsigned long flags;
241
242 spin_lock_irqsave(q->queue_lock, flags);
243 blk_plug_device(q);
244 spin_unlock_irqrestore(q->queue_lock, flags);
245}
246EXPORT_SYMBOL(blk_plug_device_unlocked);
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248/*
249 * remove the queue from the plugged list, if present. called with
250 * queue lock held and interrupts disabled.
251 */
Jens Axboe165125e2007-07-24 09:28:11 +0200252int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
254 WARN_ON(!irqs_disabled());
255
Jens Axboee48ec692008-07-03 13:18:54 +0200256 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return 0;
258
259 del_timer(&q->unplug_timer);
260 return 1;
261}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262EXPORT_SYMBOL(blk_remove_plug);
263
264/*
265 * remove the plug and let it rip..
266 */
Jens Axboe165125e2007-07-24 09:28:11 +0200267void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200269 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200271 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return;
273
Jens Axboe22e2c502005-06-27 10:55:12 +0200274 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277/**
278 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200279 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 *
281 * Description:
282 * Linux uses plugging to build bigger requests queues before letting
283 * the device have at them. If a queue is plugged, the I/O scheduler
284 * is still adding and merging requests on the queue. Once the queue
285 * gets unplugged, the request_fn defined for the queue is invoked and
286 * transfers started.
287 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200288void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200290 if (blk_queue_plugged(q)) {
291 spin_lock_irq(q->queue_lock);
292 __generic_unplug_device(q);
293 spin_unlock_irq(q->queue_lock);
294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296EXPORT_SYMBOL(generic_unplug_device);
297
298static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
299 struct page *page)
300{
Jens Axboe165125e2007-07-24 09:28:11 +0200301 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500303 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
Jens Axboe86db1e22008-01-29 14:53:40 +0100306void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Jens Axboe165125e2007-07-24 09:28:11 +0200308 struct request_queue *q =
309 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100311 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 q->unplug_fn(q);
313}
314
Jens Axboe86db1e22008-01-29 14:53:40 +0100315void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
Jens Axboe165125e2007-07-24 09:28:11 +0200317 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100319 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200320 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500323void blk_unplug(struct request_queue *q)
324{
325 /*
326 * devices don't necessarily have an ->unplug_fn defined
327 */
328 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100329 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500330 q->unplug_fn(q);
331 }
332}
333EXPORT_SYMBOL(blk_unplug);
334
Jens Axboec7c22e42008-09-13 20:26:01 +0200335static void blk_invoke_request_fn(struct request_queue *q)
336{
Jens Axboe80a4b582008-10-14 09:51:06 +0200337 if (unlikely(blk_queue_stopped(q)))
338 return;
339
Jens Axboec7c22e42008-09-13 20:26:01 +0200340 /*
341 * one level of recursion is ok and is much faster than kicking
342 * the unplug handling
343 */
344 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
345 q->request_fn(q);
346 queue_flag_clear(QUEUE_FLAG_REENTER, q);
347 } else {
348 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
349 kblockd_schedule_work(q, &q->unplug_work);
350 }
351}
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353/**
354 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200355 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 *
357 * Description:
358 * blk_start_queue() will clear the stop flag on the queue, and call
359 * the request_fn for the queue if it was in a stopped state when
360 * entered. Also see blk_stop_queue(). Queue lock must be held.
361 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200362void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200364 WARN_ON(!irqs_disabled());
365
Nick Piggin75ad23b2008-04-29 14:48:33 +0200366 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Jens Axboec7c22e42008-09-13 20:26:01 +0200367 blk_invoke_request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369EXPORT_SYMBOL(blk_start_queue);
370
371/**
372 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200373 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 *
375 * Description:
376 * The Linux block layer assumes that a block driver will consume all
377 * entries on the request queue when the request_fn strategy is called.
378 * Often this will not happen, because of hardware limitations (queue
379 * depth settings). If a device driver gets a 'queue full' response,
380 * or if it simply chooses not to queue more I/O at one point, it can
381 * call this function to prevent the request_fn from being called until
382 * the driver has signalled it's ready to go again. This happens by calling
383 * blk_start_queue() to restart queue operations. Queue lock must be held.
384 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200385void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
387 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200388 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389}
390EXPORT_SYMBOL(blk_stop_queue);
391
392/**
393 * blk_sync_queue - cancel any pending callbacks on a queue
394 * @q: the queue
395 *
396 * Description:
397 * The block layer may perform asynchronous callback activity
398 * on a queue, such as calling the unplug function after a timeout.
399 * A block device may call blk_sync_queue to ensure that any
400 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200401 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 * that its ->make_request_fn will not re-add plugging prior to calling
403 * this function.
404 *
405 */
406void blk_sync_queue(struct request_queue *q)
407{
408 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100409 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100410 cancel_work_sync(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
412EXPORT_SYMBOL(blk_sync_queue);
413
414/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200415 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200417 *
418 * Description:
419 * See @blk_run_queue. This variant must be called with the queue lock
420 * held and interrupts disabled.
421 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200423void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200426
427 /*
428 * Only recurse once to avoid overrunning the stack, let the unplug
429 * handling reinvoke the handler shortly if we already got there.
430 */
Jens Axboec7c22e42008-09-13 20:26:01 +0200431 if (!elv_queue_empty(q))
432 blk_invoke_request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200433}
434EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200435
Nick Piggin75ad23b2008-04-29 14:48:33 +0200436/**
437 * blk_run_queue - run a single device queue
438 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200439 *
440 * Description:
441 * Invoke request handling on this queue, if it has pending work to do.
442 * May be used to restart queueing when a request has completed. Also
443 * See @blk_start_queueing.
444 *
Nick Piggin75ad23b2008-04-29 14:48:33 +0200445 */
446void blk_run_queue(struct request_queue *q)
447{
448 unsigned long flags;
449
450 spin_lock_irqsave(q->queue_lock, flags);
451 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 spin_unlock_irqrestore(q->queue_lock, flags);
453}
454EXPORT_SYMBOL(blk_run_queue);
455
Jens Axboe165125e2007-07-24 09:28:11 +0200456void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500457{
458 kobject_put(&q->kobj);
459}
Al Viro483f4af2006-03-18 18:34:37 -0500460
Jens Axboe6728cb02008-01-31 13:03:55 +0100461void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500462{
Jens Axboee3335de2008-09-18 09:22:54 -0700463 /*
464 * We know we have process context here, so we can be a little
465 * cautious and ensure that pending block actions on this device
466 * are done before moving on. Going into this function, we should
467 * not have processes doing IO to this device.
468 */
469 blk_sync_queue(q);
470
Al Viro483f4af2006-03-18 18:34:37 -0500471 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200472 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500473 mutex_unlock(&q->sysfs_lock);
474
475 if (q->elevator)
476 elevator_exit(q->elevator);
477
478 blk_put_queue(q);
479}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480EXPORT_SYMBOL(blk_cleanup_queue);
481
Jens Axboe165125e2007-07-24 09:28:11 +0200482static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
484 struct request_list *rl = &q->rq;
485
486 rl->count[READ] = rl->count[WRITE] = 0;
487 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200488 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 init_waitqueue_head(&rl->wait[READ]);
490 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Christoph Lameter19460892005-06-23 00:08:19 -0700492 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
493 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495 if (!rl->rq_pool)
496 return -ENOMEM;
497
498 return 0;
499}
500
Jens Axboe165125e2007-07-24 09:28:11 +0200501struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Christoph Lameter19460892005-06-23 00:08:19 -0700503 return blk_alloc_queue_node(gfp_mask, -1);
504}
505EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Jens Axboe165125e2007-07-24 09:28:11 +0200507struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700508{
Jens Axboe165125e2007-07-24 09:28:11 +0200509 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700510 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700511
Jens Axboe8324aa92008-01-29 14:51:59 +0100512 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700513 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (!q)
515 return NULL;
516
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700517 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
518 q->backing_dev_info.unplug_io_data = q;
519 err = bdi_init(&q->backing_dev_info);
520 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100521 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700522 return NULL;
523 }
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700526 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
527 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200528 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500529
Jens Axboe8324aa92008-01-29 14:51:59 +0100530 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Al Viro483f4af2006-03-18 18:34:37 -0500532 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700533 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return q;
536}
Christoph Lameter19460892005-06-23 00:08:19 -0700537EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
539/**
540 * blk_init_queue - prepare a request queue for use with a block device
541 * @rfn: The function to be called to process requests that have been
542 * placed on the queue.
543 * @lock: Request queue spin lock
544 *
545 * Description:
546 * If a block device wishes to use the standard request handling procedures,
547 * which sorts requests and coalesces adjacent requests, then it must
548 * call blk_init_queue(). The function @rfn will be called when there
549 * are requests on the queue that need to be processed. If the device
550 * supports plugging, then @rfn may not be called immediately when requests
551 * are available on the queue, but may be called at some time later instead.
552 * Plugged queues are generally unplugged when a buffer belonging to one
553 * of the requests on the queue is needed, or due to memory pressure.
554 *
555 * @rfn is not required, or even expected, to remove all requests off the
556 * queue, but only as many as it can handle at a time. If it does leave
557 * requests on the queue, it is responsible for arranging that the requests
558 * get dealt with eventually.
559 *
560 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200561 * request queue; this lock will be taken also from interrupt context, so irq
562 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200564 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 * it didn't succeed.
566 *
567 * Note:
568 * blk_init_queue() must be paired with a blk_cleanup_queue() call
569 * when the block device is deactivated (such as at module unload).
570 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700571
Jens Axboe165125e2007-07-24 09:28:11 +0200572struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Christoph Lameter19460892005-06-23 00:08:19 -0700574 return blk_init_queue_node(rfn, lock, -1);
575}
576EXPORT_SYMBOL(blk_init_queue);
577
Jens Axboe165125e2007-07-24 09:28:11 +0200578struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700579blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
580{
Jens Axboe165125e2007-07-24 09:28:11 +0200581 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 if (!q)
584 return NULL;
585
Christoph Lameter19460892005-06-23 00:08:19 -0700586 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500587 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100588 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500589 return NULL;
590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
152587d2005-04-12 16:22:06 -0500592 /*
593 * if caller didn't supply a lock, they get per-queue locking with
594 * our embedded lock
595 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700596 if (!lock)
152587d2005-04-12 16:22:06 -0500597 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 q->prep_rq_fn = NULL;
601 q->unplug_fn = generic_unplug_device;
Kiyoshi Ueda4ee5eaf2008-09-18 10:46:13 -0400602 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER |
603 1 << QUEUE_FLAG_STACKABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 q->queue_lock = lock;
605
Milan Broz0e435ac2008-12-03 12:55:08 +0100606 blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 blk_queue_make_request(q, __make_request);
609 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
610
611 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
612 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
613
Alan Stern44ec9542007-02-20 11:01:57 -0500614 q->sg_reserved_size = INT_MAX;
615
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900616 blk_set_cmd_filter_defaults(&q->cmd_filter);
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 /*
619 * all done
620 */
621 if (!elevator_init(q, NULL)) {
622 blk_queue_congestion_threshold(q);
623 return q;
624 }
625
Al Viro8669aaf2006-03-18 13:50:00 -0500626 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return NULL;
628}
Christoph Lameter19460892005-06-23 00:08:19 -0700629EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Jens Axboe165125e2007-07-24 09:28:11 +0200631int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700633 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500634 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return 0;
636 }
637
638 return 1;
639}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Jens Axboe165125e2007-07-24 09:28:11 +0200641static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200643 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200644 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 mempool_free(rq, q->rq.rq_pool);
646}
647
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200648static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +0200649blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
652
653 if (!rq)
654 return NULL;
655
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200656 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200657
Jens Axboe49171e52006-08-10 08:59:11 +0200658 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Tejun Heocb98fc82005-10-28 08:29:39 +0200660 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200661 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200662 mempool_free(rq, q->rq.rq_pool);
663 return NULL;
664 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200665 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Tejun Heocb98fc82005-10-28 08:29:39 +0200668 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
671/*
672 * ioc_batching returns true if the ioc is a valid batching request and
673 * should be given priority access to a request.
674 */
Jens Axboe165125e2007-07-24 09:28:11 +0200675static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
677 if (!ioc)
678 return 0;
679
680 /*
681 * Make sure the process is able to allocate at least 1 request
682 * even if the batch times out, otherwise we could theoretically
683 * lose wakeups.
684 */
685 return ioc->nr_batch_requests == q->nr_batching ||
686 (ioc->nr_batch_requests > 0
687 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
688}
689
690/*
691 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
692 * will cause the process to be a "batcher" on all queues in the system. This
693 * is the behaviour we want though - once it gets a wakeup it should be given
694 * a nice run.
695 */
Jens Axboe165125e2007-07-24 09:28:11 +0200696static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
698 if (!ioc || ioc_batching(q, ioc))
699 return;
700
701 ioc->nr_batch_requests = q->nr_batching;
702 ioc->last_waited = jiffies;
703}
704
Jens Axboe165125e2007-07-24 09:28:11 +0200705static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 struct request_list *rl = &q->rq;
708
709 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -0700710 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 if (waitqueue_active(&rl->wait[rw]))
714 wake_up(&rl->wait[rw]);
715
716 blk_clear_queue_full(q, rw);
717 }
718}
719
720/*
721 * A request has just been released. Account for it, update the full and
722 * congestion status, wake up any waiters. Called under q->queue_lock.
723 */
Jens Axboe165125e2007-07-24 09:28:11 +0200724static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 struct request_list *rl = &q->rq;
727
728 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200729 if (priv)
730 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732 __freed_request(q, rw);
733
734 if (unlikely(rl->starved[rw ^ 1]))
735 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736}
737
738#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
739/*
Nick Piggind6344532005-06-28 20:45:14 -0700740 * Get a free request, queue_lock must be held.
741 * Returns NULL on failure, with queue_lock held.
742 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 */
Jens Axboe165125e2007-07-24 09:28:11 +0200744static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100745 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
747 struct request *rq = NULL;
748 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100749 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +0100750 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100751 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Jens Axboe7749a8d2006-12-13 13:02:26 +0100753 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100754 if (may_queue == ELV_MQUEUE_NO)
755 goto rq_starved;
756
757 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
758 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200759 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100760 /*
761 * The queue will fill after this allocation, so set
762 * it as full, and mark this process as "batching".
763 * This process will be allowed to complete a batch of
764 * requests, others will be blocked.
765 */
766 if (!blk_queue_full(q, rw)) {
767 ioc_set_batching(q, ioc);
768 blk_set_queue_full(q, rw);
769 } else {
770 if (may_queue != ELV_MQUEUE_MUST
771 && !ioc_batching(q, ioc)) {
772 /*
773 * The queue is full and the allocating
774 * process is not a "batcher", and not
775 * exempted by the IO scheduler
776 */
777 goto out;
778 }
779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
Thomas Maier79e2de42006-10-19 23:28:15 -0700781 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
783
Jens Axboe082cf692005-06-28 16:35:11 +0200784 /*
785 * Only allow batching queuers to allocate up to 50% over the defined
786 * limit of requests, otherwise we could have thousands of requests
787 * allocated with any setting of ->nr_requests
788 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100789 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200790 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 rl->count[rw]++;
793 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200794
Jens Axboe64521d12005-10-28 08:30:39 +0200795 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200796 if (priv)
797 rl->elvpriv++;
798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 spin_unlock_irq(q->queue_lock);
800
Jens Axboe7749a8d2006-12-13 13:02:26 +0100801 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100802 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 /*
804 * Allocation failed presumably due to memory. Undo anything
805 * we might have messed up.
806 *
807 * Allocating task should really be put onto the front of the
808 * wait queue, but this is pretty rare.
809 */
810 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +0200811 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 /*
814 * in the very unlikely event that allocation failed and no
815 * requests for this direction was pending, mark us starved
816 * so that freeing of a request in the other direction will
817 * notice us. another possible fix would be to split the
818 * rq mempool into READ and WRITE
819 */
820rq_starved:
821 if (unlikely(rl->count[rw] == 0))
822 rl->starved[rw] = 1;
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 goto out;
825 }
826
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100827 /*
828 * ioc may be NULL here, and ioc_batching will be false. That's
829 * OK, if the queue is under the request limit then requests need
830 * not count toward the nr_batch_requests limit. There will always
831 * be some limit enforced by BLK_BATCH_TIME.
832 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (ioc_batching(q, ioc))
834 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100835
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100836 trace_block_getrq(q, bio, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 return rq;
839}
840
841/*
842 * No available requests for this queue, unplug the device and wait for some
843 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700844 *
845 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 */
Jens Axboe165125e2007-07-24 09:28:11 +0200847static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200848 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
Jens Axboe7749a8d2006-12-13 13:02:26 +0100850 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 struct request *rq;
852
Jens Axboe7749a8d2006-12-13 13:02:26 +0100853 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700854 while (!rq) {
855 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200856 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 struct request_list *rl = &q->rq;
858
859 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
860 TASK_UNINTERRUPTIBLE);
861
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100862 trace_block_sleeprq(q, bio, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200864 __generic_unplug_device(q);
865 spin_unlock_irq(q->queue_lock);
866 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200868 /*
869 * After sleeping, we become a "batching" process and
870 * will be able to allocate at least one request, and
871 * up to a big batch of them for a small period time.
872 * See ioc_batching, ioc_set_batching
873 */
874 ioc = current_io_context(GFP_NOIO, q->node);
875 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100876
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200877 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 finish_wait(&rl->wait[rw], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200879
880 rq = get_request(q, rw_flags, bio, GFP_NOIO);
881 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
883 return rq;
884}
885
Jens Axboe165125e2007-07-24 09:28:11 +0200886struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
888 struct request *rq;
889
890 BUG_ON(rw != READ && rw != WRITE);
891
Nick Piggind6344532005-06-28 20:45:14 -0700892 spin_lock_irq(q->queue_lock);
893 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200894 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700895 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200896 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700897 if (!rq)
898 spin_unlock_irq(q->queue_lock);
899 }
900 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 return rq;
903}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904EXPORT_SYMBOL(blk_get_request);
905
906/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200907 * blk_start_queueing - initiate dispatch of requests to device
908 * @q: request queue to kick into gear
909 *
910 * This is basically a helper to remove the need to know whether a queue
911 * is plugged or not if someone just wants to initiate dispatch of requests
Jens Axboe80a4b582008-10-14 09:51:06 +0200912 * for this queue. Should be used to start queueing on a device outside
913 * of ->request_fn() context. Also see @blk_run_queue.
Jens Axboedc72ef42006-07-20 14:54:05 +0200914 *
915 * The queue lock must be held with interrupts disabled.
916 */
Jens Axboe165125e2007-07-24 09:28:11 +0200917void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200918{
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200919 if (!blk_queue_plugged(q)) {
920 if (unlikely(blk_queue_stopped(q)))
921 return;
Jens Axboedc72ef42006-07-20 14:54:05 +0200922 q->request_fn(q);
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200923 } else
Jens Axboedc72ef42006-07-20 14:54:05 +0200924 __generic_unplug_device(q);
925}
926EXPORT_SYMBOL(blk_start_queueing);
927
928/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 * blk_requeue_request - put a request back on queue
930 * @q: request queue where request should be inserted
931 * @rq: request to be inserted
932 *
933 * Description:
934 * Drivers often keep queueing requests until the hardware cannot accept
935 * more, when that condition happens we need to put the request back
936 * on the queue. Must be called with queue lock held.
937 */
Jens Axboe165125e2007-07-24 09:28:11 +0200938void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700940 blk_delete_timer(rq);
941 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100942 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (blk_rq_tagged(rq))
945 blk_queue_end_tag(q, rq);
946
947 elv_requeue_request(q, rq);
948}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949EXPORT_SYMBOL(blk_requeue_request);
950
951/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200952 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 * @q: request queue where request should be inserted
954 * @rq: request to be inserted
955 * @at_head: insert request at head or tail of queue
956 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 *
958 * Description:
959 * Many block devices need to execute commands asynchronously, so they don't
960 * block the whole kernel from preemption during request execution. This is
961 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200962 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
963 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 *
965 * We have the option of inserting the head or the tail of the queue.
966 * Typically we use the tail for new ioctls and so forth. We use the head
967 * of the queue for things like a QUEUE_FULL message from a device, or a
968 * host that is unable to accept a particular command.
969 */
Jens Axboe165125e2007-07-24 09:28:11 +0200970void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500971 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972{
Tejun Heo 867d1192005-04-24 02:06:05 -0500973 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 unsigned long flags;
975
976 /*
977 * tell I/O scheduler that this isn't a regular read/write (ie it
978 * must not attempt merges on this) and that it acts as a soft
979 * barrier
980 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200981 rq->cmd_type = REQ_TYPE_SPECIAL;
982 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 rq->special = data;
985
986 spin_lock_irqsave(q->queue_lock, flags);
987
988 /*
989 * If command is tagged, release the tag
990 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500991 if (blk_rq_tagged(rq))
992 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200994 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500995 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200996 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 spin_unlock_irqrestore(q->queue_lock, flags);
998}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999EXPORT_SYMBOL(blk_insert_request);
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001/*
1002 * add-request adds a request to the linked list.
1003 * queue lock is held and interrupts disabled, as we muck with the
1004 * request queue list.
1005 */
Jens Axboe6728cb02008-01-31 13:03:55 +01001006static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001008 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /*
1011 * elevator indicated where it wants this request to be
1012 * inserted at elevator_merge time
1013 */
1014 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1015}
Jens Axboe6728cb02008-01-31 13:03:55 +01001016
Tejun Heo074a7ac2008-08-25 19:56:14 +09001017static void part_round_stats_single(int cpu, struct hd_struct *part,
1018 unsigned long now)
1019{
1020 if (now == part->stamp)
1021 return;
1022
1023 if (part->in_flight) {
1024 __part_stat_add(cpu, part, time_in_queue,
1025 part->in_flight * (now - part->stamp));
1026 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1027 }
1028 part->stamp = now;
1029}
1030
1031/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001032 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1033 * @cpu: cpu number for stats access
1034 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 *
1036 * The average IO queue length and utilisation statistics are maintained
1037 * by observing the current state of the queue length and the amount of
1038 * time it has been in this state for.
1039 *
1040 * Normally, that accounting is done on IO completion, but that can result
1041 * in more than a second's worth of IO being accounted for within any one
1042 * second, leading to >100% utilisation. To deal with that, we call this
1043 * function to do a round-off before returning the results when reading
1044 * /proc/diskstats. This accounts immediately for all queue usage up to
1045 * the current jiffies and restarts the counters again.
1046 */
Tejun Heoc9959052008-08-25 19:47:21 +09001047void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001048{
1049 unsigned long now = jiffies;
1050
Tejun Heo074a7ac2008-08-25 19:56:14 +09001051 if (part->partno)
1052 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1053 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001054}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001055EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057/*
1058 * queue lock must be held
1059 */
Jens Axboe165125e2007-07-24 09:28:11 +02001060void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 if (unlikely(!q))
1063 return;
1064 if (unlikely(--req->ref_count))
1065 return;
1066
Tejun Heo8922e162005-10-20 16:23:44 +02001067 elv_completed_request(q, req);
1068
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 /*
1070 * Request may not have originated from ll_rw_blk. if not,
1071 * it didn't come out of our reserved rq pools
1072 */
Jens Axboe49171e52006-08-10 08:59:11 +02001073 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001075 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001078 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
1080 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02001081 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 }
1083}
Mike Christie6e39b692005-11-11 05:30:24 -06001084EXPORT_SYMBOL_GPL(__blk_put_request);
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086void blk_put_request(struct request *req)
1087{
Tejun Heo8922e162005-10-20 16:23:44 +02001088 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001089 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001091 spin_lock_irqsave(q->queue_lock, flags);
1092 __blk_put_request(q, req);
1093 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095EXPORT_SYMBOL(blk_put_request);
1096
Jens Axboe86db1e22008-01-29 14:53:40 +01001097void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001098{
Jens Axboec7c22e42008-09-13 20:26:01 +02001099 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001100 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001101
1102 /*
1103 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1104 */
Mike Christie6000a362008-08-19 18:45:30 -05001105 if (bio_rw_ahead(bio))
1106 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1107 REQ_FAILFAST_DRIVER);
1108 if (bio_failfast_dev(bio))
1109 req->cmd_flags |= REQ_FAILFAST_DEV;
1110 if (bio_failfast_transport(bio))
1111 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1112 if (bio_failfast_driver(bio))
1113 req->cmd_flags |= REQ_FAILFAST_DRIVER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001114
1115 /*
1116 * REQ_BARRIER implies no merging, but lets make it explicit
1117 */
David Woodhousefb2dce82008-08-05 18:01:53 +01001118 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001119 req->cmd_flags |= REQ_DISCARD;
1120 if (bio_barrier(bio))
1121 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001122 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001123 } else if (unlikely(bio_barrier(bio)))
1124 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001125
Jens Axboeb31dc662006-06-13 08:26:10 +02001126 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001127 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001128 if (bio_rw_meta(bio))
1129 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02001130
Tejun Heo52d9e672006-01-06 09:49:58 +01001131 req->errors = 0;
1132 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001133 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001134 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001135 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001136}
1137
Jens Axboe165125e2007-07-24 09:28:11 +02001138static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
Nick Piggin450991b2005-06-28 20:45:13 -07001140 struct request *req;
Tejun Heoa7384672008-11-28 13:32:03 +09001141 int el_ret, nr_sectors;
Jens Axboe51da90f2006-07-18 04:14:45 +02001142 const unsigned short prio = bio_prio(bio);
1143 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001144 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 /*
1149 * low level driver can indicate that it wants pages above a
1150 * certain limit bounced to low memory (ie for highmem, or even
1151 * ISA dma in theory)
1152 */
1153 blk_queue_bounce(q, &bio);
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 spin_lock_irq(q->queue_lock);
1156
Tejun Heoa7384672008-11-28 13:32:03 +09001157 if (unlikely(bio_barrier(bio)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 goto get_rq;
1159
1160 el_ret = elv_merge(q, &req, bio);
1161 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001162 case ELEVATOR_BACK_MERGE:
1163 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Jens Axboe6728cb02008-01-31 13:03:55 +01001165 if (!ll_back_merge_fn(q, req, bio))
1166 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001168 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001169
Jens Axboe6728cb02008-01-31 13:03:55 +01001170 req->biotail->bi_next = bio;
1171 req->biotail = bio;
1172 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1173 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001174 if (!blk_rq_cpu_valid(req))
1175 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001176 drive_stat_acct(req, 0);
1177 if (!attempt_back_merge(q, req))
1178 elv_merged_request(q, req, el_ret);
1179 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Jens Axboe6728cb02008-01-31 13:03:55 +01001181 case ELEVATOR_FRONT_MERGE:
1182 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Jens Axboe6728cb02008-01-31 13:03:55 +01001184 if (!ll_front_merge_fn(q, req, bio))
1185 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001187 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001188
Jens Axboe6728cb02008-01-31 13:03:55 +01001189 bio->bi_next = req->bio;
1190 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Jens Axboe6728cb02008-01-31 13:03:55 +01001192 /*
1193 * may not be valid. if the low level driver said
1194 * it didn't need a bounce buffer then it better
1195 * not touch req->buffer either...
1196 */
1197 req->buffer = bio_data(bio);
1198 req->current_nr_sectors = bio_cur_sectors(bio);
1199 req->hard_cur_sectors = req->current_nr_sectors;
1200 req->sector = req->hard_sector = bio->bi_sector;
1201 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1202 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001203 if (!blk_rq_cpu_valid(req))
1204 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001205 drive_stat_acct(req, 0);
1206 if (!attempt_front_merge(q, req))
1207 elv_merged_request(q, req, el_ret);
1208 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Jens Axboe6728cb02008-01-31 13:03:55 +01001210 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1211 default:
1212 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001216 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001217 * This sync check and mask will be re-done in init_request_from_bio(),
1218 * but we need to set it earlier to expose the sync flag to the
1219 * rq allocator and io schedulers.
1220 */
1221 rw_flags = bio_data_dir(bio);
1222 if (sync)
1223 rw_flags |= REQ_RW_SYNC;
1224
1225 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001226 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001227 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001228 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001229 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001230
Nick Piggin450991b2005-06-28 20:45:13 -07001231 /*
1232 * After dropping the lock and possibly sleeping here, our request
1233 * may now be mergeable after it had proven unmergeable (above).
1234 * We don't worry about that case for efficiency. It won't happen
1235 * often, and the elevators are able to handle it.
1236 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001237 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Nick Piggin450991b2005-06-28 20:45:13 -07001239 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001240 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1241 bio_flagged(bio, BIO_CPU_AFFINE))
1242 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboea31a9732008-10-17 13:58:29 +02001243 if (!blk_queue_nonrot(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001244 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 add_request(q, req);
1246out:
Jens Axboea31a9732008-10-17 13:58:29 +02001247 if (sync || blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 spin_unlock_irq(q->queue_lock);
1250 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251}
1252
1253/*
1254 * If bio->bi_dev is a partition, remap the location
1255 */
1256static inline void blk_partition_remap(struct bio *bio)
1257{
1258 struct block_device *bdev = bio->bi_bdev;
1259
Jens Axboebf2de6f2007-09-27 13:01:25 +02001260 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 bio->bi_sector += p->start_sect;
1264 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001265
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001266 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001267 bdev->bd_dev, bio->bi_sector,
1268 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
1270}
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272static void handle_bad_sector(struct bio *bio)
1273{
1274 char b[BDEVNAME_SIZE];
1275
1276 printk(KERN_INFO "attempt to access beyond end of device\n");
1277 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1278 bdevname(bio->bi_bdev, b),
1279 bio->bi_rw,
1280 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1281 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1282
1283 set_bit(BIO_EOF, &bio->bi_flags);
1284}
1285
Akinobu Mitac17bb492006-12-08 02:39:46 -08001286#ifdef CONFIG_FAIL_MAKE_REQUEST
1287
1288static DECLARE_FAULT_ATTR(fail_make_request);
1289
1290static int __init setup_fail_make_request(char *str)
1291{
1292 return setup_fault_attr(&fail_make_request, str);
1293}
1294__setup("fail_make_request=", setup_fail_make_request);
1295
1296static int should_fail_request(struct bio *bio)
1297{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001298 struct hd_struct *part = bio->bi_bdev->bd_part;
1299
1300 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001301 return should_fail(&fail_make_request, bio->bi_size);
1302
1303 return 0;
1304}
1305
1306static int __init fail_make_request_debugfs(void)
1307{
1308 return init_fault_attr_dentries(&fail_make_request,
1309 "fail_make_request");
1310}
1311
1312late_initcall(fail_make_request_debugfs);
1313
1314#else /* CONFIG_FAIL_MAKE_REQUEST */
1315
1316static inline int should_fail_request(struct bio *bio)
1317{
1318 return 0;
1319}
1320
1321#endif /* CONFIG_FAIL_MAKE_REQUEST */
1322
Jens Axboec07e2b42007-07-18 13:27:58 +02001323/*
1324 * Check whether this bio extends beyond the end of the device.
1325 */
1326static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1327{
1328 sector_t maxsector;
1329
1330 if (!nr_sectors)
1331 return 0;
1332
1333 /* Test device or partition size, when known. */
1334 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1335 if (maxsector) {
1336 sector_t sector = bio->bi_sector;
1337
1338 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1339 /*
1340 * This may well happen - the kernel calls bread()
1341 * without checking the size of the device, e.g., when
1342 * mounting a device.
1343 */
1344 handle_bad_sector(bio);
1345 return 1;
1346 }
1347 }
1348
1349 return 0;
1350}
1351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001353 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 * @bio: The bio describing the location in memory and on the device.
1355 *
1356 * generic_make_request() is used to make I/O requests of block
1357 * devices. It is passed a &struct bio, which describes the I/O that needs
1358 * to be done.
1359 *
1360 * generic_make_request() does not return any status. The
1361 * success/failure status of the request, along with notification of
1362 * completion, is delivered asynchronously through the bio->bi_end_io
1363 * function described (one day) else where.
1364 *
1365 * The caller of generic_make_request must make sure that bi_io_vec
1366 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1367 * set to describe the device address, and the
1368 * bi_end_io and optionally bi_private are set to describe how
1369 * completion notification should be signaled.
1370 *
1371 * generic_make_request and the drivers it calls may use bi_next if this
1372 * bio happens to be merged with someone else, and may change bi_dev and
1373 * bi_sector for remaps as it sees fit. So the values of these fields
1374 * should NOT be depended on after the call to generic_make_request.
1375 */
Neil Brownd89d8792007-05-01 09:53:42 +02001376static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
Jens Axboe165125e2007-07-24 09:28:11 +02001378 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001379 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001381 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001382 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Jens Axboec07e2b42007-07-18 13:27:58 +02001386 if (bio_check_eod(bio, nr_sectors))
1387 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 /*
1390 * Resolve the mapping until finished. (drivers are
1391 * still free to implement/resolve their own stacking
1392 * by explicitly returning 0)
1393 *
1394 * NOTE: we don't repeat the blk_size check for each new device.
1395 * Stacking drivers are expected to know what they are doing.
1396 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001397 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001398 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 do {
1400 char b[BDEVNAME_SIZE];
1401
1402 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001403 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 printk(KERN_ERR
1405 "generic_make_request: Trying to access "
1406 "nonexistent block-device %s (%Lu)\n",
1407 bdevname(bio->bi_bdev, b),
1408 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001409 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
1411
Jens Axboe4fa253f2007-07-18 13:13:10 +02001412 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001413 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 bdevname(bio->bi_bdev, b),
1415 bio_sectors(bio),
1416 q->max_hw_sectors);
1417 goto end_io;
1418 }
1419
Nick Pigginfde6ad22005-06-23 00:08:53 -07001420 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 goto end_io;
1422
Akinobu Mitac17bb492006-12-08 02:39:46 -08001423 if (should_fail_request(bio))
1424 goto end_io;
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 /*
1427 * If this device has partitions, remap block n
1428 * of partition p to block n+start(p) of the disk.
1429 */
1430 blk_partition_remap(bio);
1431
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001432 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1433 goto end_io;
1434
NeilBrown5ddfe962006-10-30 22:07:21 -08001435 if (old_sector != -1)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001436 trace_block_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001437 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001438
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001439 trace_block_bio_queue(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001440
NeilBrown5ddfe962006-10-30 22:07:21 -08001441 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001442 old_dev = bio->bi_bdev->bd_dev;
1443
Jens Axboec07e2b42007-07-18 13:27:58 +02001444 if (bio_check_eod(bio, nr_sectors))
1445 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001446
1447 if (bio_discard(bio) && !q->prepare_discard_fn) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001448 err = -EOPNOTSUPP;
1449 goto end_io;
1450 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 ret = q->make_request_fn(q, bio);
1453 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001454
1455 return;
1456
1457end_io:
1458 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459}
1460
Neil Brownd89d8792007-05-01 09:53:42 +02001461/*
1462 * We only want one ->make_request_fn to be active at a time,
1463 * else stack usage with stacked devices could be a problem.
1464 * So use current->bio_{list,tail} to keep a list of requests
1465 * submited by a make_request_fn function.
1466 * current->bio_tail is also used as a flag to say if
1467 * generic_make_request is currently active in this task or not.
1468 * If it is NULL, then no make_request is active. If it is non-NULL,
1469 * then a make_request is active, and new requests should be added
1470 * at the tail
1471 */
1472void generic_make_request(struct bio *bio)
1473{
1474 if (current->bio_tail) {
1475 /* make_request is active */
1476 *(current->bio_tail) = bio;
1477 bio->bi_next = NULL;
1478 current->bio_tail = &bio->bi_next;
1479 return;
1480 }
1481 /* following loop may be a bit non-obvious, and so deserves some
1482 * explanation.
1483 * Before entering the loop, bio->bi_next is NULL (as all callers
1484 * ensure that) so we have a list with a single bio.
1485 * We pretend that we have just taken it off a longer list, so
1486 * we assign bio_list to the next (which is NULL) and bio_tail
1487 * to &bio_list, thus initialising the bio_list of new bios to be
1488 * added. __generic_make_request may indeed add some more bios
1489 * through a recursive call to generic_make_request. If it
1490 * did, we find a non-NULL value in bio_list and re-enter the loop
1491 * from the top. In this case we really did just take the bio
1492 * of the top of the list (no pretending) and so fixup bio_list and
1493 * bio_tail or bi_next, and call into __generic_make_request again.
1494 *
1495 * The loop was structured like this to make only one call to
1496 * __generic_make_request (which is important as it is large and
1497 * inlined) and to keep the structure simple.
1498 */
1499 BUG_ON(bio->bi_next);
1500 do {
1501 current->bio_list = bio->bi_next;
1502 if (bio->bi_next == NULL)
1503 current->bio_tail = &current->bio_list;
1504 else
1505 bio->bi_next = NULL;
1506 __generic_make_request(bio);
1507 bio = current->bio_list;
1508 } while (bio);
1509 current->bio_tail = NULL; /* deactivate */
1510}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511EXPORT_SYMBOL(generic_make_request);
1512
1513/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001514 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1516 * @bio: The &struct bio which describes the I/O
1517 *
1518 * submit_bio() is very similar in purpose to generic_make_request(), and
1519 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001520 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
1522 */
1523void submit_bio(int rw, struct bio *bio)
1524{
1525 int count = bio_sectors(bio);
1526
Jens Axboe22e2c502005-06-27 10:55:12 +02001527 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Jens Axboebf2de6f2007-09-27 13:01:25 +02001529 /*
1530 * If it's a regular read/write or a barrier with data attached,
1531 * go through the normal accounting stuff before submission.
1532 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001533 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001534 if (rw & WRITE) {
1535 count_vm_events(PGPGOUT, count);
1536 } else {
1537 task_io_account_read(bio->bi_size);
1538 count_vm_events(PGPGIN, count);
1539 }
1540
1541 if (unlikely(block_dump)) {
1542 char b[BDEVNAME_SIZE];
1543 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001544 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001545 (rw & WRITE) ? "WRITE" : "READ",
1546 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001547 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
1550
1551 generic_make_request(bio);
1552}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553EXPORT_SYMBOL(submit_bio);
1554
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001555/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001556 * blk_rq_check_limits - Helper function to check a request for the queue limit
1557 * @q: the queue
1558 * @rq: the request being checked
1559 *
1560 * Description:
1561 * @rq may have been made based on weaker limitations of upper-level queues
1562 * in request stacking drivers, and it may violate the limitation of @q.
1563 * Since the block layer and the underlying device driver trust @rq
1564 * after it is inserted to @q, it should be checked against @q before
1565 * the insertion using this generic function.
1566 *
1567 * This function should also be useful for request stacking drivers
1568 * in some cases below, so export this fuction.
1569 * Request stacking drivers like request-based dm may change the queue
1570 * limits while requests are in the queue (e.g. dm's table swapping).
1571 * Such request stacking drivers should check those requests agaist
1572 * the new queue limits again when they dispatch those requests,
1573 * although such checkings are also done against the old queue limits
1574 * when submitting requests.
1575 */
1576int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1577{
1578 if (rq->nr_sectors > q->max_sectors ||
1579 rq->data_len > q->max_hw_sectors << 9) {
1580 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1581 return -EIO;
1582 }
1583
1584 /*
1585 * queue's settings related to segment counting like q->bounce_pfn
1586 * may differ from that of other stacking queues.
1587 * Recalculate it to check the request correctly on this queue's
1588 * limitation.
1589 */
1590 blk_recalc_rq_segments(rq);
1591 if (rq->nr_phys_segments > q->max_phys_segments ||
1592 rq->nr_phys_segments > q->max_hw_segments) {
1593 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1594 return -EIO;
1595 }
1596
1597 return 0;
1598}
1599EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1600
1601/**
1602 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1603 * @q: the queue to submit the request
1604 * @rq: the request being queued
1605 */
1606int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1607{
1608 unsigned long flags;
1609
1610 if (blk_rq_check_limits(q, rq))
1611 return -EIO;
1612
1613#ifdef CONFIG_FAIL_MAKE_REQUEST
1614 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1615 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1616 return -EIO;
1617#endif
1618
1619 spin_lock_irqsave(q->queue_lock, flags);
1620
1621 /*
1622 * Submitting request must be dequeued before calling this function
1623 * because it will be linked to another request_queue
1624 */
1625 BUG_ON(blk_queued_rq(rq));
1626
1627 drive_stat_acct(rq, 1);
1628 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1629
1630 spin_unlock_irqrestore(q->queue_lock, flags);
1631
1632 return 0;
1633}
1634EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1635
1636/**
Tejun Heo53a08802008-12-03 12:41:26 +01001637 * blkdev_dequeue_request - dequeue request and start timeout timer
1638 * @req: request to dequeue
1639 *
1640 * Dequeue @req and start timeout timer on it. This hands off the
1641 * request to the driver.
1642 *
1643 * Block internal functions which don't want to start timer should
1644 * call elv_dequeue_request().
1645 */
1646void blkdev_dequeue_request(struct request *req)
1647{
1648 elv_dequeue_request(req->q, req);
1649
1650 /*
1651 * We are now handing the request to the hardware, add the
1652 * timeout handler.
1653 */
1654 blk_add_timer(req);
1655}
1656EXPORT_SYMBOL(blkdev_dequeue_request);
1657
1658/**
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001659 * __end_that_request_first - end I/O on a request
1660 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001661 * @error: %0 for success, < %0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001662 * @nr_bytes: number of bytes to complete
1663 *
1664 * Description:
1665 * Ends I/O on a number of bytes attached to @req, and sets it up
1666 * for the next range of segments (if any) in the cluster.
1667 *
1668 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001669 * %0 - we are done with this request, call end_that_request_last()
1670 * %1 - still buffers pending for this request
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001671 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001672static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 int nr_bytes)
1674{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001675 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 struct bio *bio;
1677
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001678 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 /*
Randy Dunlap710027a2008-08-19 20:13:11 +02001681 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 * sense key with us all the way through
1683 */
1684 if (!blk_pc_request(req))
1685 req->errors = 0;
1686
Jens Axboe6728cb02008-01-31 13:03:55 +01001687 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1688 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 req->rq_disk ? req->rq_disk->disk_name : "?",
1690 (unsigned long long)req->sector);
1691 }
1692
Jens Axboed72d9042005-11-01 08:35:42 +01001693 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01001694 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001695 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001696 int cpu;
Jens Axboea3623572005-11-01 09:26:16 +01001697
Tejun Heo074a7ac2008-08-25 19:56:14 +09001698 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001699 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Tejun Heo074a7ac2008-08-25 19:56:14 +09001700 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1701 part_stat_unlock();
Jens Axboed72d9042005-11-01 08:35:42 +01001702 }
1703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 total_bytes = bio_nbytes = 0;
1705 while ((bio = req->bio) != NULL) {
1706 int nbytes;
1707
1708 if (nr_bytes >= bio->bi_size) {
1709 req->bio = bio->bi_next;
1710 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001711 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 next_idx = 0;
1713 bio_nbytes = 0;
1714 } else {
1715 int idx = bio->bi_idx + next_idx;
1716
1717 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1718 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001719 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001720 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 break;
1722 }
1723
1724 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1725 BIO_BUG_ON(nbytes > bio->bi_size);
1726
1727 /*
1728 * not a complete bvec done
1729 */
1730 if (unlikely(nbytes > nr_bytes)) {
1731 bio_nbytes += nr_bytes;
1732 total_bytes += nr_bytes;
1733 break;
1734 }
1735
1736 /*
1737 * advance to the next vector
1738 */
1739 next_idx++;
1740 bio_nbytes += nbytes;
1741 }
1742
1743 total_bytes += nbytes;
1744 nr_bytes -= nbytes;
1745
Jens Axboe6728cb02008-01-31 13:03:55 +01001746 bio = req->bio;
1747 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 /*
1749 * end more in this run, or just return 'not-done'
1750 */
1751 if (unlikely(nr_bytes <= 0))
1752 break;
1753 }
1754 }
1755
1756 /*
1757 * completely done
1758 */
1759 if (!req->bio)
1760 return 0;
1761
1762 /*
1763 * if the request wasn't completed, update state
1764 */
1765 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001766 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 bio->bi_idx += next_idx;
1768 bio_iovec(bio)->bv_offset += nr_bytes;
1769 bio_iovec(bio)->bv_len -= nr_bytes;
1770 }
1771
1772 blk_recalc_rq_sectors(req, total_bytes >> 9);
1773 blk_recalc_rq_segments(req);
1774 return 1;
1775}
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777/*
1778 * queue lock must be held
1779 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001780static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
1782 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01001783
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001784 if (blk_rq_tagged(req))
1785 blk_queue_end_tag(req->q, req);
1786
1787 if (blk_queued_rq(req))
Tejun Heo53a08802008-12-03 12:41:26 +01001788 elv_dequeue_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 if (unlikely(laptop_mode) && blk_fs_request(req))
1791 laptop_io_completion();
1792
Mike Andersone78042e2008-10-30 02:16:20 -07001793 blk_delete_timer(req);
1794
Jens Axboefd0ff8a2006-05-23 11:23:49 +02001795 /*
1796 * Account IO completion. bar_rq isn't accounted as a normal
1797 * IO on queueing nor completion. Accounting the containing
1798 * request is enough.
1799 */
1800 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01001802 const int rw = rq_data_dir(req);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001803 struct hd_struct *part;
Tejun Heoc9959052008-08-25 19:47:21 +09001804 int cpu;
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001805
Tejun Heo074a7ac2008-08-25 19:56:14 +09001806 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001807 part = disk_map_sector_rcu(disk, req->sector);
Jens Axboea3623572005-11-01 09:26:16 +01001808
Tejun Heo074a7ac2008-08-25 19:56:14 +09001809 part_stat_inc(cpu, part, ios[rw]);
1810 part_stat_add(cpu, part, ticks[rw], duration);
1811 part_round_stats(cpu, part);
1812 part_dec_in_flight(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001813
Tejun Heo074a7ac2008-08-25 19:56:14 +09001814 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001818 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001819 else {
1820 if (blk_bidi_rq(req))
1821 __blk_put_request(req->next_rq->q, req->next_rq);
1822
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 }
1825}
1826
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001827/**
1828 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001829 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001830 **/
1831unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001832{
1833 if (blk_fs_request(rq))
1834 return rq->hard_nr_sectors << 9;
1835
1836 return rq->data_len;
1837}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001838EXPORT_SYMBOL_GPL(blk_rq_bytes);
1839
1840/**
1841 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001842 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001843 **/
1844unsigned int blk_rq_cur_bytes(struct request *rq)
1845{
1846 if (blk_fs_request(rq))
1847 return rq->current_nr_sectors << 9;
1848
1849 if (rq->bio)
1850 return rq->bio->bi_size;
1851
1852 return rq->data_len;
1853}
1854EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001855
1856/**
Jens Axboea0cd1282007-09-21 10:41:07 +02001857 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001858 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001859 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001860 *
1861 * Description:
1862 * Ends I/O on the current segment of a request. If that is the only
1863 * remaining segment, the request is also completed and freed.
1864 *
Randy Dunlap710027a2008-08-19 20:13:11 +02001865 * This is a remnant of how older block drivers handled I/O completions.
1866 * Modern drivers typically end I/O on the full request in one go, unless
Jens Axboea0cd1282007-09-21 10:41:07 +02001867 * they have a residual value to account for. For that case this function
1868 * isn't really useful, unless the residual just happens to be the
1869 * full current segment. In other words, don't use this function in new
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001870 * code. Use blk_end_request() or __blk_end_request() to end a request.
Jens Axboea0cd1282007-09-21 10:41:07 +02001871 **/
1872void end_request(struct request *req, int uptodate)
1873{
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001874 int error = 0;
1875
1876 if (uptodate <= 0)
1877 error = uptodate ? uptodate : -EIO;
1878
1879 __blk_end_request(req, error, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001880}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881EXPORT_SYMBOL(end_request);
1882
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001883static int end_that_request_data(struct request *rq, int error,
1884 unsigned int nr_bytes, unsigned int bidi_bytes)
1885{
1886 if (rq->bio) {
1887 if (__end_that_request_first(rq, error, nr_bytes))
1888 return 1;
1889
1890 /* Bidi request must be completed as a whole */
1891 if (blk_bidi_rq(rq) &&
1892 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1893 return 1;
1894 }
1895
1896 return 0;
1897}
1898
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001899/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001900 * blk_end_io - Generic end_io function to complete a request.
1901 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001902 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001903 * @nr_bytes: number of bytes to complete @rq
1904 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001905 * @drv_callback: function called between completion of bios in the request
1906 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001907 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001908 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001909 *
1910 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001911 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001912 * If @rq has leftover, sets it up for the next range of segments.
1913 *
1914 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001915 * %0 - we are done with this request
1916 * %1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001917 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001918static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1919 unsigned int bidi_bytes,
1920 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001921{
1922 struct request_queue *q = rq->q;
1923 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001924
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001925 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1926 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001927
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001928 /* Special feature for tricky drivers */
1929 if (drv_callback && drv_callback(rq))
1930 return 1;
1931
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001932 add_disk_randomness(rq->rq_disk);
1933
1934 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001935 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001936 spin_unlock_irqrestore(q->queue_lock, flags);
1937
1938 return 0;
1939}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001940
1941/**
1942 * blk_end_request - Helper function for drivers to complete the request.
1943 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001944 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001945 * @nr_bytes: number of bytes to complete
1946 *
1947 * Description:
1948 * Ends I/O on a number of bytes attached to @rq.
1949 * If @rq has leftover, sets it up for the next range of segments.
1950 *
1951 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001952 * %0 - we are done with this request
1953 * %1 - still buffers pending for this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001954 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001955int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001956{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001957 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001958}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001959EXPORT_SYMBOL_GPL(blk_end_request);
1960
1961/**
1962 * __blk_end_request - Helper function for drivers to complete the request.
1963 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001964 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001965 * @nr_bytes: number of bytes to complete
1966 *
1967 * Description:
1968 * Must be called with queue lock held unlike blk_end_request().
1969 *
1970 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001971 * %0 - we are done with this request
1972 * %1 - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001973 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001974int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001975{
Jens Axboe60540162008-08-26 13:34:34 +02001976 if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
Jens Axboe051cc392008-08-08 11:06:45 +02001977 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001978
1979 add_disk_randomness(rq->rq_disk);
1980
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001981 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001982
1983 return 0;
1984}
1985EXPORT_SYMBOL_GPL(__blk_end_request);
1986
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001987/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001988 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1989 * @rq: the bidi request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001990 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001991 * @nr_bytes: number of bytes to complete @rq
1992 * @bidi_bytes: number of bytes to complete @rq->next_rq
1993 *
1994 * Description:
1995 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1996 *
1997 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001998 * %0 - we are done with this request
1999 * %1 - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002000 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002001int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
2002 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002003{
2004 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
2005}
2006EXPORT_SYMBOL_GPL(blk_end_bidi_request);
2007
2008/**
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002009 * blk_update_request - Special helper function for request stacking drivers
2010 * @rq: the request being processed
2011 * @error: %0 for success, < %0 for error
2012 * @nr_bytes: number of bytes to complete @rq
2013 *
2014 * Description:
2015 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2016 * the request structure even if @rq doesn't have leftover.
2017 * If @rq has leftover, sets it up for the next range of segments.
2018 *
2019 * This special helper function is only for request stacking drivers
2020 * (e.g. request-based dm) so that they can handle partial completion.
2021 * Actual device drivers should use blk_end_request instead.
2022 */
2023void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2024{
2025 if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2026 /*
2027 * These members are not updated in end_that_request_data()
2028 * when all bios are completed.
2029 * Update them so that the request stacking driver can find
2030 * how many bytes remain in the request later.
2031 */
2032 rq->nr_sectors = rq->hard_nr_sectors = 0;
2033 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2034 }
2035}
2036EXPORT_SYMBOL_GPL(blk_update_request);
2037
2038/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002039 * blk_end_request_callback - Special helper function for tricky drivers
2040 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02002041 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002042 * @nr_bytes: number of bytes to complete
2043 * @drv_callback: function called between completion of bios in the request
2044 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02002045 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002046 * completion of the request.
2047 *
2048 * Description:
2049 * Ends I/O on a number of bytes attached to @rq.
2050 * If @rq has leftover, sets it up for the next range of segments.
2051 *
2052 * This special helper function is used only for existing tricky drivers.
2053 * (e.g. cdrom_newpc_intr() of ide-cd)
2054 * This interface will be removed when such drivers are rewritten.
2055 * Don't use this interface in other places anymore.
2056 *
2057 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002058 * %0 - we are done with this request
2059 * %1 - this request is not freed yet.
2060 * this request still has pending buffers or
2061 * the driver doesn't want to finish this request yet.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002062 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002063int blk_end_request_callback(struct request *rq, int error,
2064 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002065 int (drv_callback)(struct request *))
2066{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002067 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002068}
2069EXPORT_SYMBOL_GPL(blk_end_request_callback);
2070
Jens Axboe86db1e22008-01-29 14:53:40 +01002071void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2072 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073{
David Woodhoused628eae2008-08-09 16:22:17 +01002074 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2075 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002076 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
David Woodhousefb2dce82008-08-05 18:01:53 +01002078 if (bio_has_data(bio)) {
2079 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002080 rq->buffer = bio_data(bio);
2081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 rq->current_nr_sectors = bio_cur_sectors(bio);
2083 rq->hard_cur_sectors = rq->current_nr_sectors;
2084 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002085 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
2087 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
NeilBrown66846572007-08-16 13:31:28 +02002089 if (bio->bi_bdev)
2090 rq->rq_disk = bio->bi_bdev->bd_disk;
2091}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002093/**
2094 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2095 * @q : the queue of the device being checked
2096 *
2097 * Description:
2098 * Check if underlying low-level drivers of a device are busy.
2099 * If the drivers want to export their busy state, they must set own
2100 * exporting function using blk_queue_lld_busy() first.
2101 *
2102 * Basically, this function is used only by request stacking drivers
2103 * to stop dispatching requests to underlying devices when underlying
2104 * devices are busy. This behavior helps more I/O merging on the queue
2105 * of the request stacking driver and prevents I/O throughput regression
2106 * on burst I/O load.
2107 *
2108 * Return:
2109 * 0 - Not busy (The request stacking driver should dispatch request)
2110 * 1 - Busy (The request stacking driver should stop dispatching request)
2111 */
2112int blk_lld_busy(struct request_queue *q)
2113{
2114 if (q->lld_busy_fn)
2115 return q->lld_busy_fn(q);
2116
2117 return 0;
2118}
2119EXPORT_SYMBOL_GPL(blk_lld_busy);
2120
Jens Axboe18887ad2008-07-28 13:08:45 +02002121int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122{
2123 return queue_work(kblockd_workqueue, work);
2124}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125EXPORT_SYMBOL(kblockd_schedule_work);
2126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127int __init blk_dev_init(void)
2128{
2129 kblockd_workqueue = create_workqueue("kblockd");
2130 if (!kblockd_workqueue)
2131 panic("Failed to create kblockd\n");
2132
2133 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002134 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Jens Axboe8324aa92008-01-29 14:51:59 +01002136 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002137 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 return 0;
2140}
2141