blob: c89883be87379d9454ab1af7cd68319e92795597 [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
Jerome Marchand26308ea2009-03-27 10:31:51 +010071 if (!blk_fs_request(rq) || !blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010072 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;
Li Zefane2494e12009-04-02 13:43:26 +0800134 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100136 rq->ref_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200138EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
NeilBrown5bb23a62007-09-27 12:46:13 +0200140static void req_bio_endio(struct request *rq, struct bio *bio,
141 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100142{
Jens Axboe165125e2007-07-24 09:28:11 +0200143 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100144
NeilBrown5bb23a62007-09-27 12:46:13 +0200145 if (&q->bar_rq != rq) {
146 if (error)
147 clear_bit(BIO_UPTODATE, &bio->bi_flags);
148 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
149 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100150
NeilBrown5bb23a62007-09-27 12:46:13 +0200151 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100152 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700153 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200154 nbytes = bio->bi_size;
155 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100156
Keith Mannthey08bafc02008-11-25 10:24:35 +0100157 if (unlikely(rq->cmd_flags & REQ_QUIET))
158 set_bit(BIO_QUIET, &bio->bi_flags);
159
NeilBrown5bb23a62007-09-27 12:46:13 +0200160 bio->bi_size -= nbytes;
161 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200162
163 if (bio_integrity(bio))
164 bio_integrity_advance(bio, nbytes);
165
NeilBrown5bb23a62007-09-27 12:46:13 +0200166 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200167 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200168 } else {
169
170 /*
171 * Okay, this is the barrier request in progress, just
172 * record the error;
173 */
174 if (error && !q->orderr)
175 q->orderr = error;
176 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100177}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179void blk_dump_rq_flags(struct request *rq, char *msg)
180{
181 int bit;
182
Jens Axboe6728cb02008-01-31 13:03:55 +0100183 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200184 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
185 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Jens Axboe6728cb02008-01-31 13:03:55 +0100187 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
188 (unsigned long long)rq->sector,
189 rq->nr_sectors,
190 rq->current_nr_sectors);
191 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
192 rq->bio, rq->biotail,
193 rq->buffer, rq->data,
194 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Jens Axboe4aff5e22006-08-10 08:44:47 +0200196 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100197 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200198 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 printk("%02x ", rq->cmd[bit]);
200 printk("\n");
201 }
202}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203EXPORT_SYMBOL(blk_dump_rq_flags);
204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205/*
206 * "plug" the device if there are no outstanding requests: this will
207 * force the transfer to start only after we have put all the requests
208 * on the list.
209 *
210 * This is called with interrupts off and no requests on the queue and
211 * with the queue lock held.
212 */
Jens Axboe165125e2007-07-24 09:28:11 +0200213void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 WARN_ON(!irqs_disabled());
216
217 /*
218 * don't plug a stopped queue, it must be paired with blk_start_queue()
219 * which will restart the queueing
220 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200221 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 return;
223
Jens Axboee48ec692008-07-03 13:18:54 +0200224 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100226 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229EXPORT_SYMBOL(blk_plug_device);
230
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200231/**
232 * blk_plug_device_unlocked - plug a device without queue lock held
233 * @q: The &struct request_queue to plug
234 *
235 * Description:
236 * Like @blk_plug_device(), but grabs the queue lock and disables
237 * interrupts.
238 **/
239void blk_plug_device_unlocked(struct request_queue *q)
240{
241 unsigned long flags;
242
243 spin_lock_irqsave(q->queue_lock, flags);
244 blk_plug_device(q);
245 spin_unlock_irqrestore(q->queue_lock, flags);
246}
247EXPORT_SYMBOL(blk_plug_device_unlocked);
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249/*
250 * remove the queue from the plugged list, if present. called with
251 * queue lock held and interrupts disabled.
252 */
Jens Axboe165125e2007-07-24 09:28:11 +0200253int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 WARN_ON(!irqs_disabled());
256
Jens Axboee48ec692008-07-03 13:18:54 +0200257 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return 0;
259
260 del_timer(&q->unplug_timer);
261 return 1;
262}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263EXPORT_SYMBOL(blk_remove_plug);
264
265/*
266 * remove the plug and let it rip..
267 */
Jens Axboe165125e2007-07-24 09:28:11 +0200268void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200270 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200272 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return;
274
Jens Axboe22e2c502005-06-27 10:55:12 +0200275 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278/**
279 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200280 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 *
282 * Description:
283 * Linux uses plugging to build bigger requests queues before letting
284 * the device have at them. If a queue is plugged, the I/O scheduler
285 * is still adding and merging requests on the queue. Once the queue
286 * gets unplugged, the request_fn defined for the queue is invoked and
287 * transfers started.
288 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200289void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200291 if (blk_queue_plugged(q)) {
292 spin_lock_irq(q->queue_lock);
293 __generic_unplug_device(q);
294 spin_unlock_irq(q->queue_lock);
295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297EXPORT_SYMBOL(generic_unplug_device);
298
299static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
300 struct page *page)
301{
Jens Axboe165125e2007-07-24 09:28:11 +0200302 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500304 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305}
306
Jens Axboe86db1e22008-01-29 14:53:40 +0100307void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
Jens Axboe165125e2007-07-24 09:28:11 +0200309 struct request_queue *q =
310 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100312 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 q->unplug_fn(q);
314}
315
Jens Axboe86db1e22008-01-29 14:53:40 +0100316void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
Jens Axboe165125e2007-07-24 09:28:11 +0200318 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100320 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200321 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322}
323
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500324void blk_unplug(struct request_queue *q)
325{
326 /*
327 * devices don't necessarily have an ->unplug_fn defined
328 */
329 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100330 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500331 q->unplug_fn(q);
332 }
333}
334EXPORT_SYMBOL(blk_unplug);
335
Jens Axboec7c22e42008-09-13 20:26:01 +0200336static void blk_invoke_request_fn(struct request_queue *q)
337{
Jens Axboe80a4b582008-10-14 09:51:06 +0200338 if (unlikely(blk_queue_stopped(q)))
339 return;
340
Jens Axboec7c22e42008-09-13 20:26:01 +0200341 /*
342 * one level of recursion is ok and is much faster than kicking
343 * the unplug handling
344 */
345 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
346 q->request_fn(q);
347 queue_flag_clear(QUEUE_FLAG_REENTER, q);
348 } else {
349 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
350 kblockd_schedule_work(q, &q->unplug_work);
351 }
352}
353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354/**
355 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200356 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 *
358 * Description:
359 * blk_start_queue() will clear the stop flag on the queue, and call
360 * the request_fn for the queue if it was in a stopped state when
361 * entered. Also see blk_stop_queue(). Queue lock must be held.
362 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200363void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200365 WARN_ON(!irqs_disabled());
366
Nick Piggin75ad23b2008-04-29 14:48:33 +0200367 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Jens Axboec7c22e42008-09-13 20:26:01 +0200368 blk_invoke_request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370EXPORT_SYMBOL(blk_start_queue);
371
372/**
373 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200374 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 *
376 * Description:
377 * The Linux block layer assumes that a block driver will consume all
378 * entries on the request queue when the request_fn strategy is called.
379 * Often this will not happen, because of hardware limitations (queue
380 * depth settings). If a device driver gets a 'queue full' response,
381 * or if it simply chooses not to queue more I/O at one point, it can
382 * call this function to prevent the request_fn from being called until
383 * the driver has signalled it's ready to go again. This happens by calling
384 * blk_start_queue() to restart queue operations. Queue lock must be held.
385 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200386void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387{
388 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200389 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391EXPORT_SYMBOL(blk_stop_queue);
392
393/**
394 * blk_sync_queue - cancel any pending callbacks on a queue
395 * @q: the queue
396 *
397 * Description:
398 * The block layer may perform asynchronous callback activity
399 * on a queue, such as calling the unplug function after a timeout.
400 * A block device may call blk_sync_queue to ensure that any
401 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200402 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 * that its ->make_request_fn will not re-add plugging prior to calling
404 * this function.
405 *
406 */
407void blk_sync_queue(struct request_queue *q)
408{
409 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100410 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100411 cancel_work_sync(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413EXPORT_SYMBOL(blk_sync_queue);
414
415/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200416 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200418 *
419 * Description:
420 * See @blk_run_queue. This variant must be called with the queue lock
421 * held and interrupts disabled.
422 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200424void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200427
428 /*
429 * Only recurse once to avoid overrunning the stack, let the unplug
430 * handling reinvoke the handler shortly if we already got there.
431 */
Jens Axboec7c22e42008-09-13 20:26:01 +0200432 if (!elv_queue_empty(q))
433 blk_invoke_request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200434}
435EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200436
Nick Piggin75ad23b2008-04-29 14:48:33 +0200437/**
438 * blk_run_queue - run a single device queue
439 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200440 *
441 * Description:
442 * Invoke request handling on this queue, if it has pending work to do.
443 * May be used to restart queueing when a request has completed. Also
444 * See @blk_start_queueing.
445 *
Nick Piggin75ad23b2008-04-29 14:48:33 +0200446 */
447void blk_run_queue(struct request_queue *q)
448{
449 unsigned long flags;
450
451 spin_lock_irqsave(q->queue_lock, flags);
452 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 spin_unlock_irqrestore(q->queue_lock, flags);
454}
455EXPORT_SYMBOL(blk_run_queue);
456
Jens Axboe165125e2007-07-24 09:28:11 +0200457void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500458{
459 kobject_put(&q->kobj);
460}
Al Viro483f4af2006-03-18 18:34:37 -0500461
Jens Axboe6728cb02008-01-31 13:03:55 +0100462void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500463{
Jens Axboee3335de2008-09-18 09:22:54 -0700464 /*
465 * We know we have process context here, so we can be a little
466 * cautious and ensure that pending block actions on this device
467 * are done before moving on. Going into this function, we should
468 * not have processes doing IO to this device.
469 */
470 blk_sync_queue(q);
471
Al Viro483f4af2006-03-18 18:34:37 -0500472 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200473 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500474 mutex_unlock(&q->sysfs_lock);
475
476 if (q->elevator)
477 elevator_exit(q->elevator);
478
479 blk_put_queue(q);
480}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481EXPORT_SYMBOL(blk_cleanup_queue);
482
Jens Axboe165125e2007-07-24 09:28:11 +0200483static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 struct request_list *rl = &q->rq;
486
Jens Axboe1faa16d2009-04-06 14:48:01 +0200487 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
488 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200489 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200490 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
491 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Christoph Lameter19460892005-06-23 00:08:19 -0700493 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
494 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 if (!rl->rq_pool)
497 return -ENOMEM;
498
499 return 0;
500}
501
Jens Axboe165125e2007-07-24 09:28:11 +0200502struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
Christoph Lameter19460892005-06-23 00:08:19 -0700504 return blk_alloc_queue_node(gfp_mask, -1);
505}
506EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Jens Axboe165125e2007-07-24 09:28:11 +0200508struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700509{
Jens Axboe165125e2007-07-24 09:28:11 +0200510 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700511 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700512
Jens Axboe8324aa92008-01-29 14:51:59 +0100513 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700514 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (!q)
516 return NULL;
517
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700518 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
519 q->backing_dev_info.unplug_io_data = q;
520 err = bdi_init(&q->backing_dev_info);
521 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100522 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700523 return NULL;
524 }
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700527 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
528 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200529 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500530
Jens Axboe8324aa92008-01-29 14:51:59 +0100531 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Al Viro483f4af2006-03-18 18:34:37 -0500533 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700534 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return q;
537}
Christoph Lameter19460892005-06-23 00:08:19 -0700538EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540/**
541 * blk_init_queue - prepare a request queue for use with a block device
542 * @rfn: The function to be called to process requests that have been
543 * placed on the queue.
544 * @lock: Request queue spin lock
545 *
546 * Description:
547 * If a block device wishes to use the standard request handling procedures,
548 * which sorts requests and coalesces adjacent requests, then it must
549 * call blk_init_queue(). The function @rfn will be called when there
550 * are requests on the queue that need to be processed. If the device
551 * supports plugging, then @rfn may not be called immediately when requests
552 * are available on the queue, but may be called at some time later instead.
553 * Plugged queues are generally unplugged when a buffer belonging to one
554 * of the requests on the queue is needed, or due to memory pressure.
555 *
556 * @rfn is not required, or even expected, to remove all requests off the
557 * queue, but only as many as it can handle at a time. If it does leave
558 * requests on the queue, it is responsible for arranging that the requests
559 * get dealt with eventually.
560 *
561 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200562 * request queue; this lock will be taken also from interrupt context, so irq
563 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200565 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 * it didn't succeed.
567 *
568 * Note:
569 * blk_init_queue() must be paired with a blk_cleanup_queue() call
570 * when the block device is deactivated (such as at module unload).
571 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700572
Jens Axboe165125e2007-07-24 09:28:11 +0200573struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Christoph Lameter19460892005-06-23 00:08:19 -0700575 return blk_init_queue_node(rfn, lock, -1);
576}
577EXPORT_SYMBOL(blk_init_queue);
578
Jens Axboe165125e2007-07-24 09:28:11 +0200579struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700580blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
581{
Jens Axboe165125e2007-07-24 09:28:11 +0200582 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 if (!q)
585 return NULL;
586
Christoph Lameter19460892005-06-23 00:08:19 -0700587 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500588 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100589 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500590 return NULL;
591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
152587d2005-04-12 16:22:06 -0500593 /*
594 * if caller didn't supply a lock, they get per-queue locking with
595 * our embedded lock
596 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700597 if (!lock)
152587d2005-04-12 16:22:06 -0500598 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 q->prep_rq_fn = NULL;
602 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100603 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 q->queue_lock = lock;
605
Jens Axboef3b144a2009-03-06 08:48:33 +0100606 /*
607 * This also sets hw/phys segments, boundary and size
608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Alan Stern44ec9542007-02-20 11:01:57 -0500611 q->sg_reserved_size = INT_MAX;
612
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900613 blk_set_cmd_filter_defaults(&q->cmd_filter);
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 /*
616 * all done
617 */
618 if (!elevator_init(q, NULL)) {
619 blk_queue_congestion_threshold(q);
620 return q;
621 }
622
Al Viro8669aaf2006-03-18 13:50:00 -0500623 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return NULL;
625}
Christoph Lameter19460892005-06-23 00:08:19 -0700626EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Jens Axboe165125e2007-07-24 09:28:11 +0200628int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700630 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500631 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return 0;
633 }
634
635 return 1;
636}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Jens Axboe165125e2007-07-24 09:28:11 +0200638static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200640 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200641 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 mempool_free(rq, q->rq.rq_pool);
643}
644
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200645static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200646blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
648 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
649
650 if (!rq)
651 return NULL;
652
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200653 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200654
Jerome Marchand42dad762009-04-22 14:01:49 +0200655 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Tejun Heocb98fc82005-10-28 08:29:39 +0200657 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200658 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200659 mempool_free(rq, q->rq.rq_pool);
660 return NULL;
661 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200662 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Tejun Heocb98fc82005-10-28 08:29:39 +0200665 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
668/*
669 * ioc_batching returns true if the ioc is a valid batching request and
670 * should be given priority access to a request.
671 */
Jens Axboe165125e2007-07-24 09:28:11 +0200672static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 if (!ioc)
675 return 0;
676
677 /*
678 * Make sure the process is able to allocate at least 1 request
679 * even if the batch times out, otherwise we could theoretically
680 * lose wakeups.
681 */
682 return ioc->nr_batch_requests == q->nr_batching ||
683 (ioc->nr_batch_requests > 0
684 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
685}
686
687/*
688 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
689 * will cause the process to be a "batcher" on all queues in the system. This
690 * is the behaviour we want though - once it gets a wakeup it should be given
691 * a nice run.
692 */
Jens Axboe165125e2007-07-24 09:28:11 +0200693static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
695 if (!ioc || ioc_batching(q, ioc))
696 return;
697
698 ioc->nr_batch_requests = q->nr_batching;
699 ioc->last_waited = jiffies;
700}
701
Jens Axboe1faa16d2009-04-06 14:48:01 +0200702static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
704 struct request_list *rl = &q->rq;
705
Jens Axboe1faa16d2009-04-06 14:48:01 +0200706 if (rl->count[sync] < queue_congestion_off_threshold(q))
707 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Jens Axboe1faa16d2009-04-06 14:48:01 +0200709 if (rl->count[sync] + 1 <= q->nr_requests) {
710 if (waitqueue_active(&rl->wait[sync]))
711 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Jens Axboe1faa16d2009-04-06 14:48:01 +0200713 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 }
715}
716
717/*
718 * A request has just been released. Account for it, update the full and
719 * congestion status, wake up any waiters. Called under q->queue_lock.
720 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200721static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 struct request_list *rl = &q->rq;
724
Jens Axboe1faa16d2009-04-06 14:48:01 +0200725 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200726 if (priv)
727 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Jens Axboe1faa16d2009-04-06 14:48:01 +0200729 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Jens Axboe1faa16d2009-04-06 14:48:01 +0200731 if (unlikely(rl->starved[sync ^ 1]))
732 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735/*
Nick Piggind6344532005-06-28 20:45:14 -0700736 * Get a free request, queue_lock must be held.
737 * Returns NULL on failure, with queue_lock held.
738 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
Jens Axboe165125e2007-07-24 09:28:11 +0200740static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100741 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
743 struct request *rq = NULL;
744 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100745 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200746 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100747 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Jens Axboe7749a8d2006-12-13 13:02:26 +0100749 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100750 if (may_queue == ELV_MQUEUE_NO)
751 goto rq_starved;
752
Jens Axboe1faa16d2009-04-06 14:48:01 +0200753 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
754 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200755 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100756 /*
757 * The queue will fill after this allocation, so set
758 * it as full, and mark this process as "batching".
759 * This process will be allowed to complete a batch of
760 * requests, others will be blocked.
761 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200762 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100763 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200764 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100765 } else {
766 if (may_queue != ELV_MQUEUE_MUST
767 && !ioc_batching(q, ioc)) {
768 /*
769 * The queue is full and the allocating
770 * process is not a "batcher", and not
771 * exempted by the IO scheduler
772 */
773 goto out;
774 }
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200777 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
779
Jens Axboe082cf692005-06-28 16:35:11 +0200780 /*
781 * Only allow batching queuers to allocate up to 50% over the defined
782 * limit of requests, otherwise we could have thousands of requests
783 * allocated with any setting of ->nr_requests
784 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200785 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200786 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100787
Jens Axboe1faa16d2009-04-06 14:48:01 +0200788 rl->count[is_sync]++;
789 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200790
Jens Axboe64521d12005-10-28 08:30:39 +0200791 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200792 if (priv)
793 rl->elvpriv++;
794
Jerome Marchand42dad762009-04-22 14:01:49 +0200795 if (blk_queue_io_stat(q))
796 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 spin_unlock_irq(q->queue_lock);
798
Jens Axboe7749a8d2006-12-13 13:02:26 +0100799 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100800 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 /*
802 * Allocation failed presumably due to memory. Undo anything
803 * we might have messed up.
804 *
805 * Allocating task should really be put onto the front of the
806 * wait queue, but this is pretty rare.
807 */
808 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200809 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
811 /*
812 * in the very unlikely event that allocation failed and no
813 * requests for this direction was pending, mark us starved
814 * so that freeing of a request in the other direction will
815 * notice us. another possible fix would be to split the
816 * rq mempool into READ and WRITE
817 */
818rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200819 if (unlikely(rl->count[is_sync] == 0))
820 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 goto out;
823 }
824
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100825 /*
826 * ioc may be NULL here, and ioc_batching will be false. That's
827 * OK, if the queue is under the request limit then requests need
828 * not count toward the nr_batch_requests limit. There will always
829 * be some limit enforced by BLK_BATCH_TIME.
830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (ioc_batching(q, ioc))
832 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100833
Jens Axboe1faa16d2009-04-06 14:48:01 +0200834 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return rq;
837}
838
839/*
840 * No available requests for this queue, unplug the device and wait for some
841 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700842 *
843 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 */
Jens Axboe165125e2007-07-24 09:28:11 +0200845static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200846 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200848 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 struct request *rq;
850
Jens Axboe7749a8d2006-12-13 13:02:26 +0100851 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700852 while (!rq) {
853 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200854 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 struct request_list *rl = &q->rq;
856
Jens Axboe1faa16d2009-04-06 14:48:01 +0200857 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 TASK_UNINTERRUPTIBLE);
859
Jens Axboe1faa16d2009-04-06 14:48:01 +0200860 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200862 __generic_unplug_device(q);
863 spin_unlock_irq(q->queue_lock);
864 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200866 /*
867 * After sleeping, we become a "batching" process and
868 * will be able to allocate at least one request, and
869 * up to a big batch of them for a small period time.
870 * See ioc_batching, ioc_set_batching
871 */
872 ioc = current_io_context(GFP_NOIO, q->node);
873 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100874
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200875 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200876 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200877
878 rq = get_request(q, rw_flags, bio, GFP_NOIO);
879 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 return rq;
882}
883
Jens Axboe165125e2007-07-24 09:28:11 +0200884struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
886 struct request *rq;
887
888 BUG_ON(rw != READ && rw != WRITE);
889
Nick Piggind6344532005-06-28 20:45:14 -0700890 spin_lock_irq(q->queue_lock);
891 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200892 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700893 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200894 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700895 if (!rq)
896 spin_unlock_irq(q->queue_lock);
897 }
898 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 return rq;
901}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902EXPORT_SYMBOL(blk_get_request);
903
904/**
Jens Axboedc72ef42006-07-20 14:54:05 +0200905 * blk_start_queueing - initiate dispatch of requests to device
906 * @q: request queue to kick into gear
907 *
908 * This is basically a helper to remove the need to know whether a queue
909 * is plugged or not if someone just wants to initiate dispatch of requests
Jens Axboe80a4b582008-10-14 09:51:06 +0200910 * for this queue. Should be used to start queueing on a device outside
911 * of ->request_fn() context. Also see @blk_run_queue.
Jens Axboedc72ef42006-07-20 14:54:05 +0200912 *
913 * The queue lock must be held with interrupts disabled.
914 */
Jens Axboe165125e2007-07-24 09:28:11 +0200915void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +0200916{
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200917 if (!blk_queue_plugged(q)) {
918 if (unlikely(blk_queue_stopped(q)))
919 return;
Jens Axboedc72ef42006-07-20 14:54:05 +0200920 q->request_fn(q);
Elias Oltmanns336c3d82008-10-01 16:02:33 +0200921 } else
Jens Axboedc72ef42006-07-20 14:54:05 +0200922 __generic_unplug_device(q);
923}
924EXPORT_SYMBOL(blk_start_queueing);
925
926/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 * blk_requeue_request - put a request back on queue
928 * @q: request queue where request should be inserted
929 * @rq: request to be inserted
930 *
931 * Description:
932 * Drivers often keep queueing requests until the hardware cannot accept
933 * more, when that condition happens we need to put the request back
934 * on the queue. Must be called with queue lock held.
935 */
Jens Axboe165125e2007-07-24 09:28:11 +0200936void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700938 blk_delete_timer(rq);
939 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100940 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (blk_rq_tagged(rq))
943 blk_queue_end_tag(q, rq);
944
945 elv_requeue_request(q, rq);
946}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947EXPORT_SYMBOL(blk_requeue_request);
948
949/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200950 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 * @q: request queue where request should be inserted
952 * @rq: request to be inserted
953 * @at_head: insert request at head or tail of queue
954 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 *
956 * Description:
957 * Many block devices need to execute commands asynchronously, so they don't
958 * block the whole kernel from preemption during request execution. This is
959 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200960 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
961 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 *
963 * We have the option of inserting the head or the tail of the queue.
964 * Typically we use the tail for new ioctls and so forth. We use the head
965 * of the queue for things like a QUEUE_FULL message from a device, or a
966 * host that is unable to accept a particular command.
967 */
Jens Axboe165125e2007-07-24 09:28:11 +0200968void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500969 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
Tejun Heo 867d1192005-04-24 02:06:05 -0500971 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 unsigned long flags;
973
974 /*
975 * tell I/O scheduler that this isn't a regular read/write (ie it
976 * must not attempt merges on this) and that it acts as a soft
977 * barrier
978 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200979 rq->cmd_type = REQ_TYPE_SPECIAL;
980 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 rq->special = data;
983
984 spin_lock_irqsave(q->queue_lock, flags);
985
986 /*
987 * If command is tagged, release the tag
988 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500989 if (blk_rq_tagged(rq))
990 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200992 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500993 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +0200994 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 spin_unlock_irqrestore(q->queue_lock, flags);
996}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997EXPORT_SYMBOL(blk_insert_request);
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/*
1000 * add-request adds a request to the linked list.
1001 * queue lock is held and interrupts disabled, as we muck with the
1002 * request queue list.
1003 */
Jens Axboe6728cb02008-01-31 13:03:55 +01001004static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001006 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 /*
1009 * elevator indicated where it wants this request to be
1010 * inserted at elevator_merge time
1011 */
1012 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1013}
Jens Axboe6728cb02008-01-31 13:03:55 +01001014
Tejun Heo074a7ac2008-08-25 19:56:14 +09001015static void part_round_stats_single(int cpu, struct hd_struct *part,
1016 unsigned long now)
1017{
1018 if (now == part->stamp)
1019 return;
1020
1021 if (part->in_flight) {
1022 __part_stat_add(cpu, part, time_in_queue,
1023 part->in_flight * (now - part->stamp));
1024 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1025 }
1026 part->stamp = now;
1027}
1028
1029/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001030 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1031 * @cpu: cpu number for stats access
1032 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 *
1034 * The average IO queue length and utilisation statistics are maintained
1035 * by observing the current state of the queue length and the amount of
1036 * time it has been in this state for.
1037 *
1038 * Normally, that accounting is done on IO completion, but that can result
1039 * in more than a second's worth of IO being accounted for within any one
1040 * second, leading to >100% utilisation. To deal with that, we call this
1041 * function to do a round-off before returning the results when reading
1042 * /proc/diskstats. This accounts immediately for all queue usage up to
1043 * the current jiffies and restarts the counters again.
1044 */
Tejun Heoc9959052008-08-25 19:47:21 +09001045void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001046{
1047 unsigned long now = jiffies;
1048
Tejun Heo074a7ac2008-08-25 19:56:14 +09001049 if (part->partno)
1050 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1051 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001052}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001053EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055/*
1056 * queue lock must be held
1057 */
Jens Axboe165125e2007-07-24 09:28:11 +02001058void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (unlikely(!q))
1061 return;
1062 if (unlikely(--req->ref_count))
1063 return;
1064
Tejun Heo8922e162005-10-20 16:23:44 +02001065 elv_completed_request(q, req);
1066
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001067 /* this is a bio leak */
1068 WARN_ON(req->bio != NULL);
1069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /*
1071 * Request may not have originated from ll_rw_blk. if not,
1072 * it didn't come out of our reserved rq pools
1073 */
Jens Axboe49171e52006-08-10 08:59:11 +02001074 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001075 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001076 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001079 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001082 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 }
1084}
Mike Christie6e39b692005-11-11 05:30:24 -06001085EXPORT_SYMBOL_GPL(__blk_put_request);
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087void blk_put_request(struct request *req)
1088{
Tejun Heo8922e162005-10-20 16:23:44 +02001089 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001090 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001092 spin_lock_irqsave(q->queue_lock, flags);
1093 __blk_put_request(q, req);
1094 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096EXPORT_SYMBOL(blk_put_request);
1097
Jens Axboe86db1e22008-01-29 14:53:40 +01001098void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001099{
Jens Axboec7c22e42008-09-13 20:26:01 +02001100 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001101 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001102
1103 /*
1104 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1105 */
Mike Christie6000a362008-08-19 18:45:30 -05001106 if (bio_rw_ahead(bio))
1107 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1108 REQ_FAILFAST_DRIVER);
1109 if (bio_failfast_dev(bio))
1110 req->cmd_flags |= REQ_FAILFAST_DEV;
1111 if (bio_failfast_transport(bio))
1112 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1113 if (bio_failfast_driver(bio))
1114 req->cmd_flags |= REQ_FAILFAST_DRIVER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001115
1116 /*
1117 * REQ_BARRIER implies no merging, but lets make it explicit
1118 */
David Woodhousefb2dce82008-08-05 18:01:53 +01001119 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001120 req->cmd_flags |= REQ_DISCARD;
1121 if (bio_barrier(bio))
1122 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001123 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001124 } else if (unlikely(bio_barrier(bio)))
1125 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01001126
Jens Axboeb31dc662006-06-13 08:26:10 +02001127 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001128 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001129 if (bio_rw_meta(bio))
1130 req->cmd_flags |= REQ_RW_META;
Jens Axboeaeb6faf2009-04-06 14:48:07 +02001131 if (bio_noidle(bio))
1132 req->cmd_flags |= REQ_NOIDLE;
Jens Axboeb31dc662006-06-13 08:26:10 +02001133
Tejun Heo52d9e672006-01-06 09:49:58 +01001134 req->errors = 0;
1135 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001136 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001137 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02001138 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001139}
1140
Jens Axboe644b2d92009-04-06 14:48:06 +02001141/*
1142 * Only disabling plugging for non-rotational devices if it does tagging
1143 * as well, otherwise we do need the proper merging
1144 */
1145static inline bool queue_should_plug(struct request_queue *q)
1146{
1147 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1148}
1149
Jens Axboe165125e2007-07-24 09:28:11 +02001150static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Nick Piggin450991b2005-06-28 20:45:13 -07001152 struct request *req;
Tejun Heoa7384672008-11-28 13:32:03 +09001153 int el_ret, nr_sectors;
Jens Axboe51da90f2006-07-18 04:14:45 +02001154 const unsigned short prio = bio_prio(bio);
1155 const int sync = bio_sync(bio);
Jens Axboe213d9412009-01-06 09:16:05 +01001156 const int unplug = bio_unplug(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001157 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 /*
1162 * low level driver can indicate that it wants pages above a
1163 * certain limit bounced to low memory (ie for highmem, or even
1164 * ISA dma in theory)
1165 */
1166 blk_queue_bounce(q, &bio);
1167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 spin_lock_irq(q->queue_lock);
1169
Tejun Heoa7384672008-11-28 13:32:03 +09001170 if (unlikely(bio_barrier(bio)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 goto get_rq;
1172
1173 el_ret = elv_merge(q, &req, bio);
1174 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001175 case ELEVATOR_BACK_MERGE:
1176 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Jens Axboe6728cb02008-01-31 13:03:55 +01001178 if (!ll_back_merge_fn(q, req, bio))
1179 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001181 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001182
Jens Axboe6728cb02008-01-31 13:03:55 +01001183 req->biotail->bi_next = bio;
1184 req->biotail = bio;
1185 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1186 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001187 if (!blk_rq_cpu_valid(req))
1188 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001189 drive_stat_acct(req, 0);
1190 if (!attempt_back_merge(q, req))
1191 elv_merged_request(q, req, el_ret);
1192 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Jens Axboe6728cb02008-01-31 13:03:55 +01001194 case ELEVATOR_FRONT_MERGE:
1195 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Jens Axboe6728cb02008-01-31 13:03:55 +01001197 if (!ll_front_merge_fn(q, req, bio))
1198 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001200 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001201
Jens Axboe6728cb02008-01-31 13:03:55 +01001202 bio->bi_next = req->bio;
1203 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Jens Axboe6728cb02008-01-31 13:03:55 +01001205 /*
1206 * may not be valid. if the low level driver said
1207 * it didn't need a bounce buffer then it better
1208 * not touch req->buffer either...
1209 */
1210 req->buffer = bio_data(bio);
1211 req->current_nr_sectors = bio_cur_sectors(bio);
1212 req->hard_cur_sectors = req->current_nr_sectors;
1213 req->sector = req->hard_sector = bio->bi_sector;
1214 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1215 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001216 if (!blk_rq_cpu_valid(req))
1217 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001218 drive_stat_acct(req, 0);
1219 if (!attempt_front_merge(q, req))
1220 elv_merged_request(q, req, el_ret);
1221 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Jens Axboe6728cb02008-01-31 13:03:55 +01001223 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1224 default:
1225 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001229 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001230 * This sync check and mask will be re-done in init_request_from_bio(),
1231 * but we need to set it earlier to expose the sync flag to the
1232 * rq allocator and io schedulers.
1233 */
1234 rw_flags = bio_data_dir(bio);
1235 if (sync)
1236 rw_flags |= REQ_RW_SYNC;
1237
1238 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001239 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001240 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001241 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001242 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001243
Nick Piggin450991b2005-06-28 20:45:13 -07001244 /*
1245 * After dropping the lock and possibly sleeping here, our request
1246 * may now be mergeable after it had proven unmergeable (above).
1247 * We don't worry about that case for efficiency. It won't happen
1248 * often, and the elevators are able to handle it.
1249 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001250 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Nick Piggin450991b2005-06-28 20:45:13 -07001252 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001253 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1254 bio_flagged(bio, BIO_CPU_AFFINE))
1255 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001256 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001257 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 add_request(q, req);
1259out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001260 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 spin_unlock_irq(q->queue_lock);
1263 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264}
1265
1266/*
1267 * If bio->bi_dev is a partition, remap the location
1268 */
1269static inline void blk_partition_remap(struct bio *bio)
1270{
1271 struct block_device *bdev = bio->bi_bdev;
1272
Jens Axboebf2de6f2007-09-27 13:01:25 +02001273 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 bio->bi_sector += p->start_sect;
1277 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001278
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001279 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001280 bdev->bd_dev, bio->bi_sector,
1281 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 }
1283}
1284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285static void handle_bad_sector(struct bio *bio)
1286{
1287 char b[BDEVNAME_SIZE];
1288
1289 printk(KERN_INFO "attempt to access beyond end of device\n");
1290 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1291 bdevname(bio->bi_bdev, b),
1292 bio->bi_rw,
1293 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1294 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1295
1296 set_bit(BIO_EOF, &bio->bi_flags);
1297}
1298
Akinobu Mitac17bb492006-12-08 02:39:46 -08001299#ifdef CONFIG_FAIL_MAKE_REQUEST
1300
1301static DECLARE_FAULT_ATTR(fail_make_request);
1302
1303static int __init setup_fail_make_request(char *str)
1304{
1305 return setup_fault_attr(&fail_make_request, str);
1306}
1307__setup("fail_make_request=", setup_fail_make_request);
1308
1309static int should_fail_request(struct bio *bio)
1310{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001311 struct hd_struct *part = bio->bi_bdev->bd_part;
1312
1313 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001314 return should_fail(&fail_make_request, bio->bi_size);
1315
1316 return 0;
1317}
1318
1319static int __init fail_make_request_debugfs(void)
1320{
1321 return init_fault_attr_dentries(&fail_make_request,
1322 "fail_make_request");
1323}
1324
1325late_initcall(fail_make_request_debugfs);
1326
1327#else /* CONFIG_FAIL_MAKE_REQUEST */
1328
1329static inline int should_fail_request(struct bio *bio)
1330{
1331 return 0;
1332}
1333
1334#endif /* CONFIG_FAIL_MAKE_REQUEST */
1335
Jens Axboec07e2b42007-07-18 13:27:58 +02001336/*
1337 * Check whether this bio extends beyond the end of the device.
1338 */
1339static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1340{
1341 sector_t maxsector;
1342
1343 if (!nr_sectors)
1344 return 0;
1345
1346 /* Test device or partition size, when known. */
1347 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1348 if (maxsector) {
1349 sector_t sector = bio->bi_sector;
1350
1351 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1352 /*
1353 * This may well happen - the kernel calls bread()
1354 * without checking the size of the device, e.g., when
1355 * mounting a device.
1356 */
1357 handle_bad_sector(bio);
1358 return 1;
1359 }
1360 }
1361
1362 return 0;
1363}
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001366 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 * @bio: The bio describing the location in memory and on the device.
1368 *
1369 * generic_make_request() is used to make I/O requests of block
1370 * devices. It is passed a &struct bio, which describes the I/O that needs
1371 * to be done.
1372 *
1373 * generic_make_request() does not return any status. The
1374 * success/failure status of the request, along with notification of
1375 * completion, is delivered asynchronously through the bio->bi_end_io
1376 * function described (one day) else where.
1377 *
1378 * The caller of generic_make_request must make sure that bi_io_vec
1379 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1380 * set to describe the device address, and the
1381 * bi_end_io and optionally bi_private are set to describe how
1382 * completion notification should be signaled.
1383 *
1384 * generic_make_request and the drivers it calls may use bi_next if this
1385 * bio happens to be merged with someone else, and may change bi_dev and
1386 * bi_sector for remaps as it sees fit. So the values of these fields
1387 * should NOT be depended on after the call to generic_make_request.
1388 */
Neil Brownd89d8792007-05-01 09:53:42 +02001389static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390{
Jens Axboe165125e2007-07-24 09:28:11 +02001391 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001392 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001394 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001395 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
1397 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Jens Axboec07e2b42007-07-18 13:27:58 +02001399 if (bio_check_eod(bio, nr_sectors))
1400 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402 /*
1403 * Resolve the mapping until finished. (drivers are
1404 * still free to implement/resolve their own stacking
1405 * by explicitly returning 0)
1406 *
1407 * NOTE: we don't repeat the blk_size check for each new device.
1408 * Stacking drivers are expected to know what they are doing.
1409 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001410 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001411 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 do {
1413 char b[BDEVNAME_SIZE];
1414
1415 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001416 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 printk(KERN_ERR
1418 "generic_make_request: Trying to access "
1419 "nonexistent block-device %s (%Lu)\n",
1420 bdevname(bio->bi_bdev, b),
1421 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001422 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
1424
Jens Axboe4fa253f2007-07-18 13:13:10 +02001425 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001426 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 bdevname(bio->bi_bdev, b),
1428 bio_sectors(bio),
1429 q->max_hw_sectors);
1430 goto end_io;
1431 }
1432
Nick Pigginfde6ad22005-06-23 00:08:53 -07001433 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 goto end_io;
1435
Akinobu Mitac17bb492006-12-08 02:39:46 -08001436 if (should_fail_request(bio))
1437 goto end_io;
1438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 /*
1440 * If this device has partitions, remap block n
1441 * of partition p to block n+start(p) of the disk.
1442 */
1443 blk_partition_remap(bio);
1444
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001445 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1446 goto end_io;
1447
NeilBrown5ddfe962006-10-30 22:07:21 -08001448 if (old_sector != -1)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001449 trace_block_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001450 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001451
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001452 trace_block_bio_queue(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001453
NeilBrown5ddfe962006-10-30 22:07:21 -08001454 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001455 old_dev = bio->bi_bdev->bd_dev;
1456
Jens Axboec07e2b42007-07-18 13:27:58 +02001457 if (bio_check_eod(bio, nr_sectors))
1458 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001459
1460 if (bio_discard(bio) && !q->prepare_discard_fn) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001461 err = -EOPNOTSUPP;
1462 goto end_io;
1463 }
Jens Axboecec07072009-01-13 15:28:32 +01001464 if (bio_barrier(bio) && bio_has_data(bio) &&
1465 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1466 err = -EOPNOTSUPP;
1467 goto end_io;
1468 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 ret = q->make_request_fn(q, bio);
1471 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001472
1473 return;
1474
1475end_io:
1476 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477}
1478
Neil Brownd89d8792007-05-01 09:53:42 +02001479/*
1480 * We only want one ->make_request_fn to be active at a time,
1481 * else stack usage with stacked devices could be a problem.
1482 * So use current->bio_{list,tail} to keep a list of requests
1483 * submited by a make_request_fn function.
1484 * current->bio_tail is also used as a flag to say if
1485 * generic_make_request is currently active in this task or not.
1486 * If it is NULL, then no make_request is active. If it is non-NULL,
1487 * then a make_request is active, and new requests should be added
1488 * at the tail
1489 */
1490void generic_make_request(struct bio *bio)
1491{
1492 if (current->bio_tail) {
1493 /* make_request is active */
1494 *(current->bio_tail) = bio;
1495 bio->bi_next = NULL;
1496 current->bio_tail = &bio->bi_next;
1497 return;
1498 }
1499 /* following loop may be a bit non-obvious, and so deserves some
1500 * explanation.
1501 * Before entering the loop, bio->bi_next is NULL (as all callers
1502 * ensure that) so we have a list with a single bio.
1503 * We pretend that we have just taken it off a longer list, so
1504 * we assign bio_list to the next (which is NULL) and bio_tail
1505 * to &bio_list, thus initialising the bio_list of new bios to be
1506 * added. __generic_make_request may indeed add some more bios
1507 * through a recursive call to generic_make_request. If it
1508 * did, we find a non-NULL value in bio_list and re-enter the loop
1509 * from the top. In this case we really did just take the bio
1510 * of the top of the list (no pretending) and so fixup bio_list and
1511 * bio_tail or bi_next, and call into __generic_make_request again.
1512 *
1513 * The loop was structured like this to make only one call to
1514 * __generic_make_request (which is important as it is large and
1515 * inlined) and to keep the structure simple.
1516 */
1517 BUG_ON(bio->bi_next);
1518 do {
1519 current->bio_list = bio->bi_next;
1520 if (bio->bi_next == NULL)
1521 current->bio_tail = &current->bio_list;
1522 else
1523 bio->bi_next = NULL;
1524 __generic_make_request(bio);
1525 bio = current->bio_list;
1526 } while (bio);
1527 current->bio_tail = NULL; /* deactivate */
1528}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529EXPORT_SYMBOL(generic_make_request);
1530
1531/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001532 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1534 * @bio: The &struct bio which describes the I/O
1535 *
1536 * submit_bio() is very similar in purpose to generic_make_request(), and
1537 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001538 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 *
1540 */
1541void submit_bio(int rw, struct bio *bio)
1542{
1543 int count = bio_sectors(bio);
1544
Jens Axboe22e2c502005-06-27 10:55:12 +02001545 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Jens Axboebf2de6f2007-09-27 13:01:25 +02001547 /*
1548 * If it's a regular read/write or a barrier with data attached,
1549 * go through the normal accounting stuff before submission.
1550 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001551 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001552 if (rw & WRITE) {
1553 count_vm_events(PGPGOUT, count);
1554 } else {
1555 task_io_account_read(bio->bi_size);
1556 count_vm_events(PGPGIN, count);
1557 }
1558
1559 if (unlikely(block_dump)) {
1560 char b[BDEVNAME_SIZE];
1561 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001562 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001563 (rw & WRITE) ? "WRITE" : "READ",
1564 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001565 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 }
1568
1569 generic_make_request(bio);
1570}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571EXPORT_SYMBOL(submit_bio);
1572
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001573/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001574 * blk_rq_check_limits - Helper function to check a request for the queue limit
1575 * @q: the queue
1576 * @rq: the request being checked
1577 *
1578 * Description:
1579 * @rq may have been made based on weaker limitations of upper-level queues
1580 * in request stacking drivers, and it may violate the limitation of @q.
1581 * Since the block layer and the underlying device driver trust @rq
1582 * after it is inserted to @q, it should be checked against @q before
1583 * the insertion using this generic function.
1584 *
1585 * This function should also be useful for request stacking drivers
1586 * in some cases below, so export this fuction.
1587 * Request stacking drivers like request-based dm may change the queue
1588 * limits while requests are in the queue (e.g. dm's table swapping).
1589 * Such request stacking drivers should check those requests agaist
1590 * the new queue limits again when they dispatch those requests,
1591 * although such checkings are also done against the old queue limits
1592 * when submitting requests.
1593 */
1594int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1595{
1596 if (rq->nr_sectors > q->max_sectors ||
1597 rq->data_len > q->max_hw_sectors << 9) {
1598 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1599 return -EIO;
1600 }
1601
1602 /*
1603 * queue's settings related to segment counting like q->bounce_pfn
1604 * may differ from that of other stacking queues.
1605 * Recalculate it to check the request correctly on this queue's
1606 * limitation.
1607 */
1608 blk_recalc_rq_segments(rq);
1609 if (rq->nr_phys_segments > q->max_phys_segments ||
1610 rq->nr_phys_segments > q->max_hw_segments) {
1611 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1612 return -EIO;
1613 }
1614
1615 return 0;
1616}
1617EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1618
1619/**
1620 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1621 * @q: the queue to submit the request
1622 * @rq: the request being queued
1623 */
1624int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1625{
1626 unsigned long flags;
1627
1628 if (blk_rq_check_limits(q, rq))
1629 return -EIO;
1630
1631#ifdef CONFIG_FAIL_MAKE_REQUEST
1632 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1633 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1634 return -EIO;
1635#endif
1636
1637 spin_lock_irqsave(q->queue_lock, flags);
1638
1639 /*
1640 * Submitting request must be dequeued before calling this function
1641 * because it will be linked to another request_queue
1642 */
1643 BUG_ON(blk_queued_rq(rq));
1644
1645 drive_stat_acct(rq, 1);
1646 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1647
1648 spin_unlock_irqrestore(q->queue_lock, flags);
1649
1650 return 0;
1651}
1652EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1653
1654/**
Tejun Heo53a08802008-12-03 12:41:26 +01001655 * blkdev_dequeue_request - dequeue request and start timeout timer
1656 * @req: request to dequeue
1657 *
1658 * Dequeue @req and start timeout timer on it. This hands off the
1659 * request to the driver.
1660 *
1661 * Block internal functions which don't want to start timer should
1662 * call elv_dequeue_request().
1663 */
1664void blkdev_dequeue_request(struct request *req)
1665{
1666 elv_dequeue_request(req->q, req);
1667
1668 /*
1669 * We are now handing the request to the hardware, add the
1670 * timeout handler.
1671 */
1672 blk_add_timer(req);
1673}
1674EXPORT_SYMBOL(blkdev_dequeue_request);
1675
Jens Axboebc58ba92009-01-23 10:54:44 +01001676static void blk_account_io_completion(struct request *req, unsigned int bytes)
1677{
Jerome Marchand26308ea2009-03-27 10:31:51 +01001678 if (!blk_do_io_stat(req))
Jens Axboebc58ba92009-01-23 10:54:44 +01001679 return;
1680
1681 if (blk_fs_request(req)) {
1682 const int rw = rq_data_dir(req);
1683 struct hd_struct *part;
1684 int cpu;
1685
1686 cpu = part_stat_lock();
1687 part = disk_map_sector_rcu(req->rq_disk, req->sector);
1688 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1689 part_stat_unlock();
1690 }
1691}
1692
1693static void blk_account_io_done(struct request *req)
1694{
Jerome Marchand26308ea2009-03-27 10:31:51 +01001695 if (!blk_do_io_stat(req))
Jens Axboebc58ba92009-01-23 10:54:44 +01001696 return;
1697
1698 /*
1699 * Account IO completion. bar_rq isn't accounted as a normal
1700 * IO on queueing nor completion. Accounting the containing
1701 * request is enough.
1702 */
1703 if (blk_fs_request(req) && req != &req->q->bar_rq) {
1704 unsigned long duration = jiffies - req->start_time;
1705 const int rw = rq_data_dir(req);
1706 struct hd_struct *part;
1707 int cpu;
1708
1709 cpu = part_stat_lock();
Jerome Marchand26308ea2009-03-27 10:31:51 +01001710 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Jens Axboebc58ba92009-01-23 10:54:44 +01001711
1712 part_stat_inc(cpu, part, ios[rw]);
1713 part_stat_add(cpu, part, ticks[rw], duration);
1714 part_round_stats(cpu, part);
1715 part_dec_in_flight(part);
1716
1717 part_stat_unlock();
1718 }
1719}
1720
Tejun Heo53a08802008-12-03 12:41:26 +01001721/**
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001722 * __end_that_request_first - end I/O on a request
1723 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001724 * @error: %0 for success, < %0 for error
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001725 * @nr_bytes: number of bytes to complete
1726 *
1727 * Description:
1728 * Ends I/O on a number of bytes attached to @req, and sets it up
1729 * for the next range of segments (if any) in the cluster.
1730 *
1731 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001732 * %0 - we are done with this request, call end_that_request_last()
1733 * %1 - still buffers pending for this request
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001734 **/
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001735static int __end_that_request_first(struct request *req, int error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 int nr_bytes)
1737{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001738 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 struct bio *bio;
1740
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001741 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 /*
Randy Dunlap710027a2008-08-19 20:13:11 +02001744 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 * sense key with us all the way through
1746 */
1747 if (!blk_pc_request(req))
1748 req->errors = 0;
1749
Jens Axboe6728cb02008-01-31 13:03:55 +01001750 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1751 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 req->rq_disk ? req->rq_disk->disk_name : "?",
1753 (unsigned long long)req->sector);
1754 }
1755
Jens Axboebc58ba92009-01-23 10:54:44 +01001756 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01001757
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 total_bytes = bio_nbytes = 0;
1759 while ((bio = req->bio) != NULL) {
1760 int nbytes;
1761
1762 if (nr_bytes >= bio->bi_size) {
1763 req->bio = bio->bi_next;
1764 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001765 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 next_idx = 0;
1767 bio_nbytes = 0;
1768 } else {
1769 int idx = bio->bi_idx + next_idx;
1770
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02001771 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001773 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02001774 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 break;
1776 }
1777
1778 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1779 BIO_BUG_ON(nbytes > bio->bi_size);
1780
1781 /*
1782 * not a complete bvec done
1783 */
1784 if (unlikely(nbytes > nr_bytes)) {
1785 bio_nbytes += nr_bytes;
1786 total_bytes += nr_bytes;
1787 break;
1788 }
1789
1790 /*
1791 * advance to the next vector
1792 */
1793 next_idx++;
1794 bio_nbytes += nbytes;
1795 }
1796
1797 total_bytes += nbytes;
1798 nr_bytes -= nbytes;
1799
Jens Axboe6728cb02008-01-31 13:03:55 +01001800 bio = req->bio;
1801 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 /*
1803 * end more in this run, or just return 'not-done'
1804 */
1805 if (unlikely(nr_bytes <= 0))
1806 break;
1807 }
1808 }
1809
1810 /*
1811 * completely done
1812 */
1813 if (!req->bio)
1814 return 0;
1815
1816 /*
1817 * if the request wasn't completed, update state
1818 */
1819 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001820 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 bio->bi_idx += next_idx;
1822 bio_iovec(bio)->bv_offset += nr_bytes;
1823 bio_iovec(bio)->bv_len -= nr_bytes;
1824 }
1825
1826 blk_recalc_rq_sectors(req, total_bytes >> 9);
1827 blk_recalc_rq_segments(req);
1828 return 1;
1829}
1830
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831/*
1832 * queue lock must be held
1833 */
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001834static void end_that_request_last(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001836 if (blk_rq_tagged(req))
1837 blk_queue_end_tag(req->q, req);
1838
1839 if (blk_queued_rq(req))
Tejun Heo53a08802008-12-03 12:41:26 +01001840 elv_dequeue_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
1842 if (unlikely(laptop_mode) && blk_fs_request(req))
1843 laptop_io_completion();
1844
Mike Andersone78042e2008-10-30 02:16:20 -07001845 blk_delete_timer(req);
1846
Jens Axboebc58ba92009-01-23 10:54:44 +01001847 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001850 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001851 else {
1852 if (blk_bidi_rq(req))
1853 __blk_put_request(req->next_rq->q, req->next_rq);
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 }
1857}
1858
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001859/**
1860 * blk_rq_bytes - Returns bytes left to complete in the entire request
Randy Dunlap5d87a052008-02-20 09:01:22 +01001861 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001862 **/
1863unsigned int blk_rq_bytes(struct request *rq)
Jens Axboea0cd1282007-09-21 10:41:07 +02001864{
1865 if (blk_fs_request(rq))
1866 return rq->hard_nr_sectors << 9;
1867
1868 return rq->data_len;
1869}
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001870EXPORT_SYMBOL_GPL(blk_rq_bytes);
1871
1872/**
1873 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
Randy Dunlap5d87a052008-02-20 09:01:22 +01001874 * @rq: the request being processed
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001875 **/
1876unsigned int blk_rq_cur_bytes(struct request *rq)
1877{
1878 if (blk_fs_request(rq))
1879 return rq->current_nr_sectors << 9;
1880
1881 if (rq->bio)
1882 return rq->bio->bi_size;
1883
1884 return rq->data_len;
1885}
1886EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
Jens Axboea0cd1282007-09-21 10:41:07 +02001887
1888/**
Jens Axboea0cd1282007-09-21 10:41:07 +02001889 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07001890 * @req: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001891 * @uptodate: error value or %0/%1 uptodate flag
Jens Axboea0cd1282007-09-21 10:41:07 +02001892 *
1893 * Description:
1894 * Ends I/O on the current segment of a request. If that is the only
1895 * remaining segment, the request is also completed and freed.
1896 *
Randy Dunlap710027a2008-08-19 20:13:11 +02001897 * This is a remnant of how older block drivers handled I/O completions.
1898 * Modern drivers typically end I/O on the full request in one go, unless
Jens Axboea0cd1282007-09-21 10:41:07 +02001899 * they have a residual value to account for. For that case this function
1900 * isn't really useful, unless the residual just happens to be the
1901 * full current segment. In other words, don't use this function in new
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001902 * code. Use blk_end_request() or __blk_end_request() to end a request.
Jens Axboea0cd1282007-09-21 10:41:07 +02001903 **/
1904void end_request(struct request *req, int uptodate)
1905{
Kiyoshi Uedad00e29f2008-10-01 10:14:46 -04001906 int error = 0;
1907
1908 if (uptodate <= 0)
1909 error = uptodate ? uptodate : -EIO;
1910
1911 __blk_end_request(req, error, req->hard_cur_sectors << 9);
Jens Axboea0cd1282007-09-21 10:41:07 +02001912}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913EXPORT_SYMBOL(end_request);
1914
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001915static int end_that_request_data(struct request *rq, int error,
1916 unsigned int nr_bytes, unsigned int bidi_bytes)
1917{
1918 if (rq->bio) {
1919 if (__end_that_request_first(rq, error, nr_bytes))
1920 return 1;
1921
1922 /* Bidi request must be completed as a whole */
1923 if (blk_bidi_rq(rq) &&
1924 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1925 return 1;
1926 }
1927
1928 return 0;
1929}
1930
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001931/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001932 * blk_end_io - Generic end_io function to complete a request.
1933 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001934 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001935 * @nr_bytes: number of bytes to complete @rq
1936 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001937 * @drv_callback: function called between completion of bios in the request
1938 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02001939 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001940 * completion of the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001941 *
1942 * Description:
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001943 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001944 * If @rq has leftover, sets it up for the next range of segments.
1945 *
1946 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001947 * %0 - we are done with this request
1948 * %1 - this request is not freed yet, it still has pending buffers.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001949 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001950static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1951 unsigned int bidi_bytes,
1952 int (drv_callback)(struct request *))
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001953{
1954 struct request_queue *q = rq->q;
1955 unsigned long flags = 0UL;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001956
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04001957 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1958 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001959
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001960 /* Special feature for tricky drivers */
1961 if (drv_callback && drv_callback(rq))
1962 return 1;
1963
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001964 add_disk_randomness(rq->rq_disk);
1965
1966 spin_lock_irqsave(q->queue_lock, flags);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001967 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001968 spin_unlock_irqrestore(q->queue_lock, flags);
1969
1970 return 0;
1971}
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001972
1973/**
1974 * blk_end_request - Helper function for drivers to complete the request.
1975 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001976 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001977 * @nr_bytes: number of bytes to complete
1978 *
1979 * Description:
1980 * Ends I/O on a number of bytes attached to @rq.
1981 * If @rq has leftover, sets it up for the next range of segments.
1982 *
1983 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02001984 * %0 - we are done with this request
1985 * %1 - still buffers pending for this request
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001986 **/
Jens Axboe22b13212008-01-31 12:36:19 +01001987int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001988{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001989 return blk_end_io(rq, error, nr_bytes, 0, NULL);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05001990}
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001991EXPORT_SYMBOL_GPL(blk_end_request);
1992
1993/**
1994 * __blk_end_request - Helper function for drivers to complete the request.
1995 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001996 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05001997 * @nr_bytes: number of bytes to complete
1998 *
1999 * Description:
2000 * Must be called with queue lock held unlike blk_end_request().
2001 *
2002 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002003 * %0 - we are done with this request
2004 * %1 - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002005 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002006int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002007{
Jens Axboe60540162008-08-26 13:34:34 +02002008 if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
Jens Axboe051cc392008-08-08 11:06:45 +02002009 return 1;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002010
2011 add_disk_randomness(rq->rq_disk);
2012
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002013 end_that_request_last(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002014
2015 return 0;
2016}
2017EXPORT_SYMBOL_GPL(__blk_end_request);
2018
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002019/**
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002020 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2021 * @rq: the bidi request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02002022 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002023 * @nr_bytes: number of bytes to complete @rq
2024 * @bidi_bytes: number of bytes to complete @rq->next_rq
2025 *
2026 * Description:
2027 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2028 *
2029 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002030 * %0 - we are done with this request
2031 * %1 - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002032 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002033int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
2034 unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002035{
2036 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
2037}
2038EXPORT_SYMBOL_GPL(blk_end_bidi_request);
2039
2040/**
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002041 * blk_update_request - Special helper function for request stacking drivers
2042 * @rq: the request being processed
2043 * @error: %0 for success, < %0 for error
2044 * @nr_bytes: number of bytes to complete @rq
2045 *
2046 * Description:
2047 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2048 * the request structure even if @rq doesn't have leftover.
2049 * If @rq has leftover, sets it up for the next range of segments.
2050 *
2051 * This special helper function is only for request stacking drivers
2052 * (e.g. request-based dm) so that they can handle partial completion.
2053 * Actual device drivers should use blk_end_request instead.
2054 */
2055void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2056{
2057 if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2058 /*
2059 * These members are not updated in end_that_request_data()
2060 * when all bios are completed.
2061 * Update them so that the request stacking driver can find
2062 * how many bytes remain in the request later.
2063 */
2064 rq->nr_sectors = rq->hard_nr_sectors = 0;
2065 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2066 }
2067}
2068EXPORT_SYMBOL_GPL(blk_update_request);
2069
2070/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002071 * blk_end_request_callback - Special helper function for tricky drivers
2072 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02002073 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002074 * @nr_bytes: number of bytes to complete
2075 * @drv_callback: function called between completion of bios in the request
2076 * and completion of the request.
Randy Dunlap710027a2008-08-19 20:13:11 +02002077 * If the callback returns non %0, this helper returns without
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002078 * completion of the request.
2079 *
2080 * Description:
2081 * Ends I/O on a number of bytes attached to @rq.
2082 * If @rq has leftover, sets it up for the next range of segments.
2083 *
2084 * This special helper function is used only for existing tricky drivers.
2085 * (e.g. cdrom_newpc_intr() of ide-cd)
2086 * This interface will be removed when such drivers are rewritten.
2087 * Don't use this interface in other places anymore.
2088 *
2089 * Return:
Randy Dunlap710027a2008-08-19 20:13:11 +02002090 * %0 - we are done with this request
2091 * %1 - this request is not freed yet.
2092 * this request still has pending buffers or
2093 * the driver doesn't want to finish this request yet.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002094 **/
Jens Axboe22b13212008-01-31 12:36:19 +01002095int blk_end_request_callback(struct request *rq, int error,
2096 unsigned int nr_bytes,
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002097 int (drv_callback)(struct request *))
2098{
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002099 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002100}
2101EXPORT_SYMBOL_GPL(blk_end_request_callback);
2102
Jens Axboe86db1e22008-01-29 14:53:40 +01002103void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2104 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
David Woodhoused628eae2008-08-09 16:22:17 +01002106 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2107 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002108 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
David Woodhousefb2dce82008-08-05 18:01:53 +01002110 if (bio_has_data(bio)) {
2111 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002112 rq->buffer = bio_data(bio);
2113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 rq->current_nr_sectors = bio_cur_sectors(bio);
2115 rq->hard_cur_sectors = rq->current_nr_sectors;
2116 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002117 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
NeilBrown66846572007-08-16 13:31:28 +02002121 if (bio->bi_bdev)
2122 rq->rq_disk = bio->bi_bdev->bd_disk;
2123}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002125/**
2126 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2127 * @q : the queue of the device being checked
2128 *
2129 * Description:
2130 * Check if underlying low-level drivers of a device are busy.
2131 * If the drivers want to export their busy state, they must set own
2132 * exporting function using blk_queue_lld_busy() first.
2133 *
2134 * Basically, this function is used only by request stacking drivers
2135 * to stop dispatching requests to underlying devices when underlying
2136 * devices are busy. This behavior helps more I/O merging on the queue
2137 * of the request stacking driver and prevents I/O throughput regression
2138 * on burst I/O load.
2139 *
2140 * Return:
2141 * 0 - Not busy (The request stacking driver should dispatch request)
2142 * 1 - Busy (The request stacking driver should stop dispatching request)
2143 */
2144int blk_lld_busy(struct request_queue *q)
2145{
2146 if (q->lld_busy_fn)
2147 return q->lld_busy_fn(q);
2148
2149 return 0;
2150}
2151EXPORT_SYMBOL_GPL(blk_lld_busy);
2152
Jens Axboe18887ad2008-07-28 13:08:45 +02002153int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154{
2155 return queue_work(kblockd_workqueue, work);
2156}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157EXPORT_SYMBOL(kblockd_schedule_work);
2158
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159int __init blk_dev_init(void)
2160{
2161 kblockd_workqueue = create_workqueue("kblockd");
2162 if (!kblockd_workqueue)
2163 panic("Failed to create kblockd\n");
2164
2165 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002166 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167
Jens Axboe8324aa92008-01-29 14:51:59 +01002168 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002169 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 return 0;
2172}
2173