blob: 0520cc70458586db66c8cd280431bb267bcc2618 [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 Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Jens Axboe2056a782006-03-23 20:00:26 +010029#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080030#include <linux/fault-inject.h>
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010031#include <trace/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Jens Axboe8324aa92008-01-29 14:51:59 +010033#include "blk.h"
34
Ingo Molnar0bfc2452008-11-26 11:59:56 +010035DEFINE_TRACE(block_plug);
36DEFINE_TRACE(block_unplug_io);
37DEFINE_TRACE(block_unplug_timer);
38DEFINE_TRACE(block_getrq);
39DEFINE_TRACE(block_sleeprq);
40DEFINE_TRACE(block_rq_requeue);
41DEFINE_TRACE(block_bio_backmerge);
42DEFINE_TRACE(block_bio_frontmerge);
43DEFINE_TRACE(block_bio_queue);
44DEFINE_TRACE(block_rq_complete);
45DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */
46EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
47
Jens Axboe165125e2007-07-24 09:28:11 +020048static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * For the allocated request tables
52 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010053static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * For queue allocation
57 */
Jens Axboe6728cb02008-01-31 13:03:55 +010058struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * Controlling structure to kblockd
62 */
Jens Axboeff856ba2006-01-09 16:02:34 +010063static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jens Axboe26b82562008-01-29 13:54:41 +010065static void drive_stat_acct(struct request *rq, int new_io)
66{
Jens Axboe28f13702008-05-07 10:15:46 +020067 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010068 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090069 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010070
Jerome Marchand26308ea2009-03-27 10:31:51 +010071 if (!blk_fs_request(rq) || !blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010072 return;
73
Tejun Heo074a7ac2008-08-25 19:56:14 +090074 cpu = part_stat_lock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +020075 part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
Tejun Heoc9959052008-08-25 19:47:21 +090076
Jens Axboe28f13702008-05-07 10:15:46 +020077 if (!new_io)
Tejun Heo074a7ac2008-08-25 19:56:14 +090078 part_stat_inc(cpu, part, merges[rw]);
Jens Axboe28f13702008-05-07 10:15:46 +020079 else {
Tejun Heo074a7ac2008-08-25 19:56:14 +090080 part_round_stats(cpu, part);
81 part_inc_in_flight(part);
Jens Axboe26b82562008-01-29 13:54:41 +010082 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020083
Tejun Heo074a7ac2008-08-25 19:56:14 +090084 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010085}
86
Jens Axboe8324aa92008-01-29 14:51:59 +010087void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 int nr;
90
91 nr = q->nr_requests - (q->nr_requests / 8) + 1;
92 if (nr > q->nr_requests)
93 nr = q->nr_requests;
94 q->nr_congestion_on = nr;
95
96 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
97 if (nr < 1)
98 nr = 1;
99 q->nr_congestion_off = nr;
100}
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102/**
103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
104 * @bdev: device
105 *
106 * Locates the passed device's request queue and returns the address of its
107 * backing_dev_info
108 *
109 * Will return NULL if the request queue cannot be located.
110 */
111struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
112{
113 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200114 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116 if (q)
117 ret = &q->backing_dev_info;
118 return ret;
119}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120EXPORT_SYMBOL(blk_get_backing_dev_info);
121
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200122void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200124 memset(rq, 0, sizeof(*rq));
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700127 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200128 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100129 rq->q = q;
130 rq->sector = rq->hard_sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200133 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800134 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100136 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900137 rq->start_time = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200139EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
NeilBrown5bb23a62007-09-27 12:46:13 +0200141static void req_bio_endio(struct request *rq, struct bio *bio,
142 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100143{
Jens Axboe165125e2007-07-24 09:28:11 +0200144 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100145
NeilBrown5bb23a62007-09-27 12:46:13 +0200146 if (&q->bar_rq != rq) {
147 if (error)
148 clear_bit(BIO_UPTODATE, &bio->bi_flags);
149 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
150 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100151
NeilBrown5bb23a62007-09-27 12:46:13 +0200152 if (unlikely(nbytes > bio->bi_size)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100153 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -0700154 __func__, nbytes, bio->bi_size);
NeilBrown5bb23a62007-09-27 12:46:13 +0200155 nbytes = bio->bi_size;
156 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100157
Keith Mannthey08bafc02008-11-25 10:24:35 +0100158 if (unlikely(rq->cmd_flags & REQ_QUIET))
159 set_bit(BIO_QUIET, &bio->bi_flags);
160
NeilBrown5bb23a62007-09-27 12:46:13 +0200161 bio->bi_size -= nbytes;
162 bio->bi_sector += (nbytes >> 9);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200163
164 if (bio_integrity(bio))
165 bio_integrity_advance(bio, nbytes);
166
NeilBrown5bb23a62007-09-27 12:46:13 +0200167 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200168 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200169 } else {
170
171 /*
172 * Okay, this is the barrier request in progress, just
173 * record the error;
174 */
175 if (error && !q->orderr)
176 q->orderr = error;
177 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100178}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180void blk_dump_rq_flags(struct request *rq, char *msg)
181{
182 int bit;
183
Jens Axboe6728cb02008-01-31 13:03:55 +0100184 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200185 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
186 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
Jens Axboe6728cb02008-01-31 13:03:55 +0100188 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
189 (unsigned long long)rq->sector,
190 rq->nr_sectors,
191 rq->current_nr_sectors);
192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
193 rq->bio, rq->biotail,
194 rq->buffer, rq->data,
195 rq->data_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Jens Axboe4aff5e22006-08-10 08:44:47 +0200197 if (blk_pc_request(rq)) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100198 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200199 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 printk("%02x ", rq->cmd[bit]);
201 printk("\n");
202 }
203}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204EXPORT_SYMBOL(blk_dump_rq_flags);
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206/*
207 * "plug" the device if there are no outstanding requests: this will
208 * force the transfer to start only after we have put all the requests
209 * on the list.
210 *
211 * This is called with interrupts off and no requests on the queue and
212 * with the queue lock held.
213 */
Jens Axboe165125e2007-07-24 09:28:11 +0200214void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 WARN_ON(!irqs_disabled());
217
218 /*
219 * don't plug a stopped queue, it must be paired with blk_start_queue()
220 * which will restart the queueing
221 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200222 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 return;
224
Jens Axboee48ec692008-07-03 13:18:54 +0200225 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100227 trace_block_plug(q);
Jens Axboe2056a782006-03-23 20:00:26 +0100228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230EXPORT_SYMBOL(blk_plug_device);
231
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200232/**
233 * blk_plug_device_unlocked - plug a device without queue lock held
234 * @q: The &struct request_queue to plug
235 *
236 * Description:
237 * Like @blk_plug_device(), but grabs the queue lock and disables
238 * interrupts.
239 **/
240void blk_plug_device_unlocked(struct request_queue *q)
241{
242 unsigned long flags;
243
244 spin_lock_irqsave(q->queue_lock, flags);
245 blk_plug_device(q);
246 spin_unlock_irqrestore(q->queue_lock, flags);
247}
248EXPORT_SYMBOL(blk_plug_device_unlocked);
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250/*
251 * remove the queue from the plugged list, if present. called with
252 * queue lock held and interrupts disabled.
253 */
Jens Axboe165125e2007-07-24 09:28:11 +0200254int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 WARN_ON(!irqs_disabled());
257
Jens Axboee48ec692008-07-03 13:18:54 +0200258 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 return 0;
260
261 del_timer(&q->unplug_timer);
262 return 1;
263}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264EXPORT_SYMBOL(blk_remove_plug);
265
266/*
267 * remove the plug and let it rip..
268 */
Jens Axboe165125e2007-07-24 09:28:11 +0200269void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +0200271 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return;
Jens Axboea31a9732008-10-17 13:58:29 +0200273 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 return;
275
Jens Axboe22e2c502005-06-27 10:55:12 +0200276 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279/**
280 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +0200281 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 *
283 * Description:
284 * Linux uses plugging to build bigger requests queues before letting
285 * the device have at them. If a queue is plugged, the I/O scheduler
286 * is still adding and merging requests on the queue. Once the queue
287 * gets unplugged, the request_fn defined for the queue is invoked and
288 * transfers started.
289 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200290void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
Jens Axboedbaf2c02008-05-07 09:48:17 +0200292 if (blk_queue_plugged(q)) {
293 spin_lock_irq(q->queue_lock);
294 __generic_unplug_device(q);
295 spin_unlock_irq(q->queue_lock);
296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298EXPORT_SYMBOL(generic_unplug_device);
299
300static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
301 struct page *page)
302{
Jens Axboe165125e2007-07-24 09:28:11 +0200303 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500305 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Jens Axboe86db1e22008-01-29 14:53:40 +0100308void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Jens Axboe165125e2007-07-24 09:28:11 +0200310 struct request_queue *q =
311 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100313 trace_block_unplug_io(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 q->unplug_fn(q);
315}
316
Jens Axboe86db1e22008-01-29 14:53:40 +0100317void blk_unplug_timeout(unsigned long data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Jens Axboe165125e2007-07-24 09:28:11 +0200319 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100321 trace_block_unplug_timer(q);
Jens Axboe18887ad2008-07-28 13:08:45 +0200322 kblockd_schedule_work(q, &q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323}
324
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500325void blk_unplug(struct request_queue *q)
326{
327 /*
328 * devices don't necessarily have an ->unplug_fn defined
329 */
330 if (q->unplug_fn) {
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100331 trace_block_unplug_io(q);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500332 q->unplug_fn(q);
333 }
334}
335EXPORT_SYMBOL(blk_unplug);
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337/**
338 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200339 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 *
341 * Description:
342 * blk_start_queue() will clear the stop flag on the queue, and call
343 * the request_fn for the queue if it was in a stopped state when
344 * entered. Also see blk_stop_queue(). Queue lock must be held.
345 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200346void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200348 WARN_ON(!irqs_disabled());
349
Nick Piggin75ad23b2008-04-29 14:48:33 +0200350 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Tejun Heoa538cd02009-04-23 11:05:17 +0900351 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353EXPORT_SYMBOL(blk_start_queue);
354
355/**
356 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200357 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 *
359 * Description:
360 * The Linux block layer assumes that a block driver will consume all
361 * entries on the request queue when the request_fn strategy is called.
362 * Often this will not happen, because of hardware limitations (queue
363 * depth settings). If a device driver gets a 'queue full' response,
364 * or if it simply chooses not to queue more I/O at one point, it can
365 * call this function to prevent the request_fn from being called until
366 * the driver has signalled it's ready to go again. This happens by calling
367 * blk_start_queue() to restart queue operations. Queue lock must be held.
368 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200369void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
371 blk_remove_plug(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200372 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374EXPORT_SYMBOL(blk_stop_queue);
375
376/**
377 * blk_sync_queue - cancel any pending callbacks on a queue
378 * @q: the queue
379 *
380 * Description:
381 * The block layer may perform asynchronous callback activity
382 * on a queue, such as calling the unplug function after a timeout.
383 * A block device may call blk_sync_queue to ensure that any
384 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200385 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 * that its ->make_request_fn will not re-add plugging prior to calling
387 * this function.
388 *
389 */
390void blk_sync_queue(struct request_queue *q)
391{
392 del_timer_sync(&q->unplug_timer);
Jens Axboe70ed28b2008-11-19 14:38:39 +0100393 del_timer_sync(&q->timeout);
Cheng Renquan64d01dc2008-12-03 12:41:39 +0100394 cancel_work_sync(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396EXPORT_SYMBOL(blk_sync_queue);
397
398/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200399 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200401 *
402 * Description:
403 * See @blk_run_queue. This variant must be called with the queue lock
404 * held and interrupts disabled.
405 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 */
Nick Piggin75ad23b2008-04-29 14:48:33 +0200407void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +0200410
Tejun Heoa538cd02009-04-23 11:05:17 +0900411 if (unlikely(blk_queue_stopped(q)))
412 return;
413
414 if (elv_queue_empty(q))
415 return;
416
Jens Axboedac07ec2006-05-11 08:20:16 +0200417 /*
418 * Only recurse once to avoid overrunning the stack, let the unplug
419 * handling reinvoke the handler shortly if we already got there.
420 */
Tejun Heoa538cd02009-04-23 11:05:17 +0900421 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
422 q->request_fn(q);
423 queue_flag_clear(QUEUE_FLAG_REENTER, q);
424 } else {
425 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
426 kblockd_schedule_work(q, &q->unplug_work);
427 }
Nick Piggin75ad23b2008-04-29 14:48:33 +0200428}
429EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200430
Nick Piggin75ad23b2008-04-29 14:48:33 +0200431/**
432 * blk_run_queue - run a single device queue
433 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200434 *
435 * Description:
436 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900437 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200438 */
439void blk_run_queue(struct request_queue *q)
440{
441 unsigned long flags;
442
443 spin_lock_irqsave(q->queue_lock, flags);
444 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 spin_unlock_irqrestore(q->queue_lock, flags);
446}
447EXPORT_SYMBOL(blk_run_queue);
448
Jens Axboe165125e2007-07-24 09:28:11 +0200449void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500450{
451 kobject_put(&q->kobj);
452}
Al Viro483f4af2006-03-18 18:34:37 -0500453
Jens Axboe6728cb02008-01-31 13:03:55 +0100454void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500455{
Jens Axboee3335de2008-09-18 09:22:54 -0700456 /*
457 * We know we have process context here, so we can be a little
458 * cautious and ensure that pending block actions on this device
459 * are done before moving on. Going into this function, we should
460 * not have processes doing IO to this device.
461 */
462 blk_sync_queue(q);
463
Al Viro483f4af2006-03-18 18:34:37 -0500464 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200465 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500466 mutex_unlock(&q->sysfs_lock);
467
468 if (q->elevator)
469 elevator_exit(q->elevator);
470
471 blk_put_queue(q);
472}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473EXPORT_SYMBOL(blk_cleanup_queue);
474
Jens Axboe165125e2007-07-24 09:28:11 +0200475static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 struct request_list *rl = &q->rq;
478
Jens Axboe1faa16d2009-04-06 14:48:01 +0200479 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
480 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200481 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200482 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
483 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Christoph Lameter19460892005-06-23 00:08:19 -0700485 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
486 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 if (!rl->rq_pool)
489 return -ENOMEM;
490
491 return 0;
492}
493
Jens Axboe165125e2007-07-24 09:28:11 +0200494struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Christoph Lameter19460892005-06-23 00:08:19 -0700496 return blk_alloc_queue_node(gfp_mask, -1);
497}
498EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Jens Axboe165125e2007-07-24 09:28:11 +0200500struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700501{
Jens Axboe165125e2007-07-24 09:28:11 +0200502 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700503 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700504
Jens Axboe8324aa92008-01-29 14:51:59 +0100505 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700506 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (!q)
508 return NULL;
509
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700510 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
511 q->backing_dev_info.unplug_io_data = q;
512 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 init_timer(&q->unplug_timer);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700519 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
520 INIT_LIST_HEAD(&q->timeout_list);
Peter Zijlstra713ada92008-10-16 13:44:57 +0200521 INIT_WORK(&q->unplug_work, blk_unplug_work);
Al Viro483f4af2006-03-18 18:34:37 -0500522
Jens Axboe8324aa92008-01-29 14:51:59 +0100523 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Al Viro483f4af2006-03-18 18:34:37 -0500525 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700526 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 return q;
529}
Christoph Lameter19460892005-06-23 00:08:19 -0700530EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
532/**
533 * blk_init_queue - prepare a request queue for use with a block device
534 * @rfn: The function to be called to process requests that have been
535 * placed on the queue.
536 * @lock: Request queue spin lock
537 *
538 * Description:
539 * If a block device wishes to use the standard request handling procedures,
540 * which sorts requests and coalesces adjacent requests, then it must
541 * call blk_init_queue(). The function @rfn will be called when there
542 * are requests on the queue that need to be processed. If the device
543 * supports plugging, then @rfn may not be called immediately when requests
544 * are available on the queue, but may be called at some time later instead.
545 * Plugged queues are generally unplugged when a buffer belonging to one
546 * of the requests on the queue is needed, or due to memory pressure.
547 *
548 * @rfn is not required, or even expected, to remove all requests off the
549 * queue, but only as many as it can handle at a time. If it does leave
550 * requests on the queue, it is responsible for arranging that the requests
551 * get dealt with eventually.
552 *
553 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200554 * request queue; this lock will be taken also from interrupt context, so irq
555 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200557 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 * it didn't succeed.
559 *
560 * Note:
561 * blk_init_queue() must be paired with a blk_cleanup_queue() call
562 * when the block device is deactivated (such as at module unload).
563 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700564
Jens Axboe165125e2007-07-24 09:28:11 +0200565struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
Christoph Lameter19460892005-06-23 00:08:19 -0700567 return blk_init_queue_node(rfn, lock, -1);
568}
569EXPORT_SYMBOL(blk_init_queue);
570
Jens Axboe165125e2007-07-24 09:28:11 +0200571struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700572blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
573{
Jens Axboe165125e2007-07-24 09:28:11 +0200574 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 if (!q)
577 return NULL;
578
Christoph Lameter19460892005-06-23 00:08:19 -0700579 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -0500580 if (blk_init_free_list(q)) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100581 kmem_cache_free(blk_requestq_cachep, q);
Al Viro8669aaf2006-03-18 13:50:00 -0500582 return NULL;
583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
152587d2005-04-12 16:22:06 -0500585 /*
586 * if caller didn't supply a lock, they get per-queue locking with
587 * our embedded lock
588 */
Neil Browne7e72bf2008-05-14 16:05:54 -0700589 if (!lock)
152587d2005-04-12 16:22:06 -0500590 lock = &q->__queue_lock;
152587d2005-04-12 16:22:06 -0500591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 q->prep_rq_fn = NULL;
594 q->unplug_fn = generic_unplug_device;
Jens Axboebc58ba92009-01-23 10:54:44 +0100595 q->queue_flags = QUEUE_FLAG_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 q->queue_lock = lock;
597
Jens Axboef3b144a2009-03-06 08:48:33 +0100598 /*
599 * This also sets hw/phys segments, boundary and size
600 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Alan Stern44ec9542007-02-20 11:01:57 -0500603 q->sg_reserved_size = INT_MAX;
604
FUJITA Tomonoriabf54392008-08-16 14:10:05 +0900605 blk_set_cmd_filter_defaults(&q->cmd_filter);
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
608 * all done
609 */
610 if (!elevator_init(q, NULL)) {
611 blk_queue_congestion_threshold(q);
612 return q;
613 }
614
Al Viro8669aaf2006-03-18 13:50:00 -0500615 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return NULL;
617}
Christoph Lameter19460892005-06-23 00:08:19 -0700618EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Jens Axboe165125e2007-07-24 09:28:11 +0200620int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700622 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500623 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return 0;
625 }
626
627 return 1;
628}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Jens Axboe165125e2007-07-24 09:28:11 +0200630static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200632 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200633 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 mempool_free(rq, q->rq.rq_pool);
635}
636
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200637static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200638blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
641
642 if (!rq)
643 return NULL;
644
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200645 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200646
Jerome Marchand42dad762009-04-22 14:01:49 +0200647 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Tejun Heocb98fc82005-10-28 08:29:39 +0200649 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200650 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200651 mempool_free(rq, q->rq.rq_pool);
652 return NULL;
653 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200654 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Tejun Heocb98fc82005-10-28 08:29:39 +0200657 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660/*
661 * ioc_batching returns true if the ioc is a valid batching request and
662 * should be given priority access to a request.
663 */
Jens Axboe165125e2007-07-24 09:28:11 +0200664static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 if (!ioc)
667 return 0;
668
669 /*
670 * Make sure the process is able to allocate at least 1 request
671 * even if the batch times out, otherwise we could theoretically
672 * lose wakeups.
673 */
674 return ioc->nr_batch_requests == q->nr_batching ||
675 (ioc->nr_batch_requests > 0
676 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
677}
678
679/*
680 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
681 * will cause the process to be a "batcher" on all queues in the system. This
682 * is the behaviour we want though - once it gets a wakeup it should be given
683 * a nice run.
684 */
Jens Axboe165125e2007-07-24 09:28:11 +0200685static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 if (!ioc || ioc_batching(q, ioc))
688 return;
689
690 ioc->nr_batch_requests = q->nr_batching;
691 ioc->last_waited = jiffies;
692}
693
Jens Axboe1faa16d2009-04-06 14:48:01 +0200694static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct request_list *rl = &q->rq;
697
Jens Axboe1faa16d2009-04-06 14:48:01 +0200698 if (rl->count[sync] < queue_congestion_off_threshold(q))
699 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Jens Axboe1faa16d2009-04-06 14:48:01 +0200701 if (rl->count[sync] + 1 <= q->nr_requests) {
702 if (waitqueue_active(&rl->wait[sync]))
703 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Jens Axboe1faa16d2009-04-06 14:48:01 +0200705 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707}
708
709/*
710 * A request has just been released. Account for it, update the full and
711 * congestion status, wake up any waiters. Called under q->queue_lock.
712 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200713static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
715 struct request_list *rl = &q->rq;
716
Jens Axboe1faa16d2009-04-06 14:48:01 +0200717 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200718 if (priv)
719 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Jens Axboe1faa16d2009-04-06 14:48:01 +0200721 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Jens Axboe1faa16d2009-04-06 14:48:01 +0200723 if (unlikely(rl->starved[sync ^ 1]))
724 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727/*
Nick Piggind6344532005-06-28 20:45:14 -0700728 * Get a free request, queue_lock must be held.
729 * Returns NULL on failure, with queue_lock held.
730 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
Jens Axboe165125e2007-07-24 09:28:11 +0200732static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100733 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
735 struct request *rq = NULL;
736 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100737 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200738 const bool is_sync = rw_is_sync(rw_flags) != 0;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100739 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Jens Axboe7749a8d2006-12-13 13:02:26 +0100741 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100742 if (may_queue == ELV_MQUEUE_NO)
743 goto rq_starved;
744
Jens Axboe1faa16d2009-04-06 14:48:01 +0200745 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
746 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200747 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100748 /*
749 * The queue will fill after this allocation, so set
750 * it as full, and mark this process as "batching".
751 * This process will be allowed to complete a batch of
752 * requests, others will be blocked.
753 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200754 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100755 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200756 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100757 } else {
758 if (may_queue != ELV_MQUEUE_MUST
759 && !ioc_batching(q, ioc)) {
760 /*
761 * The queue is full and the allocating
762 * process is not a "batcher", and not
763 * exempted by the IO scheduler
764 */
765 goto out;
766 }
767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200769 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 }
771
Jens Axboe082cf692005-06-28 16:35:11 +0200772 /*
773 * Only allow batching queuers to allocate up to 50% over the defined
774 * limit of requests, otherwise we could have thousands of requests
775 * allocated with any setting of ->nr_requests
776 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200777 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200778 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100779
Jens Axboe1faa16d2009-04-06 14:48:01 +0200780 rl->count[is_sync]++;
781 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200782
Jens Axboe64521d12005-10-28 08:30:39 +0200783 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200784 if (priv)
785 rl->elvpriv++;
786
Jerome Marchand42dad762009-04-22 14:01:49 +0200787 if (blk_queue_io_stat(q))
788 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 spin_unlock_irq(q->queue_lock);
790
Jens Axboe7749a8d2006-12-13 13:02:26 +0100791 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100792 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 /*
794 * Allocation failed presumably due to memory. Undo anything
795 * we might have messed up.
796 *
797 * Allocating task should really be put onto the front of the
798 * wait queue, but this is pretty rare.
799 */
800 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200801 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
803 /*
804 * in the very unlikely event that allocation failed and no
805 * requests for this direction was pending, mark us starved
806 * so that freeing of a request in the other direction will
807 * notice us. another possible fix would be to split the
808 * rq mempool into READ and WRITE
809 */
810rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200811 if (unlikely(rl->count[is_sync] == 0))
812 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 goto out;
815 }
816
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100817 /*
818 * ioc may be NULL here, and ioc_batching will be false. That's
819 * OK, if the queue is under the request limit then requests need
820 * not count toward the nr_batch_requests limit. There will always
821 * be some limit enforced by BLK_BATCH_TIME.
822 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 if (ioc_batching(q, ioc))
824 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100825
Jens Axboe1faa16d2009-04-06 14:48:01 +0200826 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return rq;
829}
830
831/*
832 * No available requests for this queue, unplug the device and wait for some
833 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -0700834 *
835 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 */
Jens Axboe165125e2007-07-24 09:28:11 +0200837static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200838 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200840 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct request *rq;
842
Jens Axboe7749a8d2006-12-13 13:02:26 +0100843 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700844 while (!rq) {
845 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200846 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 struct request_list *rl = &q->rq;
848
Jens Axboe1faa16d2009-04-06 14:48:01 +0200849 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 TASK_UNINTERRUPTIBLE);
851
Jens Axboe1faa16d2009-04-06 14:48:01 +0200852 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200854 __generic_unplug_device(q);
855 spin_unlock_irq(q->queue_lock);
856 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200858 /*
859 * After sleeping, we become a "batching" process and
860 * will be able to allocate at least one request, and
861 * up to a big batch of them for a small period time.
862 * See ioc_batching, ioc_set_batching
863 */
864 ioc = current_io_context(GFP_NOIO, q->node);
865 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100866
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200867 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200868 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200869
870 rq = get_request(q, rw_flags, bio, GFP_NOIO);
871 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 return rq;
874}
875
Jens Axboe165125e2007-07-24 09:28:11 +0200876struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
878 struct request *rq;
879
880 BUG_ON(rw != READ && rw != WRITE);
881
Nick Piggind6344532005-06-28 20:45:14 -0700882 spin_lock_irq(q->queue_lock);
883 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200884 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700885 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200886 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700887 if (!rq)
888 spin_unlock_irq(q->queue_lock);
889 }
890 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 return rq;
893}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894EXPORT_SYMBOL(blk_get_request);
895
896/**
897 * blk_requeue_request - put a request back on queue
898 * @q: request queue where request should be inserted
899 * @rq: request to be inserted
900 *
901 * Description:
902 * Drivers often keep queueing requests until the hardware cannot accept
903 * more, when that condition happens we need to put the request back
904 * on the queue. Must be called with queue lock held.
905 */
Jens Axboe165125e2007-07-24 09:28:11 +0200906void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700908 blk_delete_timer(rq);
909 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100910 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (blk_rq_tagged(rq))
913 blk_queue_end_tag(q, rq);
914
915 elv_requeue_request(q, rq);
916}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917EXPORT_SYMBOL(blk_requeue_request);
918
919/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200920 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 * @q: request queue where request should be inserted
922 * @rq: request to be inserted
923 * @at_head: insert request at head or tail of queue
924 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 *
926 * Description:
927 * Many block devices need to execute commands asynchronously, so they don't
928 * block the whole kernel from preemption during request execution. This is
929 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200930 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
931 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 *
933 * We have the option of inserting the head or the tail of the queue.
934 * Typically we use the tail for new ioctls and so forth. We use the head
935 * of the queue for things like a QUEUE_FULL message from a device, or a
936 * host that is unable to accept a particular command.
937 */
Jens Axboe165125e2007-07-24 09:28:11 +0200938void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500939 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Tejun Heo 867d1192005-04-24 02:06:05 -0500941 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 unsigned long flags;
943
944 /*
945 * tell I/O scheduler that this isn't a regular read/write (ie it
946 * must not attempt merges on this) and that it acts as a soft
947 * barrier
948 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200949 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951 rq->special = data;
952
953 spin_lock_irqsave(q->queue_lock, flags);
954
955 /*
956 * If command is tagged, release the tag
957 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500958 if (blk_rq_tagged(rq))
959 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200961 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -0500962 __elv_add_request(q, rq, where, 0);
Tejun Heoa7f55792009-04-23 11:05:17 +0900963 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 spin_unlock_irqrestore(q->queue_lock, flags);
965}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966EXPORT_SYMBOL(blk_insert_request);
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968/*
969 * add-request adds a request to the linked list.
970 * queue lock is held and interrupts disabled, as we muck with the
971 * request queue list.
972 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100973static inline void add_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
Jerome Marchandb238b3d2007-10-23 15:05:46 +0200975 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 /*
978 * elevator indicated where it wants this request to be
979 * inserted at elevator_merge time
980 */
981 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
982}
Jens Axboe6728cb02008-01-31 13:03:55 +0100983
Tejun Heo074a7ac2008-08-25 19:56:14 +0900984static void part_round_stats_single(int cpu, struct hd_struct *part,
985 unsigned long now)
986{
987 if (now == part->stamp)
988 return;
989
990 if (part->in_flight) {
991 __part_stat_add(cpu, part, time_in_queue,
992 part->in_flight * (now - part->stamp));
993 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
994 }
995 part->stamp = now;
996}
997
998/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +0200999 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1000 * @cpu: cpu number for stats access
1001 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 *
1003 * The average IO queue length and utilisation statistics are maintained
1004 * by observing the current state of the queue length and the amount of
1005 * time it has been in this state for.
1006 *
1007 * Normally, that accounting is done on IO completion, but that can result
1008 * in more than a second's worth of IO being accounted for within any one
1009 * second, leading to >100% utilisation. To deal with that, we call this
1010 * function to do a round-off before returning the results when reading
1011 * /proc/diskstats. This accounts immediately for all queue usage up to
1012 * the current jiffies and restarts the counters again.
1013 */
Tejun Heoc9959052008-08-25 19:47:21 +09001014void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001015{
1016 unsigned long now = jiffies;
1017
Tejun Heo074a7ac2008-08-25 19:56:14 +09001018 if (part->partno)
1019 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1020 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001021}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001022EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576af2008-02-08 11:04:35 +01001023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024/*
1025 * queue lock must be held
1026 */
Jens Axboe165125e2007-07-24 09:28:11 +02001027void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (unlikely(!q))
1030 return;
1031 if (unlikely(--req->ref_count))
1032 return;
1033
Tejun Heo8922e162005-10-20 16:23:44 +02001034 elv_completed_request(q, req);
1035
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001036 /* this is a bio leak */
1037 WARN_ON(req->bio != NULL);
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 /*
1040 * Request may not have originated from ll_rw_blk. if not,
1041 * it didn't come out of our reserved rq pools
1042 */
Jens Axboe49171e52006-08-10 08:59:11 +02001043 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001044 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001045 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001048 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
1050 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001051 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
1053}
Mike Christie6e39b692005-11-11 05:30:24 -06001054EXPORT_SYMBOL_GPL(__blk_put_request);
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056void blk_put_request(struct request *req)
1057{
Tejun Heo8922e162005-10-20 16:23:44 +02001058 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001059 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001061 spin_lock_irqsave(q->queue_lock, flags);
1062 __blk_put_request(q, req);
1063 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065EXPORT_SYMBOL(blk_put_request);
1066
Jens Axboe86db1e22008-01-29 14:53:40 +01001067void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001068{
Jens Axboec7c22e42008-09-13 20:26:01 +02001069 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001070 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001071
1072 /*
1073 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1074 */
Mike Christie6000a362008-08-19 18:45:30 -05001075 if (bio_rw_ahead(bio))
1076 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1077 REQ_FAILFAST_DRIVER);
1078 if (bio_failfast_dev(bio))
1079 req->cmd_flags |= REQ_FAILFAST_DEV;
1080 if (bio_failfast_transport(bio))
1081 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1082 if (bio_failfast_driver(bio))
1083 req->cmd_flags |= REQ_FAILFAST_DRIVER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001084
David Woodhousefb2dce82008-08-05 18:01:53 +01001085 if (unlikely(bio_discard(bio))) {
David Woodhousee17fc0a2008-08-09 16:42:20 +01001086 req->cmd_flags |= REQ_DISCARD;
1087 if (bio_barrier(bio))
1088 req->cmd_flags |= REQ_SOFTBARRIER;
David Woodhousefb2dce82008-08-05 18:01:53 +01001089 req->q->prepare_discard_fn(req->q, req);
David Woodhousee17fc0a2008-08-09 16:42:20 +01001090 } else if (unlikely(bio_barrier(bio)))
Tejun Heoe4025f62009-04-23 11:05:17 +09001091 req->cmd_flags |= REQ_HARDBARRIER;
Tejun Heo52d9e672006-01-06 09:49:58 +01001092
Jens Axboeb31dc662006-06-13 08:26:10 +02001093 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02001094 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02001095 if (bio_rw_meta(bio))
1096 req->cmd_flags |= REQ_RW_META;
Jens Axboeaeb6faf2009-04-06 14:48:07 +02001097 if (bio_noidle(bio))
1098 req->cmd_flags |= REQ_NOIDLE;
Jens Axboeb31dc662006-06-13 08:26:10 +02001099
Tejun Heo52d9e672006-01-06 09:49:58 +01001100 req->errors = 0;
1101 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001102 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001103 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001104}
1105
Jens Axboe644b2d92009-04-06 14:48:06 +02001106/*
1107 * Only disabling plugging for non-rotational devices if it does tagging
1108 * as well, otherwise we do need the proper merging
1109 */
1110static inline bool queue_should_plug(struct request_queue *q)
1111{
1112 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1113}
1114
Jens Axboe165125e2007-07-24 09:28:11 +02001115static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
Nick Piggin450991b2005-06-28 20:45:13 -07001117 struct request *req;
Tejun Heoa7384672008-11-28 13:32:03 +09001118 int el_ret, nr_sectors;
Jens Axboe51da90f2006-07-18 04:14:45 +02001119 const unsigned short prio = bio_prio(bio);
1120 const int sync = bio_sync(bio);
Jens Axboe213d9412009-01-06 09:16:05 +01001121 const int unplug = bio_unplug(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01001122 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126 /*
1127 * low level driver can indicate that it wants pages above a
1128 * certain limit bounced to low memory (ie for highmem, or even
1129 * ISA dma in theory)
1130 */
1131 blk_queue_bounce(q, &bio);
1132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 spin_lock_irq(q->queue_lock);
1134
Tejun Heoa7384672008-11-28 13:32:03 +09001135 if (unlikely(bio_barrier(bio)) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 goto get_rq;
1137
1138 el_ret = elv_merge(q, &req, bio);
1139 switch (el_ret) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001140 case ELEVATOR_BACK_MERGE:
1141 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Jens Axboe6728cb02008-01-31 13:03:55 +01001143 if (!ll_back_merge_fn(q, req, bio))
1144 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001146 trace_block_bio_backmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001147
Jens Axboe6728cb02008-01-31 13:03:55 +01001148 req->biotail->bi_next = bio;
1149 req->biotail = bio;
1150 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1151 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001152 if (!blk_rq_cpu_valid(req))
1153 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001154 drive_stat_acct(req, 0);
1155 if (!attempt_back_merge(q, req))
1156 elv_merged_request(q, req, el_ret);
1157 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Jens Axboe6728cb02008-01-31 13:03:55 +01001159 case ELEVATOR_FRONT_MERGE:
1160 BUG_ON(!rq_mergeable(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Jens Axboe6728cb02008-01-31 13:03:55 +01001162 if (!ll_front_merge_fn(q, req, bio))
1163 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001165 trace_block_bio_frontmerge(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001166
Jens Axboe6728cb02008-01-31 13:03:55 +01001167 bio->bi_next = req->bio;
1168 req->bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Jens Axboe6728cb02008-01-31 13:03:55 +01001170 /*
1171 * may not be valid. if the low level driver said
1172 * it didn't need a bounce buffer then it better
1173 * not touch req->buffer either...
1174 */
1175 req->buffer = bio_data(bio);
1176 req->current_nr_sectors = bio_cur_sectors(bio);
1177 req->hard_cur_sectors = req->current_nr_sectors;
1178 req->sector = req->hard_sector = bio->bi_sector;
1179 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1180 req->ioprio = ioprio_best(req->ioprio, prio);
Jens Axboeab780f12008-08-26 10:25:02 +02001181 if (!blk_rq_cpu_valid(req))
1182 req->cpu = bio->bi_comp_cpu;
Jens Axboe6728cb02008-01-31 13:03:55 +01001183 drive_stat_acct(req, 0);
1184 if (!attempt_front_merge(q, req))
1185 elv_merged_request(q, req, el_ret);
1186 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Jens Axboe6728cb02008-01-31 13:03:55 +01001188 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1189 default:
1190 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
1192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001194 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001195 * This sync check and mask will be re-done in init_request_from_bio(),
1196 * but we need to set it earlier to expose the sync flag to the
1197 * rq allocator and io schedulers.
1198 */
1199 rw_flags = bio_data_dir(bio);
1200 if (sync)
1201 rw_flags |= REQ_RW_SYNC;
1202
1203 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001204 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001205 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001206 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001207 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001208
Nick Piggin450991b2005-06-28 20:45:13 -07001209 /*
1210 * After dropping the lock and possibly sleeping here, our request
1211 * may now be mergeable after it had proven unmergeable (above).
1212 * We don't worry about that case for efficiency. It won't happen
1213 * often, and the elevators are able to handle it.
1214 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001215 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Nick Piggin450991b2005-06-28 20:45:13 -07001217 spin_lock_irq(q->queue_lock);
Jens Axboec7c22e42008-09-13 20:26:01 +02001218 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1219 bio_flagged(bio, BIO_CPU_AFFINE))
1220 req->cpu = blk_cpu_to_group(smp_processor_id());
Jens Axboe644b2d92009-04-06 14:48:06 +02001221 if (queue_should_plug(q) && elv_queue_empty(q))
Nick Piggin450991b2005-06-28 20:45:13 -07001222 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 add_request(q, req);
1224out:
Jens Axboe644b2d92009-04-06 14:48:06 +02001225 if (unplug || !queue_should_plug(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 __generic_unplug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 spin_unlock_irq(q->queue_lock);
1228 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230
1231/*
1232 * If bio->bi_dev is a partition, remap the location
1233 */
1234static inline void blk_partition_remap(struct bio *bio)
1235{
1236 struct block_device *bdev = bio->bi_bdev;
1237
Jens Axboebf2de6f2007-09-27 13:01:25 +02001238 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 bio->bi_sector += p->start_sect;
1242 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001243
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001244 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001245 bdev->bd_dev, bio->bi_sector,
1246 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248}
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250static void handle_bad_sector(struct bio *bio)
1251{
1252 char b[BDEVNAME_SIZE];
1253
1254 printk(KERN_INFO "attempt to access beyond end of device\n");
1255 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1256 bdevname(bio->bi_bdev, b),
1257 bio->bi_rw,
1258 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1259 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1260
1261 set_bit(BIO_EOF, &bio->bi_flags);
1262}
1263
Akinobu Mitac17bb492006-12-08 02:39:46 -08001264#ifdef CONFIG_FAIL_MAKE_REQUEST
1265
1266static DECLARE_FAULT_ATTR(fail_make_request);
1267
1268static int __init setup_fail_make_request(char *str)
1269{
1270 return setup_fault_attr(&fail_make_request, str);
1271}
1272__setup("fail_make_request=", setup_fail_make_request);
1273
1274static int should_fail_request(struct bio *bio)
1275{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001276 struct hd_struct *part = bio->bi_bdev->bd_part;
1277
1278 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001279 return should_fail(&fail_make_request, bio->bi_size);
1280
1281 return 0;
1282}
1283
1284static int __init fail_make_request_debugfs(void)
1285{
1286 return init_fault_attr_dentries(&fail_make_request,
1287 "fail_make_request");
1288}
1289
1290late_initcall(fail_make_request_debugfs);
1291
1292#else /* CONFIG_FAIL_MAKE_REQUEST */
1293
1294static inline int should_fail_request(struct bio *bio)
1295{
1296 return 0;
1297}
1298
1299#endif /* CONFIG_FAIL_MAKE_REQUEST */
1300
Jens Axboec07e2b42007-07-18 13:27:58 +02001301/*
1302 * Check whether this bio extends beyond the end of the device.
1303 */
1304static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1305{
1306 sector_t maxsector;
1307
1308 if (!nr_sectors)
1309 return 0;
1310
1311 /* Test device or partition size, when known. */
1312 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1313 if (maxsector) {
1314 sector_t sector = bio->bi_sector;
1315
1316 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1317 /*
1318 * This may well happen - the kernel calls bread()
1319 * without checking the size of the device, e.g., when
1320 * mounting a device.
1321 */
1322 handle_bad_sector(bio);
1323 return 1;
1324 }
1325 }
1326
1327 return 0;
1328}
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001331 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 * @bio: The bio describing the location in memory and on the device.
1333 *
1334 * generic_make_request() is used to make I/O requests of block
1335 * devices. It is passed a &struct bio, which describes the I/O that needs
1336 * to be done.
1337 *
1338 * generic_make_request() does not return any status. The
1339 * success/failure status of the request, along with notification of
1340 * completion, is delivered asynchronously through the bio->bi_end_io
1341 * function described (one day) else where.
1342 *
1343 * The caller of generic_make_request must make sure that bi_io_vec
1344 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1345 * set to describe the device address, and the
1346 * bi_end_io and optionally bi_private are set to describe how
1347 * completion notification should be signaled.
1348 *
1349 * generic_make_request and the drivers it calls may use bi_next if this
1350 * bio happens to be merged with someone else, and may change bi_dev and
1351 * bi_sector for remaps as it sees fit. So the values of these fields
1352 * should NOT be depended on after the call to generic_make_request.
1353 */
Neil Brownd89d8792007-05-01 09:53:42 +02001354static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
Jens Axboe165125e2007-07-24 09:28:11 +02001356 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001357 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001359 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001360 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Jens Axboec07e2b42007-07-18 13:27:58 +02001364 if (bio_check_eod(bio, nr_sectors))
1365 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 /*
1368 * Resolve the mapping until finished. (drivers are
1369 * still free to implement/resolve their own stacking
1370 * by explicitly returning 0)
1371 *
1372 * NOTE: we don't repeat the blk_size check for each new device.
1373 * Stacking drivers are expected to know what they are doing.
1374 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001375 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001376 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 do {
1378 char b[BDEVNAME_SIZE];
1379
1380 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001381 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 printk(KERN_ERR
1383 "generic_make_request: Trying to access "
1384 "nonexistent block-device %s (%Lu)\n",
1385 bdevname(bio->bi_bdev, b),
1386 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001387 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 }
1389
Jens Axboe4fa253f2007-07-18 13:13:10 +02001390 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001391 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 bdevname(bio->bi_bdev, b),
1393 bio_sectors(bio),
1394 q->max_hw_sectors);
1395 goto end_io;
1396 }
1397
Nick Pigginfde6ad22005-06-23 00:08:53 -07001398 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 goto end_io;
1400
Akinobu Mitac17bb492006-12-08 02:39:46 -08001401 if (should_fail_request(bio))
1402 goto end_io;
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 /*
1405 * If this device has partitions, remap block n
1406 * of partition p to block n+start(p) of the disk.
1407 */
1408 blk_partition_remap(bio);
1409
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001410 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1411 goto end_io;
1412
NeilBrown5ddfe962006-10-30 22:07:21 -08001413 if (old_sector != -1)
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001414 trace_block_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08001415 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001416
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001417 trace_block_bio_queue(q, bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001418
NeilBrown5ddfe962006-10-30 22:07:21 -08001419 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001420 old_dev = bio->bi_bdev->bd_dev;
1421
Jens Axboec07e2b42007-07-18 13:27:58 +02001422 if (bio_check_eod(bio, nr_sectors))
1423 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001424
1425 if (bio_discard(bio) && !q->prepare_discard_fn) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001426 err = -EOPNOTSUPP;
1427 goto end_io;
1428 }
Jens Axboecec07072009-01-13 15:28:32 +01001429 if (bio_barrier(bio) && bio_has_data(bio) &&
1430 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1431 err = -EOPNOTSUPP;
1432 goto end_io;
1433 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 ret = q->make_request_fn(q, bio);
1436 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001437
1438 return;
1439
1440end_io:
1441 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
Neil Brownd89d8792007-05-01 09:53:42 +02001444/*
1445 * We only want one ->make_request_fn to be active at a time,
1446 * else stack usage with stacked devices could be a problem.
1447 * So use current->bio_{list,tail} to keep a list of requests
1448 * submited by a make_request_fn function.
1449 * current->bio_tail is also used as a flag to say if
1450 * generic_make_request is currently active in this task or not.
1451 * If it is NULL, then no make_request is active. If it is non-NULL,
1452 * then a make_request is active, and new requests should be added
1453 * at the tail
1454 */
1455void generic_make_request(struct bio *bio)
1456{
1457 if (current->bio_tail) {
1458 /* make_request is active */
1459 *(current->bio_tail) = bio;
1460 bio->bi_next = NULL;
1461 current->bio_tail = &bio->bi_next;
1462 return;
1463 }
1464 /* following loop may be a bit non-obvious, and so deserves some
1465 * explanation.
1466 * Before entering the loop, bio->bi_next is NULL (as all callers
1467 * ensure that) so we have a list with a single bio.
1468 * We pretend that we have just taken it off a longer list, so
1469 * we assign bio_list to the next (which is NULL) and bio_tail
1470 * to &bio_list, thus initialising the bio_list of new bios to be
1471 * added. __generic_make_request may indeed add some more bios
1472 * through a recursive call to generic_make_request. If it
1473 * did, we find a non-NULL value in bio_list and re-enter the loop
1474 * from the top. In this case we really did just take the bio
1475 * of the top of the list (no pretending) and so fixup bio_list and
1476 * bio_tail or bi_next, and call into __generic_make_request again.
1477 *
1478 * The loop was structured like this to make only one call to
1479 * __generic_make_request (which is important as it is large and
1480 * inlined) and to keep the structure simple.
1481 */
1482 BUG_ON(bio->bi_next);
1483 do {
1484 current->bio_list = bio->bi_next;
1485 if (bio->bi_next == NULL)
1486 current->bio_tail = &current->bio_list;
1487 else
1488 bio->bi_next = NULL;
1489 __generic_make_request(bio);
1490 bio = current->bio_list;
1491 } while (bio);
1492 current->bio_tail = NULL; /* deactivate */
1493}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494EXPORT_SYMBOL(generic_make_request);
1495
1496/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001497 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1499 * @bio: The &struct bio which describes the I/O
1500 *
1501 * submit_bio() is very similar in purpose to generic_make_request(), and
1502 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001503 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 *
1505 */
1506void submit_bio(int rw, struct bio *bio)
1507{
1508 int count = bio_sectors(bio);
1509
Jens Axboe22e2c502005-06-27 10:55:12 +02001510 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Jens Axboebf2de6f2007-09-27 13:01:25 +02001512 /*
1513 * If it's a regular read/write or a barrier with data attached,
1514 * go through the normal accounting stuff before submission.
1515 */
Jens Axboea9c701e2008-08-08 11:04:44 +02001516 if (bio_has_data(bio)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001517 if (rw & WRITE) {
1518 count_vm_events(PGPGOUT, count);
1519 } else {
1520 task_io_account_read(bio->bi_size);
1521 count_vm_events(PGPGIN, count);
1522 }
1523
1524 if (unlikely(block_dump)) {
1525 char b[BDEVNAME_SIZE];
1526 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001527 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001528 (rw & WRITE) ? "WRITE" : "READ",
1529 (unsigned long long)bio->bi_sector,
Jens Axboe6728cb02008-01-31 13:03:55 +01001530 bdevname(bio->bi_bdev, b));
Jens Axboebf2de6f2007-09-27 13:01:25 +02001531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 }
1533
1534 generic_make_request(bio);
1535}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536EXPORT_SYMBOL(submit_bio);
1537
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001538/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001539 * blk_rq_check_limits - Helper function to check a request for the queue limit
1540 * @q: the queue
1541 * @rq: the request being checked
1542 *
1543 * Description:
1544 * @rq may have been made based on weaker limitations of upper-level queues
1545 * in request stacking drivers, and it may violate the limitation of @q.
1546 * Since the block layer and the underlying device driver trust @rq
1547 * after it is inserted to @q, it should be checked against @q before
1548 * the insertion using this generic function.
1549 *
1550 * This function should also be useful for request stacking drivers
1551 * in some cases below, so export this fuction.
1552 * Request stacking drivers like request-based dm may change the queue
1553 * limits while requests are in the queue (e.g. dm's table swapping).
1554 * Such request stacking drivers should check those requests agaist
1555 * the new queue limits again when they dispatch those requests,
1556 * although such checkings are also done against the old queue limits
1557 * when submitting requests.
1558 */
1559int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1560{
1561 if (rq->nr_sectors > q->max_sectors ||
1562 rq->data_len > q->max_hw_sectors << 9) {
1563 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1564 return -EIO;
1565 }
1566
1567 /*
1568 * queue's settings related to segment counting like q->bounce_pfn
1569 * may differ from that of other stacking queues.
1570 * Recalculate it to check the request correctly on this queue's
1571 * limitation.
1572 */
1573 blk_recalc_rq_segments(rq);
1574 if (rq->nr_phys_segments > q->max_phys_segments ||
1575 rq->nr_phys_segments > q->max_hw_segments) {
1576 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1577 return -EIO;
1578 }
1579
1580 return 0;
1581}
1582EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1583
1584/**
1585 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1586 * @q: the queue to submit the request
1587 * @rq: the request being queued
1588 */
1589int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1590{
1591 unsigned long flags;
1592
1593 if (blk_rq_check_limits(q, rq))
1594 return -EIO;
1595
1596#ifdef CONFIG_FAIL_MAKE_REQUEST
1597 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1598 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1599 return -EIO;
1600#endif
1601
1602 spin_lock_irqsave(q->queue_lock, flags);
1603
1604 /*
1605 * Submitting request must be dequeued before calling this function
1606 * because it will be linked to another request_queue
1607 */
1608 BUG_ON(blk_queued_rq(rq));
1609
1610 drive_stat_acct(rq, 1);
1611 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1612
1613 spin_unlock_irqrestore(q->queue_lock, flags);
1614
1615 return 0;
1616}
1617EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1618
1619/**
Tejun Heo53a08802008-12-03 12:41:26 +01001620 * blkdev_dequeue_request - dequeue request and start timeout timer
1621 * @req: request to dequeue
1622 *
1623 * Dequeue @req and start timeout timer on it. This hands off the
1624 * request to the driver.
1625 *
1626 * Block internal functions which don't want to start timer should
1627 * call elv_dequeue_request().
1628 */
1629void blkdev_dequeue_request(struct request *req)
1630{
1631 elv_dequeue_request(req->q, req);
1632
1633 /*
1634 * We are now handing the request to the hardware, add the
1635 * timeout handler.
1636 */
1637 blk_add_timer(req);
1638}
1639EXPORT_SYMBOL(blkdev_dequeue_request);
1640
Jens Axboebc58ba92009-01-23 10:54:44 +01001641static void blk_account_io_completion(struct request *req, unsigned int bytes)
1642{
Jerome Marchand26308ea2009-03-27 10:31:51 +01001643 if (!blk_do_io_stat(req))
Jens Axboebc58ba92009-01-23 10:54:44 +01001644 return;
1645
1646 if (blk_fs_request(req)) {
1647 const int rw = rq_data_dir(req);
1648 struct hd_struct *part;
1649 int cpu;
1650
1651 cpu = part_stat_lock();
1652 part = disk_map_sector_rcu(req->rq_disk, req->sector);
1653 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1654 part_stat_unlock();
1655 }
1656}
1657
1658static void blk_account_io_done(struct request *req)
1659{
Jerome Marchand26308ea2009-03-27 10:31:51 +01001660 if (!blk_do_io_stat(req))
Jens Axboebc58ba92009-01-23 10:54:44 +01001661 return;
1662
1663 /*
1664 * Account IO completion. bar_rq isn't accounted as a normal
1665 * IO on queueing nor completion. Accounting the containing
1666 * request is enough.
1667 */
1668 if (blk_fs_request(req) && req != &req->q->bar_rq) {
1669 unsigned long duration = jiffies - req->start_time;
1670 const int rw = rq_data_dir(req);
1671 struct hd_struct *part;
1672 int cpu;
1673
1674 cpu = part_stat_lock();
Jerome Marchand26308ea2009-03-27 10:31:51 +01001675 part = disk_map_sector_rcu(req->rq_disk, req->sector);
Jens Axboebc58ba92009-01-23 10:54:44 +01001676
1677 part_stat_inc(cpu, part, ios[rw]);
1678 part_stat_add(cpu, part, ticks[rw], duration);
1679 part_round_stats(cpu, part);
1680 part_dec_in_flight(part);
1681
1682 part_stat_unlock();
1683 }
1684}
1685
Tejun Heo53a08802008-12-03 12:41:26 +01001686/**
Tejun Heo5efccd12009-04-23 11:05:18 +09001687 * blk_rq_bytes - Returns bytes left to complete in the entire request
1688 * @rq: the request being processed
1689 **/
1690unsigned int blk_rq_bytes(struct request *rq)
1691{
1692 if (blk_fs_request(rq))
1693 return rq->hard_nr_sectors << 9;
1694
1695 return rq->data_len;
1696}
1697EXPORT_SYMBOL_GPL(blk_rq_bytes);
1698
1699/**
1700 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1701 * @rq: the request being processed
1702 **/
1703unsigned int blk_rq_cur_bytes(struct request *rq)
1704{
1705 if (blk_fs_request(rq))
1706 return rq->current_nr_sectors << 9;
1707
1708 if (rq->bio)
1709 return rq->bio->bi_size;
1710
1711 return rq->data_len;
1712}
1713EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1714
Tejun Heo158dbda2009-04-23 11:05:18 +09001715struct request *elv_next_request(struct request_queue *q)
1716{
1717 struct request *rq;
1718 int ret;
1719
1720 while ((rq = __elv_next_request(q)) != NULL) {
1721 if (!(rq->cmd_flags & REQ_STARTED)) {
1722 /*
1723 * This is the first time the device driver
1724 * sees this request (possibly after
1725 * requeueing). Notify IO scheduler.
1726 */
1727 if (blk_sorted_rq(rq))
1728 elv_activate_rq(q, rq);
1729
1730 /*
1731 * just mark as started even if we don't start
1732 * it, a request that has been delayed should
1733 * not be passed by new incoming requests
1734 */
1735 rq->cmd_flags |= REQ_STARTED;
1736 trace_block_rq_issue(q, rq);
1737 }
1738
1739 if (!q->boundary_rq || q->boundary_rq == rq) {
1740 q->end_sector = rq_end_sector(rq);
1741 q->boundary_rq = NULL;
1742 }
1743
1744 if (rq->cmd_flags & REQ_DONTPREP)
1745 break;
1746
1747 if (q->dma_drain_size && rq->data_len) {
1748 /*
1749 * make sure space for the drain appears we
1750 * know we can do this because max_hw_segments
1751 * has been adjusted to be one fewer than the
1752 * device can handle
1753 */
1754 rq->nr_phys_segments++;
1755 }
1756
1757 if (!q->prep_rq_fn)
1758 break;
1759
1760 ret = q->prep_rq_fn(q, rq);
1761 if (ret == BLKPREP_OK) {
1762 break;
1763 } else if (ret == BLKPREP_DEFER) {
1764 /*
1765 * the request may have been (partially) prepped.
1766 * we need to keep this request in the front to
1767 * avoid resource deadlock. REQ_STARTED will
1768 * prevent other fs requests from passing this one.
1769 */
1770 if (q->dma_drain_size && rq->data_len &&
1771 !(rq->cmd_flags & REQ_DONTPREP)) {
1772 /*
1773 * remove the space for the drain we added
1774 * so that we don't add it again
1775 */
1776 --rq->nr_phys_segments;
1777 }
1778
1779 rq = NULL;
1780 break;
1781 } else if (ret == BLKPREP_KILL) {
1782 rq->cmd_flags |= REQ_QUIET;
Tejun Heo40cbbb72009-04-23 11:05:19 +09001783 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001784 } else {
1785 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1786 break;
1787 }
1788 }
1789
1790 return rq;
1791}
1792EXPORT_SYMBOL(elv_next_request);
1793
1794void elv_dequeue_request(struct request_queue *q, struct request *rq)
1795{
1796 BUG_ON(list_empty(&rq->queuelist));
1797 BUG_ON(ELV_ON_HASH(rq));
1798
1799 list_del_init(&rq->queuelist);
1800
1801 /*
1802 * the time frame between a request being removed from the lists
1803 * and to it is freed is accounted as io that is in progress at
1804 * the driver side.
1805 */
1806 if (blk_account_rq(rq))
1807 q->in_flight++;
1808}
1809
Tejun Heo5efccd12009-04-23 11:05:18 +09001810/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001811 * blk_update_request - Special helper function for request stacking drivers
1812 * @rq: the request being processed
Randy Dunlap710027a2008-08-19 20:13:11 +02001813 * @error: %0 for success, < %0 for error
Tejun Heo2e60e022009-04-23 11:05:18 +09001814 * @nr_bytes: number of bytes to complete @rq
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001815 *
1816 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09001817 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
1818 * the request structure even if @rq doesn't have leftover.
1819 * If @rq has leftover, sets it up for the next range of segments.
1820 *
1821 * This special helper function is only for request stacking drivers
1822 * (e.g. request-based dm) so that they can handle partial completion.
1823 * Actual device drivers should use blk_end_request instead.
1824 *
1825 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1826 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001827 *
1828 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09001829 * %false - this request doesn't have any more data
1830 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001831 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09001832bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05001834 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 struct bio *bio;
1836
Tejun Heo2e60e022009-04-23 11:05:18 +09001837 if (!req->bio)
1838 return false;
1839
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001840 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01001841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 /*
Tejun Heo924cec72009-04-19 07:00:41 +09001843 * For fs requests, rq is just carrier of independent bio's
1844 * and each partial completion should be handled separately.
1845 * Reset per-request error on each partial completion.
1846 *
1847 * TODO: tj: This is too subtle. It would be better to let
1848 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 */
Tejun Heo924cec72009-04-19 07:00:41 +09001850 if (blk_fs_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 req->errors = 0;
1852
Jens Axboe6728cb02008-01-31 13:03:55 +01001853 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1854 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 req->rq_disk ? req->rq_disk->disk_name : "?",
1856 (unsigned long long)req->sector);
1857 }
1858
Jens Axboebc58ba92009-01-23 10:54:44 +01001859 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01001860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 total_bytes = bio_nbytes = 0;
1862 while ((bio = req->bio) != NULL) {
1863 int nbytes;
1864
1865 if (nr_bytes >= bio->bi_size) {
1866 req->bio = bio->bi_next;
1867 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02001868 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 next_idx = 0;
1870 bio_nbytes = 0;
1871 } else {
1872 int idx = bio->bi_idx + next_idx;
1873
1874 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1875 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01001876 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Harvey Harrison24c03d42008-05-01 04:35:17 -07001877 __func__, bio->bi_idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 break;
1879 }
1880
1881 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1882 BIO_BUG_ON(nbytes > bio->bi_size);
1883
1884 /*
1885 * not a complete bvec done
1886 */
1887 if (unlikely(nbytes > nr_bytes)) {
1888 bio_nbytes += nr_bytes;
1889 total_bytes += nr_bytes;
1890 break;
1891 }
1892
1893 /*
1894 * advance to the next vector
1895 */
1896 next_idx++;
1897 bio_nbytes += nbytes;
1898 }
1899
1900 total_bytes += nbytes;
1901 nr_bytes -= nbytes;
1902
Jens Axboe6728cb02008-01-31 13:03:55 +01001903 bio = req->bio;
1904 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 /*
1906 * end more in this run, or just return 'not-done'
1907 */
1908 if (unlikely(nr_bytes <= 0))
1909 break;
1910 }
1911 }
1912
1913 /*
1914 * completely done
1915 */
Tejun Heo2e60e022009-04-23 11:05:18 +09001916 if (!req->bio) {
1917 /*
1918 * Reset counters so that the request stacking driver
1919 * can find how many bytes remain in the request
1920 * later.
1921 */
1922 req->nr_sectors = req->hard_nr_sectors = 0;
1923 req->current_nr_sectors = req->hard_cur_sectors = 0;
1924 return false;
1925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926
1927 /*
1928 * if the request wasn't completed, update state
1929 */
1930 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02001931 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 bio->bi_idx += next_idx;
1933 bio_iovec(bio)->bv_offset += nr_bytes;
1934 bio_iovec(bio)->bv_len -= nr_bytes;
1935 }
1936
1937 blk_recalc_rq_sectors(req, total_bytes >> 9);
1938 blk_recalc_rq_segments(req);
Tejun Heo2e60e022009-04-23 11:05:18 +09001939 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
Tejun Heo2e60e022009-04-23 11:05:18 +09001941EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Tejun Heo2e60e022009-04-23 11:05:18 +09001943static bool blk_update_bidi_request(struct request *rq, int error,
1944 unsigned int nr_bytes,
1945 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09001946{
Tejun Heo2e60e022009-04-23 11:05:18 +09001947 if (blk_update_request(rq, error, nr_bytes))
1948 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09001949
Tejun Heo2e60e022009-04-23 11:05:18 +09001950 /* Bidi request must be completed as a whole */
1951 if (unlikely(blk_bidi_rq(rq)) &&
1952 blk_update_request(rq->next_rq, error, bidi_bytes))
1953 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09001954
Tejun Heo2e60e022009-04-23 11:05:18 +09001955 add_disk_randomness(rq->rq_disk);
1956
1957 return false;
Tejun Heo5efccd12009-04-23 11:05:18 +09001958}
1959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960/*
1961 * queue lock must be held
1962 */
Tejun Heo2e60e022009-04-23 11:05:18 +09001963static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001965 if (blk_rq_tagged(req))
1966 blk_queue_end_tag(req->q, req);
1967
1968 if (blk_queued_rq(req))
Tejun Heo53a08802008-12-03 12:41:26 +01001969 elv_dequeue_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 if (unlikely(laptop_mode) && blk_fs_request(req))
1972 laptop_io_completion();
1973
Mike Andersone78042e52008-10-30 02:16:20 -07001974 blk_delete_timer(req);
1975
Jens Axboebc58ba92009-01-23 10:54:44 +01001976 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001977
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01001979 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05001980 else {
1981 if (blk_bidi_rq(req))
1982 __blk_put_request(req->next_rq->q, req->next_rq);
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 }
1986}
1987
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05001988/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001989 * blk_end_bidi_request - Complete a bidi request
1990 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02001991 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05001992 * @nr_bytes: number of bytes to complete @rq
1993 * @bidi_bytes: number of bytes to complete @rq->next_rq
1994 *
1995 * Description:
1996 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09001997 * Drivers that supports bidi can safely call this member for any
1998 * type of request, bidi or uni. In the later case @bidi_bytes is
1999 * just ignored.
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002000 *
2001 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002002 * %false - we are done with this request
2003 * %true - still buffers pending for this request
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002004 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002005bool blk_end_bidi_request(struct request *rq, int error,
2006 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002007{
Tejun Heo2e60e022009-04-23 11:05:18 +09002008 struct request_queue *q = rq->q;
2009 unsigned long flags;
2010
2011 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2012 return true;
2013
2014 spin_lock_irqsave(q->queue_lock, flags);
2015 blk_finish_request(rq, error);
2016 spin_unlock_irqrestore(q->queue_lock, flags);
2017
2018 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002019}
2020EXPORT_SYMBOL_GPL(blk_end_bidi_request);
2021
2022/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002023 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2024 * @rq: the request to complete
2025 * @error: %0 for success, < %0 for error
2026 * @nr_bytes: number of bytes to complete @rq
2027 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002028 *
2029 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002030 * Identical to blk_end_bidi_request() except that queue lock is
2031 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002032 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002033 * Return:
2034 * %false - we are done with this request
2035 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002036 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002037bool __blk_end_bidi_request(struct request *rq, int error,
2038 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002039{
Tejun Heo2e60e022009-04-23 11:05:18 +09002040 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2041 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002042
Tejun Heo2e60e022009-04-23 11:05:18 +09002043 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002044
Tejun Heo2e60e022009-04-23 11:05:18 +09002045 return false;
Tejun Heo5efccd12009-04-23 11:05:18 +09002046}
Tejun Heo2e60e022009-04-23 11:05:18 +09002047EXPORT_SYMBOL_GPL(__blk_end_bidi_request);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002048
Jens Axboe86db1e22008-01-29 14:53:40 +01002049void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2050 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051{
David Woodhoused628eae2008-08-09 16:22:17 +01002052 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2053 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002054 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
David Woodhousefb2dce82008-08-05 18:01:53 +01002056 if (bio_has_data(bio)) {
2057 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002058 rq->buffer = bio_data(bio);
2059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 rq->current_nr_sectors = bio_cur_sectors(bio);
2061 rq->hard_cur_sectors = rq->current_nr_sectors;
2062 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002063 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
2065 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
NeilBrown66846572007-08-16 13:31:28 +02002067 if (bio->bi_bdev)
2068 rq->rq_disk = bio->bi_bdev->bd_disk;
2069}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002071/**
2072 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2073 * @q : the queue of the device being checked
2074 *
2075 * Description:
2076 * Check if underlying low-level drivers of a device are busy.
2077 * If the drivers want to export their busy state, they must set own
2078 * exporting function using blk_queue_lld_busy() first.
2079 *
2080 * Basically, this function is used only by request stacking drivers
2081 * to stop dispatching requests to underlying devices when underlying
2082 * devices are busy. This behavior helps more I/O merging on the queue
2083 * of the request stacking driver and prevents I/O throughput regression
2084 * on burst I/O load.
2085 *
2086 * Return:
2087 * 0 - Not busy (The request stacking driver should dispatch request)
2088 * 1 - Busy (The request stacking driver should stop dispatching request)
2089 */
2090int blk_lld_busy(struct request_queue *q)
2091{
2092 if (q->lld_busy_fn)
2093 return q->lld_busy_fn(q);
2094
2095 return 0;
2096}
2097EXPORT_SYMBOL_GPL(blk_lld_busy);
2098
Jens Axboe18887ad2008-07-28 13:08:45 +02002099int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100{
2101 return queue_work(kblockd_workqueue, work);
2102}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103EXPORT_SYMBOL(kblockd_schedule_work);
2104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105int __init blk_dev_init(void)
2106{
2107 kblockd_workqueue = create_workqueue("kblockd");
2108 if (!kblockd_workqueue)
2109 panic("Failed to create kblockd\n");
2110
2111 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002112 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Jens Axboe8324aa92008-01-29 14:51:59 +01002114 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002115 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return 0;
2118}
2119