blob: 5ab3ac22930ce36246d51b11121e8f44e0a70fb3 [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
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200187 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
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;
James Bottomley28018c22010-07-01 19:49:17 +0900611 q->unprep_rq_fn = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100613 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 q->queue_lock = lock;
615
Jens Axboef3b144a2009-03-06 08:48:33 +0100616 /*
617 * This also sets hw/phys segments, boundary and size
618 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Alan Stern44ec9542007-02-20 11:01:57 -0500621 q->sg_reserved_size = INT_MAX;
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 /*
624 * all done
625 */
626 if (!elevator_init(q, NULL)) {
627 blk_queue_congestion_threshold(q);
628 return q;
629 }
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 return NULL;
632}
Mike Snitzer01effb02010-05-11 08:57:42 +0200633EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Jens Axboe165125e2007-07-24 09:28:11 +0200635int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700637 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500638 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 return 0;
640 }
641
642 return 1;
643}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Jens Axboe165125e2007-07-24 09:28:11 +0200645static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200647 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200648 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 mempool_free(rq, q->rq.rq_pool);
650}
651
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200652static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200653blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
656
657 if (!rq)
658 return NULL;
659
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200660 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200661
Jerome Marchand42dad762009-04-22 14:01:49 +0200662 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Tejun Heocb98fc82005-10-28 08:29:39 +0200664 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200665 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200666 mempool_free(rq, q->rq.rq_pool);
667 return NULL;
668 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200669 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Tejun Heocb98fc82005-10-28 08:29:39 +0200672 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
675/*
676 * ioc_batching returns true if the ioc is a valid batching request and
677 * should be given priority access to a request.
678 */
Jens Axboe165125e2007-07-24 09:28:11 +0200679static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 if (!ioc)
682 return 0;
683
684 /*
685 * Make sure the process is able to allocate at least 1 request
686 * even if the batch times out, otherwise we could theoretically
687 * lose wakeups.
688 */
689 return ioc->nr_batch_requests == q->nr_batching ||
690 (ioc->nr_batch_requests > 0
691 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
692}
693
694/*
695 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
696 * will cause the process to be a "batcher" on all queues in the system. This
697 * is the behaviour we want though - once it gets a wakeup it should be given
698 * a nice run.
699 */
Jens Axboe165125e2007-07-24 09:28:11 +0200700static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 if (!ioc || ioc_batching(q, ioc))
703 return;
704
705 ioc->nr_batch_requests = q->nr_batching;
706 ioc->last_waited = jiffies;
707}
708
Jens Axboe1faa16d2009-04-06 14:48:01 +0200709static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 struct request_list *rl = &q->rq;
712
Jens Axboe1faa16d2009-04-06 14:48:01 +0200713 if (rl->count[sync] < queue_congestion_off_threshold(q))
714 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Jens Axboe1faa16d2009-04-06 14:48:01 +0200716 if (rl->count[sync] + 1 <= q->nr_requests) {
717 if (waitqueue_active(&rl->wait[sync]))
718 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Jens Axboe1faa16d2009-04-06 14:48:01 +0200720 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
722}
723
724/*
725 * A request has just been released. Account for it, update the full and
726 * congestion status, wake up any waiters. Called under q->queue_lock.
727 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200728static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
730 struct request_list *rl = &q->rq;
731
Jens Axboe1faa16d2009-04-06 14:48:01 +0200732 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200733 if (priv)
734 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Jens Axboe1faa16d2009-04-06 14:48:01 +0200736 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
Jens Axboe1faa16d2009-04-06 14:48:01 +0200738 if (unlikely(rl->starved[sync ^ 1]))
739 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740}
741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742/*
Nick Piggind6344532005-06-28 20:45:14 -0700743 * Get a free request, queue_lock must be held.
744 * Returns NULL on failure, with queue_lock held.
745 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Jens Axboe165125e2007-07-24 09:28:11 +0200747static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100748 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 struct request *rq = NULL;
751 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100752 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200753 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100754 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Jens Axboe7749a8d2006-12-13 13:02:26 +0100756 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100757 if (may_queue == ELV_MQUEUE_NO)
758 goto rq_starved;
759
Jens Axboe1faa16d2009-04-06 14:48:01 +0200760 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
761 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200762 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100763 /*
764 * The queue will fill after this allocation, so set
765 * it as full, and mark this process as "batching".
766 * This process will be allowed to complete a batch of
767 * requests, others will be blocked.
768 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200769 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100770 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200771 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100772 } else {
773 if (may_queue != ELV_MQUEUE_MUST
774 && !ioc_batching(q, ioc)) {
775 /*
776 * The queue is full and the allocating
777 * process is not a "batcher", and not
778 * exempted by the IO scheduler
779 */
780 goto out;
781 }
782 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200784 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
786
Jens Axboe082cf692005-06-28 16:35:11 +0200787 /*
788 * Only allow batching queuers to allocate up to 50% over the defined
789 * limit of requests, otherwise we could have thousands of requests
790 * allocated with any setting of ->nr_requests
791 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200792 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200793 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100794
Jens Axboe1faa16d2009-04-06 14:48:01 +0200795 rl->count[is_sync]++;
796 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200797
Jens Axboe64521d12005-10-28 08:30:39 +0200798 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200799 if (priv)
800 rl->elvpriv++;
801
Jerome Marchand42dad762009-04-22 14:01:49 +0200802 if (blk_queue_io_stat(q))
803 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 spin_unlock_irq(q->queue_lock);
805
Jens Axboe7749a8d2006-12-13 13:02:26 +0100806 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100807 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 /*
809 * Allocation failed presumably due to memory. Undo anything
810 * we might have messed up.
811 *
812 * Allocating task should really be put onto the front of the
813 * wait queue, but this is pretty rare.
814 */
815 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200816 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818 /*
819 * in the very unlikely event that allocation failed and no
820 * requests for this direction was pending, mark us starved
821 * so that freeing of a request in the other direction will
822 * notice us. another possible fix would be to split the
823 * rq mempool into READ and WRITE
824 */
825rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200826 if (unlikely(rl->count[is_sync] == 0))
827 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 goto out;
830 }
831
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100832 /*
833 * ioc may be NULL here, and ioc_batching will be false. That's
834 * OK, if the queue is under the request limit then requests need
835 * not count toward the nr_batch_requests limit. There will always
836 * be some limit enforced by BLK_BATCH_TIME.
837 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 if (ioc_batching(q, ioc))
839 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100840
Jens Axboe1faa16d2009-04-06 14:48:01 +0200841 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 return rq;
844}
845
846/*
847 * No available requests for this queue, unplug the device and wait for some
848 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700849 *
850 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 */
Jens Axboe165125e2007-07-24 09:28:11 +0200852static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200853 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200855 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 struct request *rq;
857
Jens Axboe7749a8d2006-12-13 13:02:26 +0100858 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700859 while (!rq) {
860 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200861 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 struct request_list *rl = &q->rq;
863
Jens Axboe1faa16d2009-04-06 14:48:01 +0200864 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 TASK_UNINTERRUPTIBLE);
866
Jens Axboe1faa16d2009-04-06 14:48:01 +0200867 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200869 __generic_unplug_device(q);
870 spin_unlock_irq(q->queue_lock);
871 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200873 /*
874 * After sleeping, we become a "batching" process and
875 * will be able to allocate at least one request, and
876 * up to a big batch of them for a small period time.
877 * See ioc_batching, ioc_set_batching
878 */
879 ioc = current_io_context(GFP_NOIO, q->node);
880 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100881
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200882 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200883 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200884
885 rq = get_request(q, rw_flags, bio, GFP_NOIO);
886 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888 return rq;
889}
890
Jens Axboe165125e2007-07-24 09:28:11 +0200891struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
893 struct request *rq;
894
895 BUG_ON(rw != READ && rw != WRITE);
896
Nick Piggind6344532005-06-28 20:45:14 -0700897 spin_lock_irq(q->queue_lock);
898 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200899 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700900 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200901 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700902 if (!rq)
903 spin_unlock_irq(q->queue_lock);
904 }
905 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 return rq;
908}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909EXPORT_SYMBOL(blk_get_request);
910
911/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300912 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700913 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300914 * @bio: The bio describing the memory mappings that will be submitted for IO.
915 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700916 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200917 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300918 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
919 * type commands. Where the struct request needs to be farther initialized by
920 * the caller. It is passed a &struct bio, which describes the memory info of
921 * the I/O transfer.
922 *
923 * The caller of blk_make_request must make sure that bi_io_vec
924 * are set to describe the memory buffers. That bio_data_dir() will return
925 * the needed direction of the request. (And all bio's in the passed bio-chain
926 * are properly set accordingly)
927 *
928 * If called under none-sleepable conditions, mapped bio buffers must not
929 * need bouncing, by calling the appropriate masked or flagged allocator,
930 * suitable for the target device. Otherwise the call to blk_queue_bounce will
931 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200932 *
933 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
934 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
935 * anything but the first bio in the chain. Otherwise you risk waiting for IO
936 * completion of a bio that hasn't been submitted yet, thus resulting in a
937 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
938 * of bio_alloc(), as that avoids the mempool deadlock.
939 * If possible a big IO should be split into smaller parts when allocation
940 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200941 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300942struct request *blk_make_request(struct request_queue *q, struct bio *bio,
943 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200944{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300945 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
946
947 if (unlikely(!rq))
948 return ERR_PTR(-ENOMEM);
949
950 for_each_bio(bio) {
951 struct bio *bounce_bio = bio;
952 int ret;
953
954 blk_queue_bounce(q, &bounce_bio);
955 ret = blk_rq_append_bio(q, rq, bounce_bio);
956 if (unlikely(ret)) {
957 blk_put_request(rq);
958 return ERR_PTR(ret);
959 }
960 }
961
962 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200963}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300964EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200965
966/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 * blk_requeue_request - put a request back on queue
968 * @q: request queue where request should be inserted
969 * @rq: request to be inserted
970 *
971 * Description:
972 * Drivers often keep queueing requests until the hardware cannot accept
973 * more, when that condition happens we need to put the request back
974 * on the queue. Must be called with queue lock held.
975 */
Jens Axboe165125e2007-07-24 09:28:11 +0200976void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700978 blk_delete_timer(rq);
979 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100980 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 if (blk_rq_tagged(rq))
983 blk_queue_end_tag(q, rq);
984
James Bottomleyba396a62009-05-27 14:17:08 +0200985 BUG_ON(blk_queued_rq(rq));
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 elv_requeue_request(q, rq);
988}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989EXPORT_SYMBOL(blk_requeue_request);
990
991/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200992 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 * @q: request queue where request should be inserted
994 * @rq: request to be inserted
995 * @at_head: insert request at head or tail of queue
996 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 *
998 * Description:
999 * Many block devices need to execute commands asynchronously, so they don't
1000 * block the whole kernel from preemption during request execution. This is
1001 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +02001002 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1003 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 *
1005 * We have the option of inserting the head or the tail of the queue.
1006 * Typically we use the tail for new ioctls and so forth. We use the head
1007 * of the queue for things like a QUEUE_FULL message from a device, or a
1008 * host that is unable to accept a particular command.
1009 */
Jens Axboe165125e2007-07-24 09:28:11 +02001010void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05001011 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012{
Tejun Heo 867d1192005-04-24 02:06:05 -05001013 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 unsigned long flags;
1015
1016 /*
1017 * tell I/O scheduler that this isn't a regular read/write (ie it
1018 * must not attempt merges on this) and that it acts as a soft
1019 * barrier
1020 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02001021 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 rq->special = data;
1024
1025 spin_lock_irqsave(q->queue_lock, flags);
1026
1027 /*
1028 * If command is tagged, release the tag
1029 */
Tejun Heo 867d1192005-04-24 02:06:05 -05001030 if (blk_rq_tagged(rq))
1031 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001033 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05001034 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +09001035 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 spin_unlock_irqrestore(q->queue_lock, flags);
1037}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038EXPORT_SYMBOL(blk_insert_request);
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040/*
1041 * add-request adds a request to the linked list.
1042 * queue lock is held and interrupts disabled, as we muck with the
1043 * request queue list.
1044 */
Jens Axboe6728cb02008-01-31 13:03:55 +01001045static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02001047 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 /*
1050 * elevator indicated where it wants this request to be
1051 * inserted at elevator_merge time
1052 */
1053 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1054}
Jens Axboe6728cb02008-01-31 13:03:55 +01001055
Tejun Heo074a7ac2008-08-25 19:56:14 +09001056static void part_round_stats_single(int cpu, struct hd_struct *part,
1057 unsigned long now)
1058{
1059 if (now == part->stamp)
1060 return;
1061
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001062 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001063 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001064 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001065 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1066 }
1067 part->stamp = now;
1068}
1069
1070/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001071 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1072 * @cpu: cpu number for stats access
1073 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 *
1075 * The average IO queue length and utilisation statistics are maintained
1076 * by observing the current state of the queue length and the amount of
1077 * time it has been in this state for.
1078 *
1079 * Normally, that accounting is done on IO completion, but that can result
1080 * in more than a second's worth of IO being accounted for within any one
1081 * second, leading to >100% utilisation. To deal with that, we call this
1082 * function to do a round-off before returning the results when reading
1083 * /proc/diskstats. This accounts immediately for all queue usage up to
1084 * the current jiffies and restarts the counters again.
1085 */
Tejun Heoc9959052008-08-25 19:47:21 +09001086void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001087{
1088 unsigned long now = jiffies;
1089
Tejun Heo074a7ac2008-08-25 19:56:14 +09001090 if (part->partno)
1091 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1092 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001093}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001094EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096/*
1097 * queue lock must be held
1098 */
Jens Axboe165125e2007-07-24 09:28:11 +02001099void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (unlikely(!q))
1102 return;
1103 if (unlikely(--req->ref_count))
1104 return;
1105
Tejun Heo8922e162005-10-20 16:23:44 +02001106 elv_completed_request(q, req);
1107
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001108 /* this is a bio leak */
1109 WARN_ON(req->bio != NULL);
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 /*
1112 * Request may not have originated from ll_rw_blk. if not,
1113 * it didn't come out of our reserved rq pools
1114 */
Jens Axboe49171e52006-08-10 08:59:11 +02001115 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001116 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001117 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001120 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001123 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
1125}
Mike Christie6e39b692005-11-11 05:30:24 -06001126EXPORT_SYMBOL_GPL(__blk_put_request);
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128void blk_put_request(struct request *req)
1129{
Tejun Heo8922e162005-10-20 16:23:44 +02001130 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001131 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001133 spin_lock_irqsave(q->queue_lock, flags);
1134 __blk_put_request(q, req);
1135 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137EXPORT_SYMBOL(blk_put_request);
1138
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001139/**
1140 * blk_add_request_payload - add a payload to a request
1141 * @rq: request to update
1142 * @page: page backing the payload
1143 * @len: length of the payload.
1144 *
1145 * This allows to later add a payload to an already submitted request by
1146 * a block driver. The driver needs to take care of freeing the payload
1147 * itself.
1148 *
1149 * Note that this is a quite horrible hack and nothing but handling of
1150 * discard requests should ever use it.
1151 */
1152void blk_add_request_payload(struct request *rq, struct page *page,
1153 unsigned int len)
1154{
1155 struct bio *bio = rq->bio;
1156
1157 bio->bi_io_vec->bv_page = page;
1158 bio->bi_io_vec->bv_offset = 0;
1159 bio->bi_io_vec->bv_len = len;
1160
1161 bio->bi_size = len;
1162 bio->bi_vcnt = 1;
1163 bio->bi_phys_segments = 1;
1164
1165 rq->__data_len = rq->resid_len = len;
1166 rq->nr_phys_segments = 1;
1167 rq->buffer = bio_data(bio);
1168}
1169EXPORT_SYMBOL_GPL(blk_add_request_payload);
1170
Jens Axboe86db1e22008-01-29 14:53:40 +01001171void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001172{
Jens Axboec7c22e42008-09-13 20:26:01 +02001173 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001174 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001175
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001176 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1177 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001178 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001179
Tejun Heo52d9e672006-01-06 09:49:58 +01001180 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001181 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001182 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001183 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001184}
1185
Jens Axboe644b2d92009-04-06 14:48:06 +02001186/*
1187 * Only disabling plugging for non-rotational devices if it does tagging
1188 * as well, otherwise we do need the proper merging
1189 */
1190static inline bool queue_should_plug(struct request_queue *q)
1191{
Jens Axboe79da0642010-02-23 08:40:43 +01001192 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
Jens Axboe644b2d92009-04-06 14:48:06 +02001193}
1194
Jens Axboe165125e2007-07-24 09:28:11 +02001195static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196{
Nick Piggin450991b2005-06-28 20:45:13 -07001197 struct request *req;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001198 int el_ret;
1199 unsigned int bytes = bio->bi_size;
Jens Axboe51da90f2006-07-18 04:14:45 +02001200 const unsigned short prio = bio_prio(bio);
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001201 const bool sync = (bio->bi_rw & REQ_SYNC);
1202 const bool unplug = (bio->bi_rw & REQ_UNPLUG);
Tejun Heo80a761f2009-07-03 17:48:17 +09001203 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001204 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001206 if ((bio->bi_rw & REQ_HARDBARRIER) &&
NeilBrowndb64f682009-06-30 09:35:44 +02001207 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1208 bio_endio(bio, -EOPNOTSUPP);
1209 return 0;
1210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 /*
1212 * low level driver can indicate that it wants pages above a
1213 * certain limit bounced to low memory (ie for highmem, or even
1214 * ISA dma in theory)
1215 */
1216 blk_queue_bounce(q, &bio);
1217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 spin_lock_irq(q->queue_lock);
1219
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001220 if (unlikely((bio->bi_rw & REQ_HARDBARRIER)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 goto get_rq;
1222
1223 el_ret = elv_merge(q, &req, bio);
1224 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001225 case ELEVATOR_BACK_MERGE:
1226 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
Jens Axboe6728cb02008-01-31 13:03:55 +01001228 if (!ll_back_merge_fn(q, req, bio))
1229 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001231 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001232
Tejun Heo80a761f2009-07-03 17:48:17 +09001233 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1234 blk_rq_set_mixed_merge(req);
1235
Jens Axboe6728cb02008-01-31 13:03:55 +01001236 req->biotail->bi_next = bio;
1237 req->biotail = bio;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001238 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001239 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001240 if (!blk_rq_cpu_valid(req))
1241 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001242 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001243 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001244 if (!attempt_back_merge(q, req))
1245 elv_merged_request(q, req, el_ret);
1246 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Jens Axboe6728cb02008-01-31 13:03:55 +01001248 case ELEVATOR_FRONT_MERGE:
1249 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Jens Axboe6728cb02008-01-31 13:03:55 +01001251 if (!ll_front_merge_fn(q, req, bio))
1252 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001254 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001255
Tejun Heo80a761f2009-07-03 17:48:17 +09001256 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1257 blk_rq_set_mixed_merge(req);
1258 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1259 req->cmd_flags |= ff;
1260 }
1261
Jens Axboe6728cb02008-01-31 13:03:55 +01001262 bio->bi_next = req->bio;
1263 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Jens Axboe6728cb02008-01-31 13:03:55 +01001265 /*
1266 * may not be valid. if the low level driver said
1267 * it didn't need a bounce buffer then it better
1268 * not touch req->buffer either...
1269 */
1270 req->buffer = bio_data(bio);
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001271 req->__sector = bio->bi_sector;
1272 req->__data_len += bytes;
Jens Axboe6728cb02008-01-31 13:03:55 +01001273 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001274 if (!blk_rq_cpu_valid(req))
1275 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001276 drive_stat_acct(req, 0);
Divyesh Shah812d4022010-04-08 21:14:23 -07001277 elv_bio_merged(q, req, bio);
Jens Axboe6728cb02008-01-31 13:03:55 +01001278 if (!attempt_front_merge(q, req))
1279 elv_merged_request(q, req, el_ret);
1280 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
Jens Axboe6728cb02008-01-31 13:03:55 +01001282 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1283 default:
1284 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 }
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001288 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001289 * This sync check and mask will be re-done in init_request_from_bio(),
1290 * but we need to set it earlier to expose the sync flag to the
1291 * rq allocator and io schedulers.
1292 */
1293 rw_flags = bio_data_dir(bio);
1294 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001295 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001296
1297 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001298 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001299 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001300 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001301 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001302
Nick Piggin450991b2005-06-28 20:45:13 -07001303 /*
1304 * After dropping the lock and possibly sleeping here, our request
1305 * may now be mergeable after it had proven unmergeable (above).
1306 * We don't worry about that case for efficiency. It won't happen
1307 * often, and the elevators are able to handle it.
1308 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001309 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Nick Piggin450991b2005-06-28 20:45:13 -07001311 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001312 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1313 bio_flagged(bio, BIO_CPU_AFFINE))
1314 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001315 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001316 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 add_request(q, req);
1318out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001319 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 spin_unlock_irq(q->queue_lock);
1322 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323}
1324
1325/*
1326 * If bio->bi_dev is a partition, remap the location
1327 */
1328static inline void blk_partition_remap(struct bio *bio)
1329{
1330 struct block_device *bdev = bio->bi_bdev;
1331
Jens Axboebf2de6f2007-09-27 13:01:25 +02001332 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 bio->bi_sector += p->start_sect;
1336 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001337
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001338 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001339 bdev->bd_dev,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001340 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 }
1342}
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344static void handle_bad_sector(struct bio *bio)
1345{
1346 char b[BDEVNAME_SIZE];
1347
1348 printk(KERN_INFO "attempt to access beyond end of device\n");
1349 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1350 bdevname(bio->bi_bdev, b),
1351 bio->bi_rw,
1352 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1353 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1354
1355 set_bit(BIO_EOF, &bio->bi_flags);
1356}
1357
Akinobu Mitac17bb492006-12-08 02:39:46 -08001358#ifdef CONFIG_FAIL_MAKE_REQUEST
1359
1360static DECLARE_FAULT_ATTR(fail_make_request);
1361
1362static int __init setup_fail_make_request(char *str)
1363{
1364 return setup_fault_attr(&fail_make_request, str);
1365}
1366__setup("fail_make_request=", setup_fail_make_request);
1367
1368static int should_fail_request(struct bio *bio)
1369{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001370 struct hd_struct *part = bio->bi_bdev->bd_part;
1371
1372 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001373 return should_fail(&fail_make_request, bio->bi_size);
1374
1375 return 0;
1376}
1377
1378static int __init fail_make_request_debugfs(void)
1379{
1380 return init_fault_attr_dentries(&fail_make_request,
1381 "fail_make_request");
1382}
1383
1384late_initcall(fail_make_request_debugfs);
1385
1386#else /* CONFIG_FAIL_MAKE_REQUEST */
1387
1388static inline int should_fail_request(struct bio *bio)
1389{
1390 return 0;
1391}
1392
1393#endif /* CONFIG_FAIL_MAKE_REQUEST */
1394
Jens Axboec07e2b42007-07-18 13:27:58 +02001395/*
1396 * Check whether this bio extends beyond the end of the device.
1397 */
1398static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1399{
1400 sector_t maxsector;
1401
1402 if (!nr_sectors)
1403 return 0;
1404
1405 /* Test device or partition size, when known. */
1406 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1407 if (maxsector) {
1408 sector_t sector = bio->bi_sector;
1409
1410 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1411 /*
1412 * This may well happen - the kernel calls bread()
1413 * without checking the size of the device, e.g., when
1414 * mounting a device.
1415 */
1416 handle_bad_sector(bio);
1417 return 1;
1418 }
1419 }
1420
1421 return 0;
1422}
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001425 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * @bio: The bio describing the location in memory and on the device.
1427 *
1428 * generic_make_request() is used to make I/O requests of block
1429 * devices. It is passed a &struct bio, which describes the I/O that needs
1430 * to be done.
1431 *
1432 * generic_make_request() does not return any status. The
1433 * success/failure status of the request, along with notification of
1434 * completion, is delivered asynchronously through the bio->bi_end_io
1435 * function described (one day) else where.
1436 *
1437 * The caller of generic_make_request must make sure that bi_io_vec
1438 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1439 * set to describe the device address, and the
1440 * bi_end_io and optionally bi_private are set to describe how
1441 * completion notification should be signaled.
1442 *
1443 * generic_make_request and the drivers it calls may use bi_next if this
1444 * bio happens to be merged with someone else, and may change bi_dev and
1445 * bi_sector for remaps as it sees fit. So the values of these fields
1446 * should NOT be depended on after the call to generic_make_request.
1447 */
Neil Brownd89d8792007-05-01 09:53:42 +02001448static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Jens Axboe165125e2007-07-24 09:28:11 +02001450 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001451 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001453 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001454 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Jens Axboec07e2b42007-07-18 13:27:58 +02001458 if (bio_check_eod(bio, nr_sectors))
1459 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
1461 /*
1462 * Resolve the mapping until finished. (drivers are
1463 * still free to implement/resolve their own stacking
1464 * by explicitly returning 0)
1465 *
1466 * NOTE: we don't repeat the blk_size check for each new device.
1467 * Stacking drivers are expected to know what they are doing.
1468 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001469 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001470 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 do {
1472 char b[BDEVNAME_SIZE];
1473
1474 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001475 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 printk(KERN_ERR
1477 "generic_make_request: Trying to access "
1478 "nonexistent block-device %s (%Lu)\n",
1479 bdevname(bio->bi_bdev, b),
1480 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001481 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001484 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001485 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001486 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001487 bdevname(bio->bi_bdev, b),
1488 bio_sectors(bio),
1489 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 goto end_io;
1491 }
1492
Nick Pigginfde6ad22005-06-23 00:08:53 -07001493 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 goto end_io;
1495
Akinobu Mitac17bb492006-12-08 02:39:46 -08001496 if (should_fail_request(bio))
1497 goto end_io;
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 /*
1500 * If this device has partitions, remap block n
1501 * of partition p to block n+start(p) of the disk.
1502 */
1503 blk_partition_remap(bio);
1504
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001505 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1506 goto end_io;
1507
NeilBrown5ddfe962006-10-30 22:07:21 -08001508 if (old_sector != -1)
Alan D. Brunelle22a7c312009-05-04 16:35:08 -04001509 trace_block_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001510
NeilBrown5ddfe962006-10-30 22:07:21 -08001511 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001512 old_dev = bio->bi_bdev->bd_dev;
1513
Jens Axboec07e2b42007-07-18 13:27:58 +02001514 if (bio_check_eod(bio, nr_sectors))
1515 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001516
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001517 if ((bio->bi_rw & REQ_DISCARD) && !blk_queue_discard(q)) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001518 err = -EOPNOTSUPP;
1519 goto end_io;
1520 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001521
Minchan Kim01edede2009-09-08 21:56:38 +02001522 trace_block_bio_queue(q, bio);
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 ret = q->make_request_fn(q, bio);
1525 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001526
1527 return;
1528
1529end_io:
1530 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
Neil Brownd89d8792007-05-01 09:53:42 +02001533/*
1534 * We only want one ->make_request_fn to be active at a time,
1535 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001536 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001537 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001538 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001539 * generic_make_request is currently active in this task or not.
1540 * If it is NULL, then no make_request is active. If it is non-NULL,
1541 * then a make_request is active, and new requests should be added
1542 * at the tail
1543 */
1544void generic_make_request(struct bio *bio)
1545{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001546 struct bio_list bio_list_on_stack;
1547
1548 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001549 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001550 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001551 return;
1552 }
1553 /* following loop may be a bit non-obvious, and so deserves some
1554 * explanation.
1555 * Before entering the loop, bio->bi_next is NULL (as all callers
1556 * ensure that) so we have a list with a single bio.
1557 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001558 * we assign bio_list to a pointer to the bio_list_on_stack,
1559 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001560 * added. __generic_make_request may indeed add some more bios
1561 * through a recursive call to generic_make_request. If it
1562 * did, we find a non-NULL value in bio_list and re-enter the loop
1563 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001564 * of the top of the list (no pretending) and so remove it from
1565 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001566 *
1567 * The loop was structured like this to make only one call to
1568 * __generic_make_request (which is important as it is large and
1569 * inlined) and to keep the structure simple.
1570 */
1571 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001572 bio_list_init(&bio_list_on_stack);
1573 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001574 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001575 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001577 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001578 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001579}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580EXPORT_SYMBOL(generic_make_request);
1581
1582/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001583 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1585 * @bio: The &struct bio which describes the I/O
1586 *
1587 * submit_bio() is very similar in purpose to generic_make_request(), and
1588 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001589 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 *
1591 */
1592void submit_bio(int rw, struct bio *bio)
1593{
1594 int count = bio_sectors(bio);
1595
Jens Axboe22e2c502005-06-27 10:55:12 +02001596 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Jens Axboebf2de6f2007-09-27 13:01:25 +02001598 /*
1599 * If it's a regular read/write or a barrier with data attached,
1600 * go through the normal accounting stuff before submission.
1601 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001602 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001603 if (rw & WRITE) {
1604 count_vm_events(PGPGOUT, count);
1605 } else {
1606 task_io_account_read(bio->bi_size);
1607 count_vm_events(PGPGIN, count);
1608 }
1609
1610 if (unlikely(block_dump)) {
1611 char b[BDEVNAME_SIZE];
1612 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001613 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001614 (rw & WRITE) ? "WRITE" : "READ",
1615 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001616 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 }
1619
1620 generic_make_request(bio);
1621}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622EXPORT_SYMBOL(submit_bio);
1623
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001624/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001625 * blk_rq_check_limits - Helper function to check a request for the queue limit
1626 * @q: the queue
1627 * @rq: the request being checked
1628 *
1629 * Description:
1630 * @rq may have been made based on weaker limitations of upper-level queues
1631 * in request stacking drivers, and it may violate the limitation of @q.
1632 * Since the block layer and the underlying device driver trust @rq
1633 * after it is inserted to @q, it should be checked against @q before
1634 * the insertion using this generic function.
1635 *
1636 * This function should also be useful for request stacking drivers
1637 * in some cases below, so export this fuction.
1638 * Request stacking drivers like request-based dm may change the queue
1639 * limits while requests are in the queue (e.g. dm's table swapping).
1640 * Such request stacking drivers should check those requests agaist
1641 * the new queue limits again when they dispatch those requests,
1642 * although such checkings are also done against the old queue limits
1643 * when submitting requests.
1644 */
1645int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1646{
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001647 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1648 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001649 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1650 return -EIO;
1651 }
1652
1653 /*
1654 * queue's settings related to segment counting like q->bounce_pfn
1655 * may differ from that of other stacking queues.
1656 * Recalculate it to check the request correctly on this queue's
1657 * limitation.
1658 */
1659 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001660 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001661 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1662 return -EIO;
1663 }
1664
1665 return 0;
1666}
1667EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1668
1669/**
1670 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1671 * @q: the queue to submit the request
1672 * @rq: the request being queued
1673 */
1674int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1675{
1676 unsigned long flags;
1677
1678 if (blk_rq_check_limits(q, rq))
1679 return -EIO;
1680
1681#ifdef CONFIG_FAIL_MAKE_REQUEST
1682 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1683 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1684 return -EIO;
1685#endif
1686
1687 spin_lock_irqsave(q->queue_lock, flags);
1688
1689 /*
1690 * Submitting request must be dequeued before calling this function
1691 * because it will be linked to another request_queue
1692 */
1693 BUG_ON(blk_queued_rq(rq));
1694
1695 drive_stat_acct(rq, 1);
1696 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1697
1698 spin_unlock_irqrestore(q->queue_lock, flags);
1699
1700 return 0;
1701}
1702EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1703
Tejun Heo80a761f2009-07-03 17:48:17 +09001704/**
1705 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1706 * @rq: request to examine
1707 *
1708 * Description:
1709 * A request could be merge of IOs which require different failure
1710 * handling. This function determines the number of bytes which
1711 * can be failed from the beginning of the request without
1712 * crossing into area which need to be retried further.
1713 *
1714 * Return:
1715 * The number of bytes to fail.
1716 *
1717 * Context:
1718 * queue_lock must be held.
1719 */
1720unsigned int blk_rq_err_bytes(const struct request *rq)
1721{
1722 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1723 unsigned int bytes = 0;
1724 struct bio *bio;
1725
1726 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1727 return blk_rq_bytes(rq);
1728
1729 /*
1730 * Currently the only 'mixing' which can happen is between
1731 * different fastfail types. We can safely fail portions
1732 * which have all the failfast bits that the first one has -
1733 * the ones which are at least as eager to fail as the first
1734 * one.
1735 */
1736 for (bio = rq->bio; bio; bio = bio->bi_next) {
1737 if ((bio->bi_rw & ff) != ff)
1738 break;
1739 bytes += bio->bi_size;
1740 }
1741
1742 /* this could lead to infinite loop */
1743 BUG_ON(blk_rq_bytes(rq) && !bytes);
1744 return bytes;
1745}
1746EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1747
Jens Axboebc58ba92009-01-23 10:54:44 +01001748static void blk_account_io_completion(struct request *req, unsigned int bytes)
1749{
Jens Axboec2553b52009-04-24 08:10:11 +02001750 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001751 const int rw = rq_data_dir(req);
1752 struct hd_struct *part;
1753 int cpu;
1754
1755 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001756 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001757 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1758 part_stat_unlock();
1759 }
1760}
1761
1762static void blk_account_io_done(struct request *req)
1763{
Jens Axboebc58ba92009-01-23 10:54:44 +01001764 /*
1765 * Account IO completion. bar_rq isn't accounted as a normal
1766 * IO on queueing nor completion. Accounting the containing
1767 * request is enough.
1768 */
Jens Axboec2553b52009-04-24 08:10:11 +02001769 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001770 unsigned long duration = jiffies - req->start_time;
1771 const int rw = rq_data_dir(req);
1772 struct hd_struct *part;
1773 int cpu;
1774
1775 cpu = part_stat_lock();
Tejun Heo83096eb2009-05-07 22:24:39 +09001776 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001777
1778 part_stat_inc(cpu, part, ios[rw]);
1779 part_stat_add(cpu, part, ticks[rw], duration);
1780 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001781 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001782
1783 part_stat_unlock();
1784 }
1785}
1786
Tejun Heo53a08802008-12-03 12:41:26 +01001787/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001788 * blk_peek_request - peek at the top of a request queue
1789 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001790 *
1791 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001792 * Return the request at the top of @q. The returned request
1793 * should be started using blk_start_request() before LLD starts
1794 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001795 *
1796 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001797 * Pointer to the request at the top of @q if available. Null
1798 * otherwise.
1799 *
1800 * Context:
1801 * queue_lock must be held.
1802 */
1803struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001804{
1805 struct request *rq;
1806 int ret;
1807
1808 while ((rq = __elv_next_request(q)) != NULL) {
1809 if (!(rq->cmd_flags & REQ_STARTED)) {
1810 /*
1811 * This is the first time the device driver
1812 * sees this request (possibly after
1813 * requeueing). Notify IO scheduler.
1814 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001815 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001816 elv_activate_rq(q, rq);
1817
1818 /*
1819 * just mark as started even if we don't start
1820 * it, a request that has been delayed should
1821 * not be passed by new incoming requests
1822 */
1823 rq->cmd_flags |= REQ_STARTED;
1824 trace_block_rq_issue(q, rq);
1825 }
1826
1827 if (!q->boundary_rq || q->boundary_rq == rq) {
1828 q->end_sector = rq_end_sector(rq);
1829 q->boundary_rq = NULL;
1830 }
1831
1832 if (rq->cmd_flags & REQ_DONTPREP)
1833 break;
1834
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001835 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001836 /*
1837 * make sure space for the drain appears we
1838 * know we can do this because max_hw_segments
1839 * has been adjusted to be one fewer than the
1840 * device can handle
1841 */
1842 rq->nr_phys_segments++;
1843 }
1844
1845 if (!q->prep_rq_fn)
1846 break;
1847
1848 ret = q->prep_rq_fn(q, rq);
1849 if (ret == BLKPREP_OK) {
1850 break;
1851 } else if (ret == BLKPREP_DEFER) {
1852 /*
1853 * the request may have been (partially) prepped.
1854 * we need to keep this request in the front to
1855 * avoid resource deadlock. REQ_STARTED will
1856 * prevent other fs requests from passing this one.
1857 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001858 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001859 !(rq->cmd_flags & REQ_DONTPREP)) {
1860 /*
1861 * remove the space for the drain we added
1862 * so that we don't add it again
1863 */
1864 --rq->nr_phys_segments;
1865 }
1866
1867 rq = NULL;
1868 break;
1869 } else if (ret == BLKPREP_KILL) {
1870 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001871 /*
1872 * Mark this request as started so we don't trigger
1873 * any debug logic in the end I/O path.
1874 */
1875 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001876 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001877 } else {
1878 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1879 break;
1880 }
1881 }
1882
1883 return rq;
1884}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001885EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001886
Tejun Heo9934c8c2009-05-08 11:54:16 +09001887void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001888{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001889 struct request_queue *q = rq->q;
1890
Tejun Heo158dbda2009-04-23 11:05:18 +09001891 BUG_ON(list_empty(&rq->queuelist));
1892 BUG_ON(ELV_ON_HASH(rq));
1893
1894 list_del_init(&rq->queuelist);
1895
1896 /*
1897 * the time frame between a request being removed from the lists
1898 * and to it is freed is accounted as io that is in progress at
1899 * the driver side.
1900 */
Divyesh Shah91952912010-04-01 15:01:41 -07001901 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001902 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001903 set_io_start_time_ns(rq);
1904 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001905}
1906
Tejun Heo5efccd12009-04-23 11:05:18 +09001907/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001908 * blk_start_request - start request processing on the driver
1909 * @req: request to dequeue
1910 *
1911 * Description:
1912 * Dequeue @req and start timeout timer on it. This hands off the
1913 * request to the driver.
1914 *
1915 * Block internal functions which don't want to start timer should
1916 * call blk_dequeue_request().
1917 *
1918 * Context:
1919 * queue_lock must be held.
1920 */
1921void blk_start_request(struct request *req)
1922{
1923 blk_dequeue_request(req);
1924
1925 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001926 * We are now handing the request to the hardware, initialize
1927 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001928 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001929 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001930 if (unlikely(blk_bidi_rq(req)))
1931 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1932
Tejun Heo9934c8c2009-05-08 11:54:16 +09001933 blk_add_timer(req);
1934}
1935EXPORT_SYMBOL(blk_start_request);
1936
1937/**
1938 * blk_fetch_request - fetch a request from a request queue
1939 * @q: request queue to fetch a request from
1940 *
1941 * Description:
1942 * Return the request at the top of @q. The request is started on
1943 * return and LLD can start processing it immediately.
1944 *
1945 * Return:
1946 * Pointer to the request at the top of @q if available. Null
1947 * otherwise.
1948 *
1949 * Context:
1950 * queue_lock must be held.
1951 */
1952struct request *blk_fetch_request(struct request_queue *q)
1953{
1954 struct request *rq;
1955
1956 rq = blk_peek_request(q);
1957 if (rq)
1958 blk_start_request(rq);
1959 return rq;
1960}
1961EXPORT_SYMBOL(blk_fetch_request);
1962
1963/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001964 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001965 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001966 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001967 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001968 *
1969 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001970 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1971 * the request structure even if @req doesn't have leftover.
1972 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001973 *
1974 * This special helper function is only for request stacking drivers
1975 * (e.g. request-based dm) so that they can handle partial completion.
1976 * Actual device drivers should use blk_end_request instead.
1977 *
1978 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1979 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001980 *
1981 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09001982 * %false - this request doesn't have any more data
1983 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001984 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09001985bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001987 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 struct bio *bio;
1989
Tejun Heo2e60e022009-04-23 11:05:18 +09001990 if (!req->bio)
1991 return false;
1992
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001993 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 /*
Tejun Heo6f414692009-04-19 07:00:41 +09001996 * For fs requests, rq is just carrier of independent bio's
1997 * and each partial completion should be handled separately.
1998 * Reset per-request error on each partial completion.
1999 *
2000 * TODO: tj: This is too subtle. It would be better to let
2001 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002003 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 req->errors = 0;
2005
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002006 if (error && req->cmd_type == REQ_TYPE_FS &&
2007 !(req->cmd_flags & REQ_QUIET)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01002008 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 req->rq_disk ? req->rq_disk->disk_name : "?",
Tejun Heo83096eb2009-05-07 22:24:39 +09002010 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 }
2012
Jens Axboebc58ba92009-01-23 10:54:44 +01002013 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002014
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 total_bytes = bio_nbytes = 0;
2016 while ((bio = req->bio) != NULL) {
2017 int nbytes;
2018
2019 if (nr_bytes >= bio->bi_size) {
2020 req->bio = bio->bi_next;
2021 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002022 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 next_idx = 0;
2024 bio_nbytes = 0;
2025 } else {
2026 int idx = bio->bi_idx + next_idx;
2027
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002028 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002030 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002031 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 break;
2033 }
2034
2035 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2036 BIO_BUG_ON(nbytes > bio->bi_size);
2037
2038 /*
2039 * not a complete bvec done
2040 */
2041 if (unlikely(nbytes > nr_bytes)) {
2042 bio_nbytes += nr_bytes;
2043 total_bytes += nr_bytes;
2044 break;
2045 }
2046
2047 /*
2048 * advance to the next vector
2049 */
2050 next_idx++;
2051 bio_nbytes += nbytes;
2052 }
2053
2054 total_bytes += nbytes;
2055 nr_bytes -= nbytes;
2056
Jens Axboe6728cb02008-01-31 13:03:55 +01002057 bio = req->bio;
2058 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 /*
2060 * end more in this run, or just return 'not-done'
2061 */
2062 if (unlikely(nr_bytes <= 0))
2063 break;
2064 }
2065 }
2066
2067 /*
2068 * completely done
2069 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002070 if (!req->bio) {
2071 /*
2072 * Reset counters so that the request stacking driver
2073 * can find how many bytes remain in the request
2074 * later.
2075 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002076 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002077 return false;
2078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
2080 /*
2081 * if the request wasn't completed, update state
2082 */
2083 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002084 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 bio->bi_idx += next_idx;
2086 bio_iovec(bio)->bv_offset += nr_bytes;
2087 bio_iovec(bio)->bv_len -= nr_bytes;
2088 }
2089
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002090 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002091 req->buffer = bio_data(req->bio);
2092
2093 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002094 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002095 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002096
Tejun Heo80a761f2009-07-03 17:48:17 +09002097 /* mixed attributes always follow the first bio */
2098 if (req->cmd_flags & REQ_MIXED_MERGE) {
2099 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2100 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2101 }
2102
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002103 /*
2104 * If total number of sectors is less than the first segment
2105 * size, something has gone terribly wrong.
2106 */
2107 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2108 printk(KERN_ERR "blk: request botched\n");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002109 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002110 }
2111
2112 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002114
Tejun Heo2e60e022009-04-23 11:05:18 +09002115 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116}
Tejun Heo2e60e022009-04-23 11:05:18 +09002117EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Tejun Heo2e60e022009-04-23 11:05:18 +09002119static bool blk_update_bidi_request(struct request *rq, int error,
2120 unsigned int nr_bytes,
2121 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002122{
Tejun Heo2e60e022009-04-23 11:05:18 +09002123 if (blk_update_request(rq, error, nr_bytes))
2124 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002125
Tejun Heo2e60e022009-04-23 11:05:18 +09002126 /* Bidi request must be completed as a whole */
2127 if (unlikely(blk_bidi_rq(rq)) &&
2128 blk_update_request(rq->next_rq, error, bidi_bytes))
2129 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002130
Jens Axboee2e1a142010-06-09 10:42:09 +02002131 if (blk_queue_add_random(rq->q))
2132 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002133
2134 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135}
2136
James Bottomley28018c22010-07-01 19:49:17 +09002137/**
2138 * blk_unprep_request - unprepare a request
2139 * @req: the request
2140 *
2141 * This function makes a request ready for complete resubmission (or
2142 * completion). It happens only after all error handling is complete,
2143 * so represents the appropriate moment to deallocate any resources
2144 * that were allocated to the request in the prep_rq_fn. The queue
2145 * lock is held when calling this.
2146 */
2147void blk_unprep_request(struct request *req)
2148{
2149 struct request_queue *q = req->q;
2150
2151 req->cmd_flags &= ~REQ_DONTPREP;
2152 if (q->unprep_rq_fn)
2153 q->unprep_rq_fn(q, req);
2154}
2155EXPORT_SYMBOL_GPL(blk_unprep_request);
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157/*
2158 * queue lock must be held
2159 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002160static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002162 if (blk_rq_tagged(req))
2163 blk_queue_end_tag(req->q, req);
2164
James Bottomleyba396a62009-05-27 14:17:08 +02002165 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002167 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002168 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
Mike Andersone78042e2008-10-30 02:16:20 -07002170 blk_delete_timer(req);
2171
James Bottomley28018c22010-07-01 19:49:17 +09002172 if (req->cmd_flags & REQ_DONTPREP)
2173 blk_unprep_request(req);
2174
2175
Jens Axboebc58ba92009-01-23 10:54:44 +01002176 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002177
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002179 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002180 else {
2181 if (blk_bidi_rq(req))
2182 __blk_put_request(req->next_rq->q, req->next_rq);
2183
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 }
2186}
2187
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002188/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002189 * blk_end_bidi_request - Complete a bidi request
2190 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002191 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002192 * @nr_bytes: number of bytes to complete @rq
2193 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002194 *
2195 * Description:
2196 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002197 * Drivers that supports bidi can safely call this member for any
2198 * type of request, bidi or uni. In the later case @bidi_bytes is
2199 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002200 *
2201 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002202 * %false - we are done with this request
2203 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002204 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002205static bool blk_end_bidi_request(struct request *rq, int error,
2206 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002207{
2208 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002209 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002210
Tejun Heo2e60e022009-04-23 11:05:18 +09002211 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2212 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002213
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002214 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002215 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002216 spin_unlock_irqrestore(q->queue_lock, flags);
2217
Tejun Heo2e60e022009-04-23 11:05:18 +09002218 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002219}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002220
2221/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002222 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2223 * @rq: the request to complete
2224 * @error: %0 for success, < %0 for error
2225 * @nr_bytes: number of bytes to complete @rq
2226 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002227 *
2228 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002229 * Identical to blk_end_bidi_request() except that queue lock is
2230 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002231 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002232 * Return:
2233 * %false - we are done with this request
2234 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002235 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002236static bool __blk_end_bidi_request(struct request *rq, int error,
2237 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002238{
Tejun Heo2e60e022009-04-23 11:05:18 +09002239 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2240 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002241
Tejun Heo2e60e022009-04-23 11:05:18 +09002242 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002243
Tejun Heo2e60e022009-04-23 11:05:18 +09002244 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002245}
2246
2247/**
2248 * blk_end_request - Helper function for drivers to complete the request.
2249 * @rq: the request being processed
2250 * @error: %0 for success, < %0 for error
2251 * @nr_bytes: number of bytes to complete
2252 *
2253 * Description:
2254 * Ends I/O on a number of bytes attached to @rq.
2255 * If @rq has leftover, sets it up for the next range of segments.
2256 *
2257 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002258 * %false - we are done with this request
2259 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002260 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002261bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002262{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002263 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002264}
Jens Axboe56ad1742009-07-28 22:11:24 +02002265EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002266
2267/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002268 * blk_end_request_all - Helper function for drives to finish the request.
2269 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002270 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002271 *
2272 * Description:
2273 * Completely finish @rq.
2274 */
2275void blk_end_request_all(struct request *rq, int error)
2276{
2277 bool pending;
2278 unsigned int bidi_bytes = 0;
2279
2280 if (unlikely(blk_bidi_rq(rq)))
2281 bidi_bytes = blk_rq_bytes(rq->next_rq);
2282
2283 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2284 BUG_ON(pending);
2285}
Jens Axboe56ad1742009-07-28 22:11:24 +02002286EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002287
2288/**
2289 * blk_end_request_cur - Helper function to finish the current request chunk.
2290 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002291 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002292 *
2293 * Description:
2294 * Complete the current consecutively mapped chunk from @rq.
2295 *
2296 * Return:
2297 * %false - we are done with this request
2298 * %true - still buffers pending for this request
2299 */
2300bool blk_end_request_cur(struct request *rq, int error)
2301{
2302 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2303}
Jens Axboe56ad1742009-07-28 22:11:24 +02002304EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002305
2306/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002307 * blk_end_request_err - Finish a request till the next failure boundary.
2308 * @rq: the request to finish till the next failure boundary for
2309 * @error: must be negative errno
2310 *
2311 * Description:
2312 * Complete @rq till the next failure boundary.
2313 *
2314 * Return:
2315 * %false - we are done with this request
2316 * %true - still buffers pending for this request
2317 */
2318bool blk_end_request_err(struct request *rq, int error)
2319{
2320 WARN_ON(error >= 0);
2321 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2322}
2323EXPORT_SYMBOL_GPL(blk_end_request_err);
2324
2325/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002326 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002327 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002328 * @error: %0 for success, < %0 for error
2329 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002330 *
2331 * Description:
2332 * Must be called with queue lock held unlike blk_end_request().
2333 *
2334 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002335 * %false - we are done with this request
2336 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002337 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002338bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002339{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002340 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002341}
Jens Axboe56ad1742009-07-28 22:11:24 +02002342EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002343
2344/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002345 * __blk_end_request_all - Helper function for drives to finish the request.
2346 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002347 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002348 *
2349 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002350 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002351 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002352void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002353{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002354 bool pending;
2355 unsigned int bidi_bytes = 0;
2356
2357 if (unlikely(blk_bidi_rq(rq)))
2358 bidi_bytes = blk_rq_bytes(rq->next_rq);
2359
2360 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2361 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002362}
Jens Axboe56ad1742009-07-28 22:11:24 +02002363EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002364
2365/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002366 * __blk_end_request_cur - Helper function to finish the current request chunk.
2367 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002368 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002369 *
2370 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002371 * Complete the current consecutively mapped chunk from @rq. Must
2372 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002373 *
2374 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002375 * %false - we are done with this request
2376 * %true - still buffers pending for this request
2377 */
2378bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002379{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002380 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002381}
Jens Axboe56ad1742009-07-28 22:11:24 +02002382EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002383
Tejun Heo80a761f2009-07-03 17:48:17 +09002384/**
2385 * __blk_end_request_err - Finish a request till the next failure boundary.
2386 * @rq: the request to finish till the next failure boundary for
2387 * @error: must be negative errno
2388 *
2389 * Description:
2390 * Complete @rq till the next failure boundary. Must be called
2391 * with queue lock held.
2392 *
2393 * Return:
2394 * %false - we are done with this request
2395 * %true - still buffers pending for this request
2396 */
2397bool __blk_end_request_err(struct request *rq, int error)
2398{
2399 WARN_ON(error >= 0);
2400 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2401}
2402EXPORT_SYMBOL_GPL(__blk_end_request_err);
2403
Jens Axboe86db1e22008-01-29 14:53:40 +01002404void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2405 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002407 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002408 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
David Woodhousefb2dce82008-08-05 18:01:53 +01002410 if (bio_has_data(bio)) {
2411 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002412 rq->buffer = bio_data(bio);
2413 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002414 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
NeilBrown66846572007-08-16 13:31:28 +02002417 if (bio->bi_bdev)
2418 rq->rq_disk = bio->bi_bdev->bd_disk;
2419}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002421#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2422/**
2423 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2424 * @rq: the request to be flushed
2425 *
2426 * Description:
2427 * Flush all pages in @rq.
2428 */
2429void rq_flush_dcache_pages(struct request *rq)
2430{
2431 struct req_iterator iter;
2432 struct bio_vec *bvec;
2433
2434 rq_for_each_segment(bvec, rq, iter)
2435 flush_dcache_page(bvec->bv_page);
2436}
2437EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2438#endif
2439
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002440/**
2441 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2442 * @q : the queue of the device being checked
2443 *
2444 * Description:
2445 * Check if underlying low-level drivers of a device are busy.
2446 * If the drivers want to export their busy state, they must set own
2447 * exporting function using blk_queue_lld_busy() first.
2448 *
2449 * Basically, this function is used only by request stacking drivers
2450 * to stop dispatching requests to underlying devices when underlying
2451 * devices are busy. This behavior helps more I/O merging on the queue
2452 * of the request stacking driver and prevents I/O throughput regression
2453 * on burst I/O load.
2454 *
2455 * Return:
2456 * 0 - Not busy (The request stacking driver should dispatch request)
2457 * 1 - Busy (The request stacking driver should stop dispatching request)
2458 */
2459int blk_lld_busy(struct request_queue *q)
2460{
2461 if (q->lld_busy_fn)
2462 return q->lld_busy_fn(q);
2463
2464 return 0;
2465}
2466EXPORT_SYMBOL_GPL(blk_lld_busy);
2467
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002468/**
2469 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2470 * @rq: the clone request to be cleaned up
2471 *
2472 * Description:
2473 * Free all bios in @rq for a cloned request.
2474 */
2475void blk_rq_unprep_clone(struct request *rq)
2476{
2477 struct bio *bio;
2478
2479 while ((bio = rq->bio) != NULL) {
2480 rq->bio = bio->bi_next;
2481
2482 bio_put(bio);
2483 }
2484}
2485EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2486
2487/*
2488 * Copy attributes of the original request to the clone request.
2489 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2490 */
2491static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2492{
2493 dst->cpu = src->cpu;
2494 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2495 dst->cmd_type = src->cmd_type;
2496 dst->__sector = blk_rq_pos(src);
2497 dst->__data_len = blk_rq_bytes(src);
2498 dst->nr_phys_segments = src->nr_phys_segments;
2499 dst->ioprio = src->ioprio;
2500 dst->extra_len = src->extra_len;
2501}
2502
2503/**
2504 * blk_rq_prep_clone - Helper function to setup clone request
2505 * @rq: the request to be setup
2506 * @rq_src: original request to be cloned
2507 * @bs: bio_set that bios for clone are allocated from
2508 * @gfp_mask: memory allocation mask for bio
2509 * @bio_ctr: setup function to be called for each clone bio.
2510 * Returns %0 for success, non %0 for failure.
2511 * @data: private data to be passed to @bio_ctr
2512 *
2513 * Description:
2514 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2515 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2516 * are not copied, and copying such parts is the caller's responsibility.
2517 * Also, pages which the original bios are pointing to are not copied
2518 * and the cloned bios just point same pages.
2519 * So cloned bios must be completed before original bios, which means
2520 * the caller must complete @rq before @rq_src.
2521 */
2522int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2523 struct bio_set *bs, gfp_t gfp_mask,
2524 int (*bio_ctr)(struct bio *, struct bio *, void *),
2525 void *data)
2526{
2527 struct bio *bio, *bio_src;
2528
2529 if (!bs)
2530 bs = fs_bio_set;
2531
2532 blk_rq_init(NULL, rq);
2533
2534 __rq_for_each_bio(bio_src, rq_src) {
2535 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2536 if (!bio)
2537 goto free_and_out;
2538
2539 __bio_clone(bio, bio_src);
2540
2541 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002542 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002543 goto free_and_out;
2544
2545 if (bio_ctr && bio_ctr(bio, bio_src, data))
2546 goto free_and_out;
2547
2548 if (rq->bio) {
2549 rq->biotail->bi_next = bio;
2550 rq->biotail = bio;
2551 } else
2552 rq->bio = rq->biotail = bio;
2553 }
2554
2555 __blk_rq_prep_clone(rq, rq_src);
2556
2557 return 0;
2558
2559free_and_out:
2560 if (bio)
2561 bio_free(bio, bs);
2562 blk_rq_unprep_clone(rq);
2563
2564 return -ENOMEM;
2565}
2566EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2567
Jens Axboe18887ad2008-07-28 13:08:45 +02002568int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569{
2570 return queue_work(kblockd_workqueue, work);
2571}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572EXPORT_SYMBOL(kblockd_schedule_work);
2573
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574int __init blk_dev_init(void)
2575{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002576 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2577 sizeof(((struct request *)0)->cmd_flags));
2578
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 kblockd_workqueue = create_workqueue("kblockd");
2580 if (!kblockd_workqueue)
2581 panic("Failed to create kblockd\n");
2582
2583 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002584 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
Jens Axboe8324aa92008-01-29 14:51:59 +01002586 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002587 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 return 0;
2590}