blob: f84cce42fc58da2dc0bc48186cbd9f6bd1d7dfc3 [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>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Li Zefan55782132009-06-09 13:43:05 +080030
31#define CREATE_TRACE_POINTS
32#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Jens Axboe8324aa92008-01-29 14:51:59 +010034#include "blk.h"
35
Ingo Molnar0bfc2452008-11-26 11:59:56 +010036EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020037EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080038EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010039
Jens Axboe165125e2007-07-24 09:28:11 +020040static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42/*
43 * For the allocated request tables
44 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010045static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47/*
48 * For queue allocation
49 */
Jens Axboe6728cb02008-01-31 13:03:55 +010050struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * Controlling structure to kblockd
54 */
Jens Axboeff856ba2006-01-09 16:02:34 +010055static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Jens Axboe26b82562008-01-29 13:54:41 +010057static void drive_stat_acct(struct request *rq, int new_io)
58{
Jens Axboe28f13702008-05-07 10:15:46 +020059 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010060 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090061 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010062
Jens Axboec2553b52009-04-24 08:10:11 +020063 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010064 return;
65
Tejun Heo074a7ac2008-08-25 19:56:14 +090066 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +090067 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Tejun Heoc9959052008-08-25 19:47:21 +090068
Jens Axboe28f13702008-05-07 10:15:46 +020069 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 part_stat_inc(cpu, part, merges[rw]);
Jens Axboe28f13702008-05-07 10:15:46 +020071 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090072 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020073 part_inc_in_flight(part, rw);
Jens Axboe26b82562008-01-29 13:54:41 +010074 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020075
Tejun Heo074a7ac2008-08-25 19:56:14 +090076 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010077}
78
Jens Axboe8324aa92008-01-29 14:51:59 +010079void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
81 int nr;
82
83 nr = q->nr_requests - (q->nr_requests / 8) + 1;
84 if (nr > q->nr_requests)
85 nr = q->nr_requests;
86 q->nr_congestion_on = nr;
87
88 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
89 if (nr < 1)
90 nr = 1;
91 q->nr_congestion_off = nr;
92}
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/**
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
96 * @bdev: device
97 *
98 * Locates the passed device's request queue and returns the address of its
99 * backing_dev_info
100 *
101 * Will return NULL if the request queue cannot be located.
102 */
103struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
104{
105 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200106 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108 if (q)
109 ret = &q->backing_dev_info;
110 return ret;
111}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112EXPORT_SYMBOL(blk_get_backing_dev_info);
113
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200114void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200116 memset(rq, 0, sizeof(*rq));
117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700119 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200120 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100121 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900122 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200123 INIT_HLIST_NODE(&rq->hash);
124 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200125 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800126 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100127 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100128 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900129 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700130 set_start_time_ns(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200132EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
NeilBrown5bb23a62007-09-27 12:46:13 +0200134static void req_bio_endio(struct request *rq, struct bio *bio,
135 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100136{
Jens Axboe165125e2007-07-24 09:28:11 +0200137 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100138
NeilBrown5bb23a62007-09-27 12:46:13 +0200139 if (&q->bar_rq != rq) {
140 if (error)
141 clear_bit(BIO_UPTODATE, &bio->bi_flags);
142 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
143 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100144
NeilBrown5bb23a62007-09-27 12:46:13 +0200145 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100146 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700147 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200148 nbytes = bio->bi_size;
149 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100150
Keith Mannthey08bafc02008-11-25 10:24:35 +0100151 if (unlikely(rq->cmd_flags & REQ_QUIET))
152 set_bit(BIO_QUIET, &bio->bi_flags);
153
NeilBrown5bb23a62007-09-27 12:46:13 +0200154 bio->bi_size -= nbytes;
155 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200156
157 if (bio_integrity(bio))
158 bio_integrity_advance(bio, nbytes);
159
NeilBrown5bb23a62007-09-27 12:46:13 +0200160 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200161 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200162 } else {
163
164 /*
165 * Okay, this is the barrier request in progress, just
166 * record the error;
167 */
168 if (error && !q->orderr)
169 q->orderr = error;
170 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100171}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173void blk_dump_rq_flags(struct request *rq, char *msg)
174{
175 int bit;
176
Jens Axboe6728cb02008-01-31 13:03:55 +0100177 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200178 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
179 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Tejun Heo83096eb2009-05-07 22:24:39 +0900181 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
182 (unsigned long long)blk_rq_pos(rq),
183 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900185 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Jens Axboe4aff5e22006-08-10 08:44:47 +0200187 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100188 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200189 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 printk("%02x ", rq->cmd[bit]);
191 printk("\n");
192 }
193}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194EXPORT_SYMBOL(blk_dump_rq_flags);
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/*
197 * "plug" the device if there are no outstanding requests: this will
198 * force the transfer to start only after we have put all the requests
199 * on the list.
200 *
201 * This is called with interrupts off and no requests on the queue and
202 * with the queue lock held.
203 */
Jens Axboe165125e2007-07-24 09:28:11 +0200204void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
206 WARN_ON(!irqs_disabled());
207
208 /*
209 * don't plug a stopped queue, it must be paired with blk_start_queue()
210 * which will restart the queueing
211 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200212 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 return;
214
Jens Axboee48ec692008-07-03 13:18:54 +0200215 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100217 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220EXPORT_SYMBOL(blk_plug_device);
221
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200222/**
223 * blk_plug_device_unlocked - plug a device without queue lock held
224 * @q: The &struct request_queue to plug
225 *
226 * Description:
227 * Like @blk_plug_device(), but grabs the queue lock and disables
228 * interrupts.
229 **/
230void blk_plug_device_unlocked(struct request_queue *q)
231{
232 unsigned long flags;
233
234 spin_lock_irqsave(q->queue_lock, flags);
235 blk_plug_device(q);
236 spin_unlock_irqrestore(q->queue_lock, flags);
237}
238EXPORT_SYMBOL(blk_plug_device_unlocked);
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240/*
241 * remove the queue from the plugged list, if present. called with
242 * queue lock held and interrupts disabled.
243 */
Jens Axboe165125e2007-07-24 09:28:11 +0200244int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
246 WARN_ON(!irqs_disabled());
247
Jens Axboee48ec692008-07-03 13:18:54 +0200248 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 return 0;
250
251 del_timer(&q->unplug_timer);
252 return 1;
253}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254EXPORT_SYMBOL(blk_remove_plug);
255
256/*
257 * remove the plug and let it rip..
258 */
Jens Axboe165125e2007-07-24 09:28:11 +0200259void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200261 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200263 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return;
265
Jens Axboe22e2c502005-06-27 10:55:12 +0200266 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269/**
270 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200271 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 *
273 * Description:
274 * Linux uses plugging to build bigger requests queues before letting
275 * the device have at them. If a queue is plugged, the I/O scheduler
276 * is still adding and merging requests on the queue. Once the queue
277 * gets unplugged, the request_fn defined for the queue is invoked and
278 * transfers started.
279 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200280void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200282 if (blk_queue_plugged(q)) {
283 spin_lock_irq(q->queue_lock);
284 __generic_unplug_device(q);
285 spin_unlock_irq(q->queue_lock);
286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288EXPORT_SYMBOL(generic_unplug_device);
289
290static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
291 struct page *page)
292{
Jens Axboe165125e2007-07-24 09:28:11 +0200293 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500295 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
Jens Axboe86db1e22008-01-29 14:53:40 +0100298void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Jens Axboe165125e2007-07-24 09:28:11 +0200300 struct request_queue *q =
301 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100303 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 q->unplug_fn(q);
305}
306
Jens Axboe86db1e22008-01-29 14:53:40 +0100307void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
Jens Axboe165125e2007-07-24 09:28:11 +0200309 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100311 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200312 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
314
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500315void blk_unplug(struct request_queue *q)
316{
317 /*
318 * devices don't necessarily have an ->unplug_fn defined
319 */
320 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100321 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500322 q->unplug_fn(q);
323 }
324}
325EXPORT_SYMBOL(blk_unplug);
326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327/**
328 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200329 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 *
331 * Description:
332 * blk_start_queue() will clear the stop flag on the queue, and call
333 * the request_fn for the queue if it was in a stopped state when
334 * entered. Also see blk_stop_queue(). Queue lock must be held.
335 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200336void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200338 WARN_ON(!irqs_disabled());
339
Nick Piggin75ad23b2008-04-29 14:48:33 +0200340 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heoa538cd02009-04-23 11:05:17 +0900341 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343EXPORT_SYMBOL(blk_start_queue);
344
345/**
346 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200347 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 *
349 * Description:
350 * The Linux block layer assumes that a block driver will consume all
351 * entries on the request queue when the request_fn strategy is called.
352 * Often this will not happen, because of hardware limitations (queue
353 * depth settings). If a device driver gets a 'queue full' response,
354 * or if it simply chooses not to queue more I/O at one point, it can
355 * call this function to prevent the request_fn from being called until
356 * the driver has signalled it's ready to go again. This happens by calling
357 * blk_start_queue() to restart queue operations. Queue lock must be held.
358 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200359void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
361 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200362 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364EXPORT_SYMBOL(blk_stop_queue);
365
366/**
367 * blk_sync_queue - cancel any pending callbacks on a queue
368 * @q: the queue
369 *
370 * Description:
371 * The block layer may perform asynchronous callback activity
372 * on a queue, such as calling the unplug function after a timeout.
373 * A block device may call blk_sync_queue to ensure that any
374 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200375 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 * that its ->make_request_fn will not re-add plugging prior to calling
377 * this function.
378 *
379 */
380void blk_sync_queue(struct request_queue *q)
381{
382 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100383 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100384 cancel_work_sync(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386EXPORT_SYMBOL(blk_sync_queue);
387
388/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200389 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200391 *
392 * Description:
393 * See @blk_run_queue. This variant must be called with the queue lock
394 * held and interrupts disabled.
395 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200397void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200400
Tejun Heoa538cd02009-04-23 11:05:17 +0900401 if (unlikely(blk_queue_stopped(q)))
402 return;
403
404 if (elv_queue_empty(q))
405 return;
406
Jens Axboedac07ec2006-05-11 08:20:16 +0200407 /*
408 * Only recurse once to avoid overrunning the stack, let the unplug
409 * handling reinvoke the handler shortly if we already got there.
410 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900411 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
412 q->request_fn(q);
413 queue_flag_clear(QUEUE_FLAG_REENTER, q);
414 } else {
415 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
416 kblockd_schedule_work(q, &q->unplug_work);
417 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200418}
419EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200420
Nick Piggin75ad23b2008-04-29 14:48:33 +0200421/**
422 * blk_run_queue - run a single device queue
423 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200424 *
425 * Description:
426 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900427 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200428 */
429void blk_run_queue(struct request_queue *q)
430{
431 unsigned long flags;
432
433 spin_lock_irqsave(q->queue_lock, flags);
434 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 spin_unlock_irqrestore(q->queue_lock, flags);
436}
437EXPORT_SYMBOL(blk_run_queue);
438
Jens Axboe165125e2007-07-24 09:28:11 +0200439void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500440{
441 kobject_put(&q->kobj);
442}
Al Viro483f4af2006-03-18 18:34:37 -0500443
Jens Axboe6728cb02008-01-31 13:03:55 +0100444void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500445{
Jens Axboee3335de2008-09-18 09:22:54 -0700446 /*
447 * We know we have process context here, so we can be a little
448 * cautious and ensure that pending block actions on this device
449 * are done before moving on. Going into this function, we should
450 * not have processes doing IO to this device.
451 */
452 blk_sync_queue(q);
453
Matthew Garrett31373d02010-04-06 14:25:14 +0200454 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500455 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200456 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500457 mutex_unlock(&q->sysfs_lock);
458
459 if (q->elevator)
460 elevator_exit(q->elevator);
461
462 blk_put_queue(q);
463}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464EXPORT_SYMBOL(blk_cleanup_queue);
465
Jens Axboe165125e2007-07-24 09:28:11 +0200466static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct request_list *rl = &q->rq;
469
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400470 if (unlikely(rl->rq_pool))
471 return 0;
472
Jens Axboe1faa16d2009-04-06 14:48:01 +0200473 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
474 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200475 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200476 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
477 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Christoph Lameter19460892005-06-23 00:08:19 -0700479 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
480 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 if (!rl->rq_pool)
483 return -ENOMEM;
484
485 return 0;
486}
487
Jens Axboe165125e2007-07-24 09:28:11 +0200488struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Christoph Lameter19460892005-06-23 00:08:19 -0700490 return blk_alloc_queue_node(gfp_mask, -1);
491}
492EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Jens Axboe165125e2007-07-24 09:28:11 +0200494struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700495{
Jens Axboe165125e2007-07-24 09:28:11 +0200496 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700497 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700498
Jens Axboe8324aa92008-01-29 14:51:59 +0100499 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700500 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (!q)
502 return NULL;
503
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700504 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
505 q->backing_dev_info.unplug_io_data = q;
Jens Axboe0989a022009-06-12 14:42:56 +0200506 q->backing_dev_info.ra_pages =
507 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
508 q->backing_dev_info.state = 0;
509 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200510 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200511
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700512 err = bdi_init(&q->backing_dev_info);
513 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100514 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700515 return NULL;
516 }
517
Matthew Garrett31373d02010-04-06 14:25:14 +0200518 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
519 laptop_mode_timer_fn, (unsigned long) q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700521 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
522 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200523 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500524
Jens Axboe8324aa92008-01-29 14:51:59 +0100525 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Al Viro483f4af2006-03-18 18:34:37 -0500527 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700528 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return q;
531}
Christoph Lameter19460892005-06-23 00:08:19 -0700532EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534/**
535 * blk_init_queue - prepare a request queue for use with a block device
536 * @rfn: The function to be called to process requests that have been
537 * placed on the queue.
538 * @lock: Request queue spin lock
539 *
540 * Description:
541 * If a block device wishes to use the standard request handling procedures,
542 * which sorts requests and coalesces adjacent requests, then it must
543 * call blk_init_queue(). The function @rfn will be called when there
544 * are requests on the queue that need to be processed. If the device
545 * supports plugging, then @rfn may not be called immediately when requests
546 * are available on the queue, but may be called at some time later instead.
547 * Plugged queues are generally unplugged when a buffer belonging to one
548 * of the requests on the queue is needed, or due to memory pressure.
549 *
550 * @rfn is not required, or even expected, to remove all requests off the
551 * queue, but only as many as it can handle at a time. If it does leave
552 * requests on the queue, it is responsible for arranging that the requests
553 * get dealt with eventually.
554 *
555 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200556 * request queue; this lock will be taken also from interrupt context, so irq
557 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200559 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 * it didn't succeed.
561 *
562 * Note:
563 * blk_init_queue() must be paired with a blk_cleanup_queue() call
564 * when the block device is deactivated (such as at module unload).
565 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700566
Jens Axboe165125e2007-07-24 09:28:11 +0200567struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Christoph Lameter19460892005-06-23 00:08:19 -0700569 return blk_init_queue_node(rfn, lock, -1);
570}
571EXPORT_SYMBOL(blk_init_queue);
572
Jens Axboe165125e2007-07-24 09:28:11 +0200573struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700574blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
575{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600576 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600578 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
579 if (!uninit_q)
580 return NULL;
581
582 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
583 if (!q)
584 blk_cleanup_queue(uninit_q);
585
586 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200587}
588EXPORT_SYMBOL(blk_init_queue_node);
589
590struct request_queue *
591blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
592 spinlock_t *lock)
593{
594 return blk_init_allocated_queue_node(q, rfn, lock, -1);
595}
596EXPORT_SYMBOL(blk_init_allocated_queue);
597
598struct request_queue *
599blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
600 spinlock_t *lock, int node_id)
601{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (!q)
603 return NULL;
604
Christoph Lameter19460892005-06-23 00:08:19 -0700605 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600606 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500607 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 q->prep_rq_fn = NULL;
611 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100612 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 q->queue_lock = lock;
614
Jens Axboef3b144a2009-03-06 08:48:33 +0100615 /*
616 * This also sets hw/phys segments, boundary and size
617 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Alan Stern44ec9542007-02-20 11:01:57 -0500620 q->sg_reserved_size = INT_MAX;
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 /*
623 * all done
624 */
625 if (!elevator_init(q, NULL)) {
626 blk_queue_congestion_threshold(q);
627 return q;
628 }
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 return NULL;
631}
Mike Snitzer01effb02010-05-11 08:57:42 +0200632EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Jens Axboe165125e2007-07-24 09:28:11 +0200634int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700636 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500637 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return 0;
639 }
640
641 return 1;
642}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Jens Axboe165125e2007-07-24 09:28:11 +0200644static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200646 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200647 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 mempool_free(rq, q->rq.rq_pool);
649}
650
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200651static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200652blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
655
656 if (!rq)
657 return NULL;
658
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200659 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200660
Jerome Marchand42dad762009-04-22 14:01:49 +0200661 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Tejun Heocb98fc82005-10-28 08:29:39 +0200663 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200664 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200665 mempool_free(rq, q->rq.rq_pool);
666 return NULL;
667 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200668 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Tejun Heocb98fc82005-10-28 08:29:39 +0200671 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
674/*
675 * ioc_batching returns true if the ioc is a valid batching request and
676 * should be given priority access to a request.
677 */
Jens Axboe165125e2007-07-24 09:28:11 +0200678static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
680 if (!ioc)
681 return 0;
682
683 /*
684 * Make sure the process is able to allocate at least 1 request
685 * even if the batch times out, otherwise we could theoretically
686 * lose wakeups.
687 */
688 return ioc->nr_batch_requests == q->nr_batching ||
689 (ioc->nr_batch_requests > 0
690 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
691}
692
693/*
694 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
695 * will cause the process to be a "batcher" on all queues in the system. This
696 * is the behaviour we want though - once it gets a wakeup it should be given
697 * a nice run.
698 */
Jens Axboe165125e2007-07-24 09:28:11 +0200699static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
701 if (!ioc || ioc_batching(q, ioc))
702 return;
703
704 ioc->nr_batch_requests = q->nr_batching;
705 ioc->last_waited = jiffies;
706}
707
Jens Axboe1faa16d2009-04-06 14:48:01 +0200708static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709{
710 struct request_list *rl = &q->rq;
711
Jens Axboe1faa16d2009-04-06 14:48:01 +0200712 if (rl->count[sync] < queue_congestion_off_threshold(q))
713 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Jens Axboe1faa16d2009-04-06 14:48:01 +0200715 if (rl->count[sync] + 1 <= q->nr_requests) {
716 if (waitqueue_active(&rl->wait[sync]))
717 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Jens Axboe1faa16d2009-04-06 14:48:01 +0200719 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
721}
722
723/*
724 * A request has just been released. Account for it, update the full and
725 * congestion status, wake up any waiters. Called under q->queue_lock.
726 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200727static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 struct request_list *rl = &q->rq;
730
Jens Axboe1faa16d2009-04-06 14:48:01 +0200731 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200732 if (priv)
733 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Jens Axboe1faa16d2009-04-06 14:48:01 +0200735 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Jens Axboe1faa16d2009-04-06 14:48:01 +0200737 if (unlikely(rl->starved[sync ^ 1]))
738 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741/*
Nick Piggind6344532005-06-28 20:45:14 -0700742 * Get a free request, queue_lock must be held.
743 * Returns NULL on failure, with queue_lock held.
744 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 */
Jens Axboe165125e2007-07-24 09:28:11 +0200746static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100747 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
749 struct request *rq = NULL;
750 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100751 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200752 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100753 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Jens Axboe7749a8d2006-12-13 13:02:26 +0100755 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100756 if (may_queue == ELV_MQUEUE_NO)
757 goto rq_starved;
758
Jens Axboe1faa16d2009-04-06 14:48:01 +0200759 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
760 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200761 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100762 /*
763 * The queue will fill after this allocation, so set
764 * it as full, and mark this process as "batching".
765 * This process will be allowed to complete a batch of
766 * requests, others will be blocked.
767 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200768 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100769 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200770 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100771 } else {
772 if (may_queue != ELV_MQUEUE_MUST
773 && !ioc_batching(q, ioc)) {
774 /*
775 * The queue is full and the allocating
776 * process is not a "batcher", and not
777 * exempted by the IO scheduler
778 */
779 goto out;
780 }
781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200783 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 }
785
Jens Axboe082cf692005-06-28 16:35:11 +0200786 /*
787 * Only allow batching queuers to allocate up to 50% over the defined
788 * limit of requests, otherwise we could have thousands of requests
789 * allocated with any setting of ->nr_requests
790 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200791 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200792 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100793
Jens Axboe1faa16d2009-04-06 14:48:01 +0200794 rl->count[is_sync]++;
795 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200796
Jens Axboe64521d12005-10-28 08:30:39 +0200797 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200798 if (priv)
799 rl->elvpriv++;
800
Jerome Marchand42dad762009-04-22 14:01:49 +0200801 if (blk_queue_io_stat(q))
802 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 spin_unlock_irq(q->queue_lock);
804
Jens Axboe7749a8d2006-12-13 13:02:26 +0100805 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100806 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 /*
808 * Allocation failed presumably due to memory. Undo anything
809 * we might have messed up.
810 *
811 * Allocating task should really be put onto the front of the
812 * wait queue, but this is pretty rare.
813 */
814 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200815 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817 /*
818 * in the very unlikely event that allocation failed and no
819 * requests for this direction was pending, mark us starved
820 * so that freeing of a request in the other direction will
821 * notice us. another possible fix would be to split the
822 * rq mempool into READ and WRITE
823 */
824rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200825 if (unlikely(rl->count[is_sync] == 0))
826 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 goto out;
829 }
830
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100831 /*
832 * ioc may be NULL here, and ioc_batching will be false. That's
833 * OK, if the queue is under the request limit then requests need
834 * not count toward the nr_batch_requests limit. There will always
835 * be some limit enforced by BLK_BATCH_TIME.
836 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (ioc_batching(q, ioc))
838 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100839
Jens Axboe1faa16d2009-04-06 14:48:01 +0200840 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 return rq;
843}
844
845/*
846 * No available requests for this queue, unplug the device and wait for some
847 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700848 *
849 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Jens Axboe165125e2007-07-24 09:28:11 +0200851static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200852 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200854 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 struct request *rq;
856
Jens Axboe7749a8d2006-12-13 13:02:26 +0100857 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700858 while (!rq) {
859 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200860 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 struct request_list *rl = &q->rq;
862
Jens Axboe1faa16d2009-04-06 14:48:01 +0200863 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 TASK_UNINTERRUPTIBLE);
865
Jens Axboe1faa16d2009-04-06 14:48:01 +0200866 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200868 __generic_unplug_device(q);
869 spin_unlock_irq(q->queue_lock);
870 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200872 /*
873 * After sleeping, we become a "batching" process and
874 * will be able to allocate at least one request, and
875 * up to a big batch of them for a small period time.
876 * See ioc_batching, ioc_set_batching
877 */
878 ioc = current_io_context(GFP_NOIO, q->node);
879 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100880
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200881 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200882 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200883
884 rq = get_request(q, rw_flags, bio, GFP_NOIO);
885 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 return rq;
888}
889
Jens Axboe165125e2007-07-24 09:28:11 +0200890struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
892 struct request *rq;
893
894 BUG_ON(rw != READ && rw != WRITE);
895
Nick Piggind6344532005-06-28 20:45:14 -0700896 spin_lock_irq(q->queue_lock);
897 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200898 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700899 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200900 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700901 if (!rq)
902 spin_unlock_irq(q->queue_lock);
903 }
904 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 return rq;
907}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908EXPORT_SYMBOL(blk_get_request);
909
910/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300911 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700912 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300913 * @bio: The bio describing the memory mappings that will be submitted for IO.
914 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700915 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200916 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300917 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
918 * type commands. Where the struct request needs to be farther initialized by
919 * the caller. It is passed a &struct bio, which describes the memory info of
920 * the I/O transfer.
921 *
922 * The caller of blk_make_request must make sure that bi_io_vec
923 * are set to describe the memory buffers. That bio_data_dir() will return
924 * the needed direction of the request. (And all bio's in the passed bio-chain
925 * are properly set accordingly)
926 *
927 * If called under none-sleepable conditions, mapped bio buffers must not
928 * need bouncing, by calling the appropriate masked or flagged allocator,
929 * suitable for the target device. Otherwise the call to blk_queue_bounce will
930 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200931 *
932 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
933 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
934 * anything but the first bio in the chain. Otherwise you risk waiting for IO
935 * completion of a bio that hasn't been submitted yet, thus resulting in a
936 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
937 * of bio_alloc(), as that avoids the mempool deadlock.
938 * If possible a big IO should be split into smaller parts when allocation
939 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200940 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300941struct request *blk_make_request(struct request_queue *q, struct bio *bio,
942 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200943{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300944 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
945
946 if (unlikely(!rq))
947 return ERR_PTR(-ENOMEM);
948
949 for_each_bio(bio) {
950 struct bio *bounce_bio = bio;
951 int ret;
952
953 blk_queue_bounce(q, &bounce_bio);
954 ret = blk_rq_append_bio(q, rq, bounce_bio);
955 if (unlikely(ret)) {
956 blk_put_request(rq);
957 return ERR_PTR(ret);
958 }
959 }
960
961 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200962}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300963EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200964
965/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 * blk_requeue_request - put a request back on queue
967 * @q: request queue where request should be inserted
968 * @rq: request to be inserted
969 *
970 * Description:
971 * Drivers often keep queueing requests until the hardware cannot accept
972 * more, when that condition happens we need to put the request back
973 * on the queue. Must be called with queue lock held.
974 */
Jens Axboe165125e2007-07-24 09:28:11 +0200975void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700977 blk_delete_timer(rq);
978 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100979 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (blk_rq_tagged(rq))
982 blk_queue_end_tag(q, rq);
983
James Bottomleyba396a62009-05-27 14:17:08 +0200984 BUG_ON(blk_queued_rq(rq));
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 elv_requeue_request(q, rq);
987}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988EXPORT_SYMBOL(blk_requeue_request);
989
990/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200991 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 * @q: request queue where request should be inserted
993 * @rq: request to be inserted
994 * @at_head: insert request at head or tail of queue
995 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 *
997 * Description:
998 * Many block devices need to execute commands asynchronously, so they don't
999 * block the whole kernel from preemption during request execution. This is
1000 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001001 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1002 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 *
1004 * We have the option of inserting the head or the tail of the queue.
1005 * Typically we use the tail for new ioctls and so forth. We use the head
1006 * of the queue for things like a QUEUE_FULL message from a device, or a
1007 * host that is unable to accept a particular command.
1008 */
Jens Axboe165125e2007-07-24 09:28:11 +02001009void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001010 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
Tejun Heo 867d1192005-04-24 02:06:05 -05001012 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 unsigned long flags;
1014
1015 /*
1016 * tell I/O scheduler that this isn't a regular read/write (ie it
1017 * must not attempt merges on this) and that it acts as a soft
1018 * barrier
1019 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001020 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 rq->special = data;
1023
1024 spin_lock_irqsave(q->queue_lock, flags);
1025
1026 /*
1027 * If command is tagged, release the tag
1028 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001029 if (blk_rq_tagged(rq))
1030 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001032 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05001033 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +09001034 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 spin_unlock_irqrestore(q->queue_lock, flags);
1036}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037EXPORT_SYMBOL(blk_insert_request);
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039/*
1040 * add-request adds a request to the linked list.
1041 * queue lock is held and interrupts disabled, as we muck with the
1042 * request queue list.
1043 */
Jens Axboe6728cb02008-01-31 13:03:55 +01001044static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001046 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /*
1049 * elevator indicated where it wants this request to be
1050 * inserted at elevator_merge time
1051 */
1052 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1053}
Jens Axboe6728cb02008-01-31 13:03:55 +01001054
Tejun Heo074a7ac2008-08-25 19:56:14 +09001055static void part_round_stats_single(int cpu, struct hd_struct *part,
1056 unsigned long now)
1057{
1058 if (now == part->stamp)
1059 return;
1060
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001061 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001062 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001063 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001064 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1065 }
1066 part->stamp = now;
1067}
1068
1069/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001070 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1071 * @cpu: cpu number for stats access
1072 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 *
1074 * The average IO queue length and utilisation statistics are maintained
1075 * by observing the current state of the queue length and the amount of
1076 * time it has been in this state for.
1077 *
1078 * Normally, that accounting is done on IO completion, but that can result
1079 * in more than a second's worth of IO being accounted for within any one
1080 * second, leading to >100% utilisation. To deal with that, we call this
1081 * function to do a round-off before returning the results when reading
1082 * /proc/diskstats. This accounts immediately for all queue usage up to
1083 * the current jiffies and restarts the counters again.
1084 */
Tejun Heoc9959052008-08-25 19:47:21 +09001085void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001086{
1087 unsigned long now = jiffies;
1088
Tejun Heo074a7ac2008-08-25 19:56:14 +09001089 if (part->partno)
1090 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1091 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001092}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001093EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095/*
1096 * queue lock must be held
1097 */
Jens Axboe165125e2007-07-24 09:28:11 +02001098void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (unlikely(!q))
1101 return;
1102 if (unlikely(--req->ref_count))
1103 return;
1104
Tejun Heo8922e162005-10-20 16:23:44 +02001105 elv_completed_request(q, req);
1106
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001107 /* this is a bio leak */
1108 WARN_ON(req->bio != NULL);
1109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 /*
1111 * Request may not have originated from ll_rw_blk. if not,
1112 * it didn't come out of our reserved rq pools
1113 */
Jens Axboe49171e52006-08-10 08:59:11 +02001114 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001115 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001116 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001119 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001122 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
1124}
Mike Christie6e39b692005-11-11 05:30:24 -06001125EXPORT_SYMBOL_GPL(__blk_put_request);
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127void blk_put_request(struct request *req)
1128{
Tejun Heo8922e162005-10-20 16:23:44 +02001129 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001130 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001132 spin_lock_irqsave(q->queue_lock, flags);
1133 __blk_put_request(q, req);
1134 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136EXPORT_SYMBOL(blk_put_request);
1137
Jens Axboe86db1e22008-01-29 14:53:40 +01001138void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001139{
Jens Axboec7c22e42008-09-13 20:26:01 +02001140 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001141 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001142
1143 /*
Tejun Heoa82afdf2009-07-03 17:48:16 +09001144 * Inherit FAILFAST from bio (for read-ahead, and explicit
1145 * FAILFAST). FAILFAST flags are identical for req and bio.
Tejun Heo52d9e672006-01-06 09:49:58 +01001146 */
Jens Axboe1f98a132009-09-11 14:32:04 +02001147 if (bio_rw_flagged(bio, BIO_RW_AHEAD))
Tejun Heoa82afdf2009-07-03 17:48:16 +09001148 req->cmd_flags |= REQ_FAILFAST_MASK;
1149 else
1150 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
Tejun Heo52d9e672006-01-06 09:49:58 +01001151
Jens Axboe1f98a132009-09-11 14:32:04 +02001152 if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001153 req->cmd_flags |= REQ_DISCARD;
Jens Axboe1f98a132009-09-11 14:32:04 +02001154 if (bio_rw_flagged(bio, BIO_RW_BARRIER))
David Woodhousee17fc0a2008-08-09 16:42:20 +01001155 req->cmd_flags |= REQ_SOFTBARRIER;
Jens Axboe1f98a132009-09-11 14:32:04 +02001156 } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
Tejun Heoe4025f62009-04-23 11:05:17 +09001157 req->cmd_flags |= REQ_HARDBARRIER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001158
Jens Axboe1f98a132009-09-11 14:32:04 +02001159 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001160 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe1f98a132009-09-11 14:32:04 +02001161 if (bio_rw_flagged(bio, BIO_RW_META))
Jens Axboe5404bc72006-08-10 09:01:02 +02001162 req->cmd_flags |= REQ_RW_META;
Jens Axboe1f98a132009-09-11 14:32:04 +02001163 if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
Jens Axboeaeb6faf2009-04-06 14:48:07 +02001164 req->cmd_flags |= REQ_NOIDLE;
Jens Axboeb31dc662006-06-13 08:26:10 +02001165
Tejun Heo52d9e672006-01-06 09:49:58 +01001166 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001167 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001168 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001169 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001170}
1171
Jens Axboe644b2d92009-04-06 14:48:06 +02001172/*
1173 * Only disabling plugging for non-rotational devices if it does tagging
1174 * as well, otherwise we do need the proper merging
1175 */
1176static inline bool queue_should_plug(struct request_queue *q)
1177{
Jens Axboe79da0642010-02-23 08:40:43 +01001178 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
Jens Axboe644b2d92009-04-06 14:48:06 +02001179}
1180
Jens Axboe165125e2007-07-24 09:28:11 +02001181static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
Nick Piggin450991b2005-06-28 20:45:13 -07001183 struct request *req;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001184 int el_ret;
1185 unsigned int bytes = bio->bi_size;
Jens Axboe51da90f2006-07-18 04:14:45 +02001186 const unsigned short prio = bio_prio(bio);
Jens Axboe1f98a132009-09-11 14:32:04 +02001187 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1188 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
Tejun Heo80a761f2009-07-03 17:48:17 +09001189 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001190 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Mark McLoughlin6cafb122009-10-24 14:14:31 +02001192 if (bio_rw_flagged(bio, BIO_RW_BARRIER) &&
NeilBrowndb64f682009-06-30 09:35:44 +02001193 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1194 bio_endio(bio, -EOPNOTSUPP);
1195 return 0;
1196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 /*
1198 * low level driver can indicate that it wants pages above a
1199 * certain limit bounced to low memory (ie for highmem, or even
1200 * ISA dma in theory)
1201 */
1202 blk_queue_bounce(q, &bio);
1203
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 spin_lock_irq(q->queue_lock);
1205
Jens Axboe1f98a132009-09-11 14:32:04 +02001206 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 goto get_rq;
1208
1209 el_ret = elv_merge(q, &req, bio);
1210 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001211 case ELEVATOR_BACK_MERGE:
1212 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Jens Axboe6728cb02008-01-31 13:03:55 +01001214 if (!ll_back_merge_fn(q, req, bio))
1215 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001217 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001218
Tejun Heo80a761f2009-07-03 17:48:17 +09001219 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1220 blk_rq_set_mixed_merge(req);
1221
Jens Axboe6728cb02008-01-31 13:03:55 +01001222 req->biotail->bi_next = bio;
1223 req->biotail = bio;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001224 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001225 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001226 if (!blk_rq_cpu_valid(req))
1227 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001228 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001229 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001230 if (!attempt_back_merge(q, req))
1231 elv_merged_request(q, req, el_ret);
1232 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Jens Axboe6728cb02008-01-31 13:03:55 +01001234 case ELEVATOR_FRONT_MERGE:
1235 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Jens Axboe6728cb02008-01-31 13:03:55 +01001237 if (!ll_front_merge_fn(q, req, bio))
1238 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001240 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001241
Tejun Heo80a761f2009-07-03 17:48:17 +09001242 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1243 blk_rq_set_mixed_merge(req);
1244 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1245 req->cmd_flags |= ff;
1246 }
1247
Jens Axboe6728cb02008-01-31 13:03:55 +01001248 bio->bi_next = req->bio;
1249 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Jens Axboe6728cb02008-01-31 13:03:55 +01001251 /*
1252 * may not be valid. if the low level driver said
1253 * it didn't need a bounce buffer then it better
1254 * not touch req->buffer either...
1255 */
1256 req->buffer = bio_data(bio);
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001257 req->__sector = bio->bi_sector;
1258 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001259 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001260 if (!blk_rq_cpu_valid(req))
1261 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001262 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001263 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001264 if (!attempt_front_merge(q, req))
1265 elv_merged_request(q, req, el_ret);
1266 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Jens Axboe6728cb02008-01-31 13:03:55 +01001268 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1269 default:
1270 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001274 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001275 * This sync check and mask will be re-done in init_request_from_bio(),
1276 * but we need to set it earlier to expose the sync flag to the
1277 * rq allocator and io schedulers.
1278 */
1279 rw_flags = bio_data_dir(bio);
1280 if (sync)
1281 rw_flags |= REQ_RW_SYNC;
1282
1283 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001284 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001285 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001286 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001287 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001288
Nick Piggin450991b2005-06-28 20:45:13 -07001289 /*
1290 * After dropping the lock and possibly sleeping here, our request
1291 * may now be mergeable after it had proven unmergeable (above).
1292 * We don't worry about that case for efficiency. It won't happen
1293 * often, and the elevators are able to handle it.
1294 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001295 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Nick Piggin450991b2005-06-28 20:45:13 -07001297 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001298 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1299 bio_flagged(bio, BIO_CPU_AFFINE))
1300 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001301 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001302 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 add_request(q, req);
1304out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001305 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 spin_unlock_irq(q->queue_lock);
1308 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309}
1310
1311/*
1312 * If bio->bi_dev is a partition, remap the location
1313 */
1314static inline void blk_partition_remap(struct bio *bio)
1315{
1316 struct block_device *bdev = bio->bi_bdev;
1317
Jens Axboebf2de6f2007-09-27 13:01:25 +02001318 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 bio->bi_sector += p->start_sect;
1322 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001323
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001324 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001325 bdev->bd_dev,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001326 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 }
1328}
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330static void handle_bad_sector(struct bio *bio)
1331{
1332 char b[BDEVNAME_SIZE];
1333
1334 printk(KERN_INFO "attempt to access beyond end of device\n");
1335 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1336 bdevname(bio->bi_bdev, b),
1337 bio->bi_rw,
1338 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1339 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1340
1341 set_bit(BIO_EOF, &bio->bi_flags);
1342}
1343
Akinobu Mitac17bb492006-12-08 02:39:46 -08001344#ifdef CONFIG_FAIL_MAKE_REQUEST
1345
1346static DECLARE_FAULT_ATTR(fail_make_request);
1347
1348static int __init setup_fail_make_request(char *str)
1349{
1350 return setup_fault_attr(&fail_make_request, str);
1351}
1352__setup("fail_make_request=", setup_fail_make_request);
1353
1354static int should_fail_request(struct bio *bio)
1355{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001356 struct hd_struct *part = bio->bi_bdev->bd_part;
1357
1358 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001359 return should_fail(&fail_make_request, bio->bi_size);
1360
1361 return 0;
1362}
1363
1364static int __init fail_make_request_debugfs(void)
1365{
1366 return init_fault_attr_dentries(&fail_make_request,
1367 "fail_make_request");
1368}
1369
1370late_initcall(fail_make_request_debugfs);
1371
1372#else /* CONFIG_FAIL_MAKE_REQUEST */
1373
1374static inline int should_fail_request(struct bio *bio)
1375{
1376 return 0;
1377}
1378
1379#endif /* CONFIG_FAIL_MAKE_REQUEST */
1380
Jens Axboec07e2b42007-07-18 13:27:58 +02001381/*
1382 * Check whether this bio extends beyond the end of the device.
1383 */
1384static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1385{
1386 sector_t maxsector;
1387
1388 if (!nr_sectors)
1389 return 0;
1390
1391 /* Test device or partition size, when known. */
1392 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1393 if (maxsector) {
1394 sector_t sector = bio->bi_sector;
1395
1396 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1397 /*
1398 * This may well happen - the kernel calls bread()
1399 * without checking the size of the device, e.g., when
1400 * mounting a device.
1401 */
1402 handle_bad_sector(bio);
1403 return 1;
1404 }
1405 }
1406
1407 return 0;
1408}
1409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001411 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 * @bio: The bio describing the location in memory and on the device.
1413 *
1414 * generic_make_request() is used to make I/O requests of block
1415 * devices. It is passed a &struct bio, which describes the I/O that needs
1416 * to be done.
1417 *
1418 * generic_make_request() does not return any status. The
1419 * success/failure status of the request, along with notification of
1420 * completion, is delivered asynchronously through the bio->bi_end_io
1421 * function described (one day) else where.
1422 *
1423 * The caller of generic_make_request must make sure that bi_io_vec
1424 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1425 * set to describe the device address, and the
1426 * bi_end_io and optionally bi_private are set to describe how
1427 * completion notification should be signaled.
1428 *
1429 * generic_make_request and the drivers it calls may use bi_next if this
1430 * bio happens to be merged with someone else, and may change bi_dev and
1431 * bi_sector for remaps as it sees fit. So the values of these fields
1432 * should NOT be depended on after the call to generic_make_request.
1433 */
Neil Brownd89d8792007-05-01 09:53:42 +02001434static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Jens Axboe165125e2007-07-24 09:28:11 +02001436 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001437 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001439 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001440 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Jens Axboec07e2b42007-07-18 13:27:58 +02001444 if (bio_check_eod(bio, nr_sectors))
1445 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
1447 /*
1448 * Resolve the mapping until finished. (drivers are
1449 * still free to implement/resolve their own stacking
1450 * by explicitly returning 0)
1451 *
1452 * NOTE: we don't repeat the blk_size check for each new device.
1453 * Stacking drivers are expected to know what they are doing.
1454 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001455 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001456 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 do {
1458 char b[BDEVNAME_SIZE];
1459
1460 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001461 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 printk(KERN_ERR
1463 "generic_make_request: Trying to access "
1464 "nonexistent block-device %s (%Lu)\n",
1465 bdevname(bio->bi_bdev, b),
1466 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001467 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 }
1469
Christoph Hellwig67efc922009-09-30 13:54:20 +02001470 if (unlikely(!bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1471 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001472 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001473 bdevname(bio->bi_bdev, b),
1474 bio_sectors(bio),
1475 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 goto end_io;
1477 }
1478
Nick Pigginfde6ad22005-06-23 00:08:53 -07001479 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 goto end_io;
1481
Akinobu Mitac17bb492006-12-08 02:39:46 -08001482 if (should_fail_request(bio))
1483 goto end_io;
1484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 /*
1486 * If this device has partitions, remap block n
1487 * of partition p to block n+start(p) of the disk.
1488 */
1489 blk_partition_remap(bio);
1490
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001491 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1492 goto end_io;
1493
NeilBrown5ddfe962006-10-30 22:07:21 -08001494 if (old_sector != -1)
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001495 trace_block_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001496
NeilBrown5ddfe962006-10-30 22:07:21 -08001497 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001498 old_dev = bio->bi_bdev->bd_dev;
1499
Jens Axboec07e2b42007-07-18 13:27:58 +02001500 if (bio_check_eod(bio, nr_sectors))
1501 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001502
Jens Axboe1f98a132009-09-11 14:32:04 +02001503 if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
Christoph Hellwigc15227d2009-09-30 13:52:12 +02001504 !blk_queue_discard(q)) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001505 err = -EOPNOTSUPP;
1506 goto end_io;
1507 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001508
Minchan Kim01edede2009-09-08 21:56:38 +02001509 trace_block_bio_queue(q, bio);
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 ret = q->make_request_fn(q, bio);
1512 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001513
1514 return;
1515
1516end_io:
1517 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518}
1519
Neil Brownd89d8792007-05-01 09:53:42 +02001520/*
1521 * We only want one ->make_request_fn to be active at a time,
1522 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001523 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001524 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001525 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001526 * generic_make_request is currently active in this task or not.
1527 * If it is NULL, then no make_request is active. If it is non-NULL,
1528 * then a make_request is active, and new requests should be added
1529 * at the tail
1530 */
1531void generic_make_request(struct bio *bio)
1532{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001533 struct bio_list bio_list_on_stack;
1534
1535 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001536 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001537 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001538 return;
1539 }
1540 /* following loop may be a bit non-obvious, and so deserves some
1541 * explanation.
1542 * Before entering the loop, bio->bi_next is NULL (as all callers
1543 * ensure that) so we have a list with a single bio.
1544 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001545 * we assign bio_list to a pointer to the bio_list_on_stack,
1546 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001547 * added. __generic_make_request may indeed add some more bios
1548 * through a recursive call to generic_make_request. If it
1549 * did, we find a non-NULL value in bio_list and re-enter the loop
1550 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001551 * of the top of the list (no pretending) and so remove it from
1552 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001553 *
1554 * The loop was structured like this to make only one call to
1555 * __generic_make_request (which is important as it is large and
1556 * inlined) and to keep the structure simple.
1557 */
1558 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001559 bio_list_init(&bio_list_on_stack);
1560 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001561 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001562 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001563 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001564 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001565 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001566}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567EXPORT_SYMBOL(generic_make_request);
1568
1569/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001570 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1572 * @bio: The &struct bio which describes the I/O
1573 *
1574 * submit_bio() is very similar in purpose to generic_make_request(), and
1575 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001576 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 *
1578 */
1579void submit_bio(int rw, struct bio *bio)
1580{
1581 int count = bio_sectors(bio);
1582
Jens Axboe22e2c502005-06-27 10:55:12 +02001583 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
Jens Axboebf2de6f2007-09-27 13:01:25 +02001585 /*
1586 * If it's a regular read/write or a barrier with data attached,
1587 * go through the normal accounting stuff before submission.
1588 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001589 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001590 if (rw & WRITE) {
1591 count_vm_events(PGPGOUT, count);
1592 } else {
1593 task_io_account_read(bio->bi_size);
1594 count_vm_events(PGPGIN, count);
1595 }
1596
1597 if (unlikely(block_dump)) {
1598 char b[BDEVNAME_SIZE];
1599 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001600 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001601 (rw & WRITE) ? "WRITE" : "READ",
1602 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001603 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 }
1606
1607 generic_make_request(bio);
1608}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609EXPORT_SYMBOL(submit_bio);
1610
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001611/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001612 * blk_rq_check_limits - Helper function to check a request for the queue limit
1613 * @q: the queue
1614 * @rq: the request being checked
1615 *
1616 * Description:
1617 * @rq may have been made based on weaker limitations of upper-level queues
1618 * in request stacking drivers, and it may violate the limitation of @q.
1619 * Since the block layer and the underlying device driver trust @rq
1620 * after it is inserted to @q, it should be checked against @q before
1621 * the insertion using this generic function.
1622 *
1623 * This function should also be useful for request stacking drivers
1624 * in some cases below, so export this fuction.
1625 * Request stacking drivers like request-based dm may change the queue
1626 * limits while requests are in the queue (e.g. dm's table swapping).
1627 * Such request stacking drivers should check those requests agaist
1628 * the new queue limits again when they dispatch those requests,
1629 * although such checkings are also done against the old queue limits
1630 * when submitting requests.
1631 */
1632int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1633{
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001634 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1635 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001636 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1637 return -EIO;
1638 }
1639
1640 /*
1641 * queue's settings related to segment counting like q->bounce_pfn
1642 * may differ from that of other stacking queues.
1643 * Recalculate it to check the request correctly on this queue's
1644 * limitation.
1645 */
1646 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001647 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001648 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1649 return -EIO;
1650 }
1651
1652 return 0;
1653}
1654EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1655
1656/**
1657 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1658 * @q: the queue to submit the request
1659 * @rq: the request being queued
1660 */
1661int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1662{
1663 unsigned long flags;
1664
1665 if (blk_rq_check_limits(q, rq))
1666 return -EIO;
1667
1668#ifdef CONFIG_FAIL_MAKE_REQUEST
1669 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1670 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1671 return -EIO;
1672#endif
1673
1674 spin_lock_irqsave(q->queue_lock, flags);
1675
1676 /*
1677 * Submitting request must be dequeued before calling this function
1678 * because it will be linked to another request_queue
1679 */
1680 BUG_ON(blk_queued_rq(rq));
1681
1682 drive_stat_acct(rq, 1);
1683 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1684
1685 spin_unlock_irqrestore(q->queue_lock, flags);
1686
1687 return 0;
1688}
1689EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1690
Tejun Heo80a761f2009-07-03 17:48:17 +09001691/**
1692 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1693 * @rq: request to examine
1694 *
1695 * Description:
1696 * A request could be merge of IOs which require different failure
1697 * handling. This function determines the number of bytes which
1698 * can be failed from the beginning of the request without
1699 * crossing into area which need to be retried further.
1700 *
1701 * Return:
1702 * The number of bytes to fail.
1703 *
1704 * Context:
1705 * queue_lock must be held.
1706 */
1707unsigned int blk_rq_err_bytes(const struct request *rq)
1708{
1709 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1710 unsigned int bytes = 0;
1711 struct bio *bio;
1712
1713 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1714 return blk_rq_bytes(rq);
1715
1716 /*
1717 * Currently the only 'mixing' which can happen is between
1718 * different fastfail types. We can safely fail portions
1719 * which have all the failfast bits that the first one has -
1720 * the ones which are at least as eager to fail as the first
1721 * one.
1722 */
1723 for (bio = rq->bio; bio; bio = bio->bi_next) {
1724 if ((bio->bi_rw & ff) != ff)
1725 break;
1726 bytes += bio->bi_size;
1727 }
1728
1729 /* this could lead to infinite loop */
1730 BUG_ON(blk_rq_bytes(rq) && !bytes);
1731 return bytes;
1732}
1733EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1734
Jens Axboebc58ba92009-01-23 10:54:44 +01001735static void blk_account_io_completion(struct request *req, unsigned int bytes)
1736{
Jens Axboec2553b52009-04-24 08:10:11 +02001737 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001738 const int rw = rq_data_dir(req);
1739 struct hd_struct *part;
1740 int cpu;
1741
1742 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001743 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001744 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1745 part_stat_unlock();
1746 }
1747}
1748
1749static void blk_account_io_done(struct request *req)
1750{
Jens Axboebc58ba92009-01-23 10:54:44 +01001751 /*
1752 * Account IO completion. bar_rq isn't accounted as a normal
1753 * IO on queueing nor completion. Accounting the containing
1754 * request is enough.
1755 */
Jens Axboec2553b52009-04-24 08:10:11 +02001756 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001757 unsigned long duration = jiffies - req->start_time;
1758 const int rw = rq_data_dir(req);
1759 struct hd_struct *part;
1760 int cpu;
1761
1762 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001763 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001764
1765 part_stat_inc(cpu, part, ios[rw]);
1766 part_stat_add(cpu, part, ticks[rw], duration);
1767 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001768 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001769
1770 part_stat_unlock();
1771 }
1772}
1773
Tejun Heo53a08802008-12-03 12:41:26 +01001774/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001775 * blk_peek_request - peek at the top of a request queue
1776 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001777 *
1778 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001779 * Return the request at the top of @q. The returned request
1780 * should be started using blk_start_request() before LLD starts
1781 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001782 *
1783 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001784 * Pointer to the request at the top of @q if available. Null
1785 * otherwise.
1786 *
1787 * Context:
1788 * queue_lock must be held.
1789 */
1790struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001791{
1792 struct request *rq;
1793 int ret;
1794
1795 while ((rq = __elv_next_request(q)) != NULL) {
1796 if (!(rq->cmd_flags & REQ_STARTED)) {
1797 /*
1798 * This is the first time the device driver
1799 * sees this request (possibly after
1800 * requeueing). Notify IO scheduler.
1801 */
1802 if (blk_sorted_rq(rq))
1803 elv_activate_rq(q, rq);
1804
1805 /*
1806 * just mark as started even if we don't start
1807 * it, a request that has been delayed should
1808 * not be passed by new incoming requests
1809 */
1810 rq->cmd_flags |= REQ_STARTED;
1811 trace_block_rq_issue(q, rq);
1812 }
1813
1814 if (!q->boundary_rq || q->boundary_rq == rq) {
1815 q->end_sector = rq_end_sector(rq);
1816 q->boundary_rq = NULL;
1817 }
1818
1819 if (rq->cmd_flags & REQ_DONTPREP)
1820 break;
1821
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001822 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001823 /*
1824 * make sure space for the drain appears we
1825 * know we can do this because max_hw_segments
1826 * has been adjusted to be one fewer than the
1827 * device can handle
1828 */
1829 rq->nr_phys_segments++;
1830 }
1831
1832 if (!q->prep_rq_fn)
1833 break;
1834
1835 ret = q->prep_rq_fn(q, rq);
1836 if (ret == BLKPREP_OK) {
1837 break;
1838 } else if (ret == BLKPREP_DEFER) {
1839 /*
1840 * the request may have been (partially) prepped.
1841 * we need to keep this request in the front to
1842 * avoid resource deadlock. REQ_STARTED will
1843 * prevent other fs requests from passing this one.
1844 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001845 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001846 !(rq->cmd_flags & REQ_DONTPREP)) {
1847 /*
1848 * remove the space for the drain we added
1849 * so that we don't add it again
1850 */
1851 --rq->nr_phys_segments;
1852 }
1853
1854 rq = NULL;
1855 break;
1856 } else if (ret == BLKPREP_KILL) {
1857 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001858 /*
1859 * Mark this request as started so we don't trigger
1860 * any debug logic in the end I/O path.
1861 */
1862 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001863 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001864 } else {
1865 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1866 break;
1867 }
1868 }
1869
1870 return rq;
1871}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001872EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001873
Tejun Heo9934c8c2009-05-08 11:54:16 +09001874void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001875{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001876 struct request_queue *q = rq->q;
1877
Tejun Heo158dbda2009-04-23 11:05:18 +09001878 BUG_ON(list_empty(&rq->queuelist));
1879 BUG_ON(ELV_ON_HASH(rq));
1880
1881 list_del_init(&rq->queuelist);
1882
1883 /*
1884 * the time frame between a request being removed from the lists
1885 * and to it is freed is accounted as io that is in progress at
1886 * the driver side.
1887 */
Divyesh Shah91952912010-04-01 15:01:41 -07001888 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001889 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001890 set_io_start_time_ns(rq);
1891 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001892}
1893
Tejun Heo5efccd12009-04-23 11:05:18 +09001894/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001895 * blk_start_request - start request processing on the driver
1896 * @req: request to dequeue
1897 *
1898 * Description:
1899 * Dequeue @req and start timeout timer on it. This hands off the
1900 * request to the driver.
1901 *
1902 * Block internal functions which don't want to start timer should
1903 * call blk_dequeue_request().
1904 *
1905 * Context:
1906 * queue_lock must be held.
1907 */
1908void blk_start_request(struct request *req)
1909{
1910 blk_dequeue_request(req);
1911
1912 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001913 * We are now handing the request to the hardware, initialize
1914 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001915 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001916 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001917 if (unlikely(blk_bidi_rq(req)))
1918 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1919
Tejun Heo9934c8c2009-05-08 11:54:16 +09001920 blk_add_timer(req);
1921}
1922EXPORT_SYMBOL(blk_start_request);
1923
1924/**
1925 * blk_fetch_request - fetch a request from a request queue
1926 * @q: request queue to fetch a request from
1927 *
1928 * Description:
1929 * Return the request at the top of @q. The request is started on
1930 * return and LLD can start processing it immediately.
1931 *
1932 * Return:
1933 * Pointer to the request at the top of @q if available. Null
1934 * otherwise.
1935 *
1936 * Context:
1937 * queue_lock must be held.
1938 */
1939struct request *blk_fetch_request(struct request_queue *q)
1940{
1941 struct request *rq;
1942
1943 rq = blk_peek_request(q);
1944 if (rq)
1945 blk_start_request(rq);
1946 return rq;
1947}
1948EXPORT_SYMBOL(blk_fetch_request);
1949
1950/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001951 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001952 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001953 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001954 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001955 *
1956 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001957 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1958 * the request structure even if @req doesn't have leftover.
1959 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001960 *
1961 * This special helper function is only for request stacking drivers
1962 * (e.g. request-based dm) so that they can handle partial completion.
1963 * Actual device drivers should use blk_end_request instead.
1964 *
1965 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1966 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001967 *
1968 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09001969 * %false - this request doesn't have any more data
1970 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001971 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09001972bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001974 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 struct bio *bio;
1976
Tejun Heo2e60e022009-04-23 11:05:18 +09001977 if (!req->bio)
1978 return false;
1979
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001980 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001981
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 /*
Tejun Heo6f414692009-04-19 07:00:41 +09001983 * For fs requests, rq is just carrier of independent bio's
1984 * and each partial completion should be handled separately.
1985 * Reset per-request error on each partial completion.
1986 *
1987 * TODO: tj: This is too subtle. It would be better to let
1988 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 */
Tejun Heo6f414692009-04-19 07:00:41 +09001990 if (blk_fs_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 req->errors = 0;
1992
Jens Axboe6728cb02008-01-31 13:03:55 +01001993 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1994 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09001996 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 }
1998
Jens Axboebc58ba92009-01-23 10:54:44 +01001999 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002000
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 total_bytes = bio_nbytes = 0;
2002 while ((bio = req->bio) != NULL) {
2003 int nbytes;
2004
2005 if (nr_bytes >= bio->bi_size) {
2006 req->bio = bio->bi_next;
2007 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002008 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 next_idx = 0;
2010 bio_nbytes = 0;
2011 } else {
2012 int idx = bio->bi_idx + next_idx;
2013
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002014 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002016 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002017 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 break;
2019 }
2020
2021 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2022 BIO_BUG_ON(nbytes > bio->bi_size);
2023
2024 /*
2025 * not a complete bvec done
2026 */
2027 if (unlikely(nbytes > nr_bytes)) {
2028 bio_nbytes += nr_bytes;
2029 total_bytes += nr_bytes;
2030 break;
2031 }
2032
2033 /*
2034 * advance to the next vector
2035 */
2036 next_idx++;
2037 bio_nbytes += nbytes;
2038 }
2039
2040 total_bytes += nbytes;
2041 nr_bytes -= nbytes;
2042
Jens Axboe6728cb02008-01-31 13:03:55 +01002043 bio = req->bio;
2044 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /*
2046 * end more in this run, or just return 'not-done'
2047 */
2048 if (unlikely(nr_bytes <= 0))
2049 break;
2050 }
2051 }
2052
2053 /*
2054 * completely done
2055 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002056 if (!req->bio) {
2057 /*
2058 * Reset counters so that the request stacking driver
2059 * can find how many bytes remain in the request
2060 * later.
2061 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002062 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002063 return false;
2064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066 /*
2067 * if the request wasn't completed, update state
2068 */
2069 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002070 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 bio->bi_idx += next_idx;
2072 bio_iovec(bio)->bv_offset += nr_bytes;
2073 bio_iovec(bio)->bv_len -= nr_bytes;
2074 }
2075
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002076 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002077 req->buffer = bio_data(req->bio);
2078
2079 /* update sector only for requests with clear definition of sector */
2080 if (blk_fs_request(req) || blk_discard_rq(req))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002081 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002082
Tejun Heo80a761f2009-07-03 17:48:17 +09002083 /* mixed attributes always follow the first bio */
2084 if (req->cmd_flags & REQ_MIXED_MERGE) {
2085 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2086 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2087 }
2088
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002089 /*
2090 * If total number of sectors is less than the first segment
2091 * size, something has gone terribly wrong.
2092 */
2093 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2094 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002095 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002096 }
2097
2098 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002100
Tejun Heo2e60e022009-04-23 11:05:18 +09002101 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102}
Tejun Heo2e60e022009-04-23 11:05:18 +09002103EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Tejun Heo2e60e022009-04-23 11:05:18 +09002105static bool blk_update_bidi_request(struct request *rq, int error,
2106 unsigned int nr_bytes,
2107 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002108{
Tejun Heo2e60e022009-04-23 11:05:18 +09002109 if (blk_update_request(rq, error, nr_bytes))
2110 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002111
Tejun Heo2e60e022009-04-23 11:05:18 +09002112 /* Bidi request must be completed as a whole */
2113 if (unlikely(blk_bidi_rq(rq)) &&
2114 blk_update_request(rq->next_rq, error, bidi_bytes))
2115 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002116
Tejun Heo2e60e022009-04-23 11:05:18 +09002117 add_disk_randomness(rq->rq_disk);
2118
2119 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
2121
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122/*
2123 * queue lock must be held
2124 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002125static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002127 if (blk_rq_tagged(req))
2128 blk_queue_end_tag(req->q, req);
2129
James Bottomleyba396a62009-05-27 14:17:08 +02002130 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132 if (unlikely(laptop_mode) && blk_fs_request(req))
Matthew Garrett31373d02010-04-06 14:25:14 +02002133 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Mike Andersone78042e2008-10-30 02:16:20 -07002135 blk_delete_timer(req);
2136
Jens Axboebc58ba92009-01-23 10:54:44 +01002137 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002140 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002141 else {
2142 if (blk_bidi_rq(req))
2143 __blk_put_request(req->next_rq->q, req->next_rq);
2144
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 }
2147}
2148
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002149/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002150 * blk_end_bidi_request - Complete a bidi request
2151 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002152 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002153 * @nr_bytes: number of bytes to complete @rq
2154 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002155 *
2156 * Description:
2157 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002158 * Drivers that supports bidi can safely call this member for any
2159 * type of request, bidi or uni. In the later case @bidi_bytes is
2160 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002161 *
2162 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002163 * %false - we are done with this request
2164 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002165 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002166static bool blk_end_bidi_request(struct request *rq, int error,
2167 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002168{
2169 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002170 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002171
Tejun Heo2e60e022009-04-23 11:05:18 +09002172 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2173 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002174
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002175 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002176 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002177 spin_unlock_irqrestore(q->queue_lock, flags);
2178
Tejun Heo2e60e022009-04-23 11:05:18 +09002179 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002180}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002181
2182/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002183 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2184 * @rq: the request to complete
2185 * @error: %0 for success, < %0 for error
2186 * @nr_bytes: number of bytes to complete @rq
2187 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002188 *
2189 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002190 * Identical to blk_end_bidi_request() except that queue lock is
2191 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002192 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002193 * Return:
2194 * %false - we are done with this request
2195 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002196 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002197static bool __blk_end_bidi_request(struct request *rq, int error,
2198 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002199{
Tejun Heo2e60e022009-04-23 11:05:18 +09002200 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2201 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002202
Tejun Heo2e60e022009-04-23 11:05:18 +09002203 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002204
Tejun Heo2e60e022009-04-23 11:05:18 +09002205 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002206}
2207
2208/**
2209 * blk_end_request - Helper function for drivers to complete the request.
2210 * @rq: the request being processed
2211 * @error: %0 for success, < %0 for error
2212 * @nr_bytes: number of bytes to complete
2213 *
2214 * Description:
2215 * Ends I/O on a number of bytes attached to @rq.
2216 * If @rq has leftover, sets it up for the next range of segments.
2217 *
2218 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002219 * %false - we are done with this request
2220 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002221 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002222bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002223{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002224 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002225}
Jens Axboe56ad1742009-07-28 22:11:24 +02002226EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002227
2228/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002229 * blk_end_request_all - Helper function for drives to finish the request.
2230 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002231 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002232 *
2233 * Description:
2234 * Completely finish @rq.
2235 */
2236void blk_end_request_all(struct request *rq, int error)
2237{
2238 bool pending;
2239 unsigned int bidi_bytes = 0;
2240
2241 if (unlikely(blk_bidi_rq(rq)))
2242 bidi_bytes = blk_rq_bytes(rq->next_rq);
2243
2244 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2245 BUG_ON(pending);
2246}
Jens Axboe56ad1742009-07-28 22:11:24 +02002247EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002248
2249/**
2250 * blk_end_request_cur - Helper function to finish the current request chunk.
2251 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002252 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002253 *
2254 * Description:
2255 * Complete the current consecutively mapped chunk from @rq.
2256 *
2257 * Return:
2258 * %false - we are done with this request
2259 * %true - still buffers pending for this request
2260 */
2261bool blk_end_request_cur(struct request *rq, int error)
2262{
2263 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2264}
Jens Axboe56ad1742009-07-28 22:11:24 +02002265EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002266
2267/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002268 * blk_end_request_err - Finish a request till the next failure boundary.
2269 * @rq: the request to finish till the next failure boundary for
2270 * @error: must be negative errno
2271 *
2272 * Description:
2273 * Complete @rq till the next failure boundary.
2274 *
2275 * Return:
2276 * %false - we are done with this request
2277 * %true - still buffers pending for this request
2278 */
2279bool blk_end_request_err(struct request *rq, int error)
2280{
2281 WARN_ON(error >= 0);
2282 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2283}
2284EXPORT_SYMBOL_GPL(blk_end_request_err);
2285
2286/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002287 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002288 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002289 * @error: %0 for success, < %0 for error
2290 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002291 *
2292 * Description:
2293 * Must be called with queue lock held unlike blk_end_request().
2294 *
2295 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002296 * %false - we are done with this request
2297 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002298 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002299bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002300{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002301 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002302}
Jens Axboe56ad1742009-07-28 22:11:24 +02002303EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002304
2305/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002306 * __blk_end_request_all - Helper function for drives to finish the request.
2307 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002308 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002309 *
2310 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002311 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002312 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002313void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002314{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002315 bool pending;
2316 unsigned int bidi_bytes = 0;
2317
2318 if (unlikely(blk_bidi_rq(rq)))
2319 bidi_bytes = blk_rq_bytes(rq->next_rq);
2320
2321 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2322 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002323}
Jens Axboe56ad1742009-07-28 22:11:24 +02002324EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002325
2326/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002327 * __blk_end_request_cur - Helper function to finish the current request chunk.
2328 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002329 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002330 *
2331 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002332 * Complete the current consecutively mapped chunk from @rq. Must
2333 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002334 *
2335 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2338 */
2339bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002340{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002341 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002342}
Jens Axboe56ad1742009-07-28 22:11:24 +02002343EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002344
Tejun Heo80a761f2009-07-03 17:48:17 +09002345/**
2346 * __blk_end_request_err - Finish a request till the next failure boundary.
2347 * @rq: the request to finish till the next failure boundary for
2348 * @error: must be negative errno
2349 *
2350 * Description:
2351 * Complete @rq till the next failure boundary. Must be called
2352 * with queue lock held.
2353 *
2354 * Return:
2355 * %false - we are done with this request
2356 * %true - still buffers pending for this request
2357 */
2358bool __blk_end_request_err(struct request *rq, int error)
2359{
2360 WARN_ON(error >= 0);
2361 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2362}
2363EXPORT_SYMBOL_GPL(__blk_end_request_err);
2364
Jens Axboe86db1e22008-01-29 14:53:40 +01002365void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2366 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002368 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2369 rq->cmd_flags |= bio->bi_rw & REQ_RW;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370
David Woodhousefb2dce82008-08-05 18:01:53 +01002371 if (bio_has_data(bio)) {
2372 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002373 rq->buffer = bio_data(bio);
2374 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002375 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
NeilBrown66846572007-08-16 13:31:28 +02002378 if (bio->bi_bdev)
2379 rq->rq_disk = bio->bi_bdev->bd_disk;
2380}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002382#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2383/**
2384 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2385 * @rq: the request to be flushed
2386 *
2387 * Description:
2388 * Flush all pages in @rq.
2389 */
2390void rq_flush_dcache_pages(struct request *rq)
2391{
2392 struct req_iterator iter;
2393 struct bio_vec *bvec;
2394
2395 rq_for_each_segment(bvec, rq, iter)
2396 flush_dcache_page(bvec->bv_page);
2397}
2398EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2399#endif
2400
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002401/**
2402 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2403 * @q : the queue of the device being checked
2404 *
2405 * Description:
2406 * Check if underlying low-level drivers of a device are busy.
2407 * If the drivers want to export their busy state, they must set own
2408 * exporting function using blk_queue_lld_busy() first.
2409 *
2410 * Basically, this function is used only by request stacking drivers
2411 * to stop dispatching requests to underlying devices when underlying
2412 * devices are busy. This behavior helps more I/O merging on the queue
2413 * of the request stacking driver and prevents I/O throughput regression
2414 * on burst I/O load.
2415 *
2416 * Return:
2417 * 0 - Not busy (The request stacking driver should dispatch request)
2418 * 1 - Busy (The request stacking driver should stop dispatching request)
2419 */
2420int blk_lld_busy(struct request_queue *q)
2421{
2422 if (q->lld_busy_fn)
2423 return q->lld_busy_fn(q);
2424
2425 return 0;
2426}
2427EXPORT_SYMBOL_GPL(blk_lld_busy);
2428
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002429/**
2430 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2431 * @rq: the clone request to be cleaned up
2432 *
2433 * Description:
2434 * Free all bios in @rq for a cloned request.
2435 */
2436void blk_rq_unprep_clone(struct request *rq)
2437{
2438 struct bio *bio;
2439
2440 while ((bio = rq->bio) != NULL) {
2441 rq->bio = bio->bi_next;
2442
2443 bio_put(bio);
2444 }
2445}
2446EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2447
2448/*
2449 * Copy attributes of the original request to the clone request.
2450 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2451 */
2452static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2453{
2454 dst->cpu = src->cpu;
2455 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2456 dst->cmd_type = src->cmd_type;
2457 dst->__sector = blk_rq_pos(src);
2458 dst->__data_len = blk_rq_bytes(src);
2459 dst->nr_phys_segments = src->nr_phys_segments;
2460 dst->ioprio = src->ioprio;
2461 dst->extra_len = src->extra_len;
2462}
2463
2464/**
2465 * blk_rq_prep_clone - Helper function to setup clone request
2466 * @rq: the request to be setup
2467 * @rq_src: original request to be cloned
2468 * @bs: bio_set that bios for clone are allocated from
2469 * @gfp_mask: memory allocation mask for bio
2470 * @bio_ctr: setup function to be called for each clone bio.
2471 * Returns %0 for success, non %0 for failure.
2472 * @data: private data to be passed to @bio_ctr
2473 *
2474 * Description:
2475 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2476 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2477 * are not copied, and copying such parts is the caller's responsibility.
2478 * Also, pages which the original bios are pointing to are not copied
2479 * and the cloned bios just point same pages.
2480 * So cloned bios must be completed before original bios, which means
2481 * the caller must complete @rq before @rq_src.
2482 */
2483int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2484 struct bio_set *bs, gfp_t gfp_mask,
2485 int (*bio_ctr)(struct bio *, struct bio *, void *),
2486 void *data)
2487{
2488 struct bio *bio, *bio_src;
2489
2490 if (!bs)
2491 bs = fs_bio_set;
2492
2493 blk_rq_init(NULL, rq);
2494
2495 __rq_for_each_bio(bio_src, rq_src) {
2496 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2497 if (!bio)
2498 goto free_and_out;
2499
2500 __bio_clone(bio, bio_src);
2501
2502 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002503 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002504 goto free_and_out;
2505
2506 if (bio_ctr && bio_ctr(bio, bio_src, data))
2507 goto free_and_out;
2508
2509 if (rq->bio) {
2510 rq->biotail->bi_next = bio;
2511 rq->biotail = bio;
2512 } else
2513 rq->bio = rq->biotail = bio;
2514 }
2515
2516 __blk_rq_prep_clone(rq, rq_src);
2517
2518 return 0;
2519
2520free_and_out:
2521 if (bio)
2522 bio_free(bio, bs);
2523 blk_rq_unprep_clone(rq);
2524
2525 return -ENOMEM;
2526}
2527EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2528
Jens Axboe18887ad2008-07-28 13:08:45 +02002529int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530{
2531 return queue_work(kblockd_workqueue, work);
2532}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533EXPORT_SYMBOL(kblockd_schedule_work);
2534
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535int __init blk_dev_init(void)
2536{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002537 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2538 sizeof(((struct request *)0)->cmd_flags));
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 kblockd_workqueue = create_workqueue("kblockd");
2541 if (!kblockd_workqueue)
2542 panic("Failed to create kblockd\n");
2543
2544 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002545 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Jens Axboe8324aa92008-01-29 14:51:59 +01002547 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002548 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 return 0;
2551}