blob: 90e1ffdeb415914ffaad20b31f54c8e22d05ce79 [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>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Li Zefan55782132009-06-09 13:43:05 +080031
32#define CREATE_TRACE_POINTS
33#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Jens Axboe8324aa92008-01-29 14:51:59 +010035#include "blk.h"
36
Mike Snitzerd07335e2010-11-16 12:52:38 +010037EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020038EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080039EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010040
Jens Axboe165125e2007-07-24 09:28:11 +020041static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For the allocated request tables
45 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010046static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
49 * For queue allocation
50 */
Jens Axboe6728cb02008-01-31 13:03:55 +010051struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * Controlling structure to kblockd
55 */
Jens Axboeff856ba2006-01-09 16:02:34 +010056static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboe26b82562008-01-29 13:54:41 +010058static void drive_stat_acct(struct request *rq, int new_io)
59{
Jens Axboe28f13702008-05-07 10:15:46 +020060 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010061 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090062 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010063
Jens Axboec2553b52009-04-24 08:10:11 +020064 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010065 return;
66
Tejun Heo074a7ac2008-08-25 19:56:14 +090067 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090068
Jerome Marchand09e099d2011-01-05 16:57:38 +010069 if (!new_io) {
70 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090071 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010072 } else {
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010074 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010075 /*
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
78 *
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
82 */
83 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010084 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010085 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090086 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020087 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010088 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010089 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020090
Tejun Heo074a7ac2008-08-25 19:56:14 +090091 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010092}
93
Jens Axboe8324aa92008-01-29 14:51:59 +010094void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095{
96 int nr;
97
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
107}
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/**
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * @bdev: device
112 *
113 * Locates the passed device's request queue and returns the address of its
114 * backing_dev_info
115 *
116 * Will return NULL if the request queue cannot be located.
117 */
118struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119{
120 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200121 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123 if (q)
124 ret = &q->backing_dev_info;
125 return ret;
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127EXPORT_SYMBOL(blk_get_backing_dev_info);
128
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200129void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200131 memset(rq, 0, sizeof(*rq));
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700134 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200135 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100136 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900137 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200140 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800141 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100142 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100143 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900144 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700145 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100146 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200148EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
NeilBrown5bb23a62007-09-27 12:46:13 +0200150static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100152{
Tejun Heo143a87f2011-01-25 12:43:52 +0100153 if (error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100157
Tejun Heo143a87f2011-01-25 12:43:52 +0100158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200162 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100163
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
166
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
169
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
172
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
Jens Axboe6728cb02008-01-31 13:03:55 +0100182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Tejun Heo83096eb2009-05-07 22:24:39 +0900186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100193 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 printk("%02x ", rq->cmd[bit]);
196 printk("\n");
197 }
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199EXPORT_SYMBOL(blk_dump_rq_flags);
200
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500201static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200202{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500203 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200204
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200207 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500208 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 *
216 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
220 */
221void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Jens Axboe4521cc42011-04-18 11:36:39 +0200223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500226EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/**
229 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200230 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 *
232 * Description:
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200237void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200239 WARN_ON(!irqs_disabled());
240
Nick Piggin75ad23b2008-04-29 14:48:33 +0200241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200242 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244EXPORT_SYMBOL(blk_start_queue);
245
246/**
247 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200248 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 *
250 * Description:
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200260void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
Jens Axboead3d9d72011-03-25 16:58:59 +0100262 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265EXPORT_SYMBOL(blk_stop_queue);
266
267/**
268 * blk_sync_queue - cancel any pending callbacks on a queue
269 * @q: the queue
270 *
271 * Description:
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200276 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * that its ->make_request_fn will not re-add plugging prior to calling
278 * this function.
279 *
Vivek Goyalda527772011-03-02 19:05:33 -0500280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
283 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
285void blk_sync_queue(struct request_queue *q)
286{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100287 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500288 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290EXPORT_SYMBOL(blk_sync_queue);
291
292/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200293 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200295 *
296 * Description:
297 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200298 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200300void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Tejun Heoa538cd02009-04-23 11:05:17 +0900302 if (unlikely(blk_queue_stopped(q)))
303 return;
304
Jens Axboec21e6be2011-04-19 13:32:46 +0200305 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200306}
307EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200308
Nick Piggin75ad23b2008-04-29 14:48:33 +0200309/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
312 *
313 * Description:
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us.
316 */
317void blk_run_queue_async(struct request_queue *q)
318{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200322 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200323}
Jens Axboec21e6be2011-04-19 13:32:46 +0200324EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200325
326/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200329 *
330 * Description:
331 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900332 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200333 */
334void blk_run_queue(struct request_queue *q)
335{
336 unsigned long flags;
337
338 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200339 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 spin_unlock_irqrestore(q->queue_lock, flags);
341}
342EXPORT_SYMBOL(blk_run_queue);
343
Jens Axboe165125e2007-07-24 09:28:11 +0200344void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500345{
346 kobject_put(&q->kobj);
347}
Jens Axboed86e0e82011-05-27 07:44:43 +0200348EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500349
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500350/*
351 * Note: If a driver supplied the queue lock, it should not zap that lock
352 * unexpectedly as some queue cleanup components like elevator_exit() and
353 * blk_throtl_exit() need queue lock.
354 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100355void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500356{
Jens Axboee3335de2008-09-18 09:22:54 -0700357 /*
358 * We know we have process context here, so we can be a little
359 * cautious and ensure that pending block actions on this device
360 * are done before moving on. Going into this function, we should
361 * not have processes doing IO to this device.
362 */
363 blk_sync_queue(q);
364
Matthew Garrett31373d02010-04-06 14:25:14 +0200365 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500366 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200367 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500368 mutex_unlock(&q->sysfs_lock);
369
370 if (q->elevator)
371 elevator_exit(q->elevator);
372
Vivek Goyalda527772011-03-02 19:05:33 -0500373 blk_throtl_exit(q);
374
Al Viro483f4af2006-03-18 18:34:37 -0500375 blk_put_queue(q);
376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377EXPORT_SYMBOL(blk_cleanup_queue);
378
Jens Axboe165125e2007-07-24 09:28:11 +0200379static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
381 struct request_list *rl = &q->rq;
382
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400383 if (unlikely(rl->rq_pool))
384 return 0;
385
Jens Axboe1faa16d2009-04-06 14:48:01 +0200386 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
387 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200388 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200389 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
390 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Christoph Lameter19460892005-06-23 00:08:19 -0700392 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
393 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 if (!rl->rq_pool)
396 return -ENOMEM;
397
398 return 0;
399}
400
Jens Axboe165125e2007-07-24 09:28:11 +0200401struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
Christoph Lameter19460892005-06-23 00:08:19 -0700403 return blk_alloc_queue_node(gfp_mask, -1);
404}
405EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Jens Axboe165125e2007-07-24 09:28:11 +0200407struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700408{
Jens Axboe165125e2007-07-24 09:28:11 +0200409 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700410 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700411
Jens Axboe8324aa92008-01-29 14:51:59 +0100412 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700413 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (!q)
415 return NULL;
416
Jens Axboe0989a022009-06-12 14:42:56 +0200417 q->backing_dev_info.ra_pages =
418 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
419 q->backing_dev_info.state = 0;
420 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200421 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200422
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700423 err = bdi_init(&q->backing_dev_info);
424 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100425 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700426 return NULL;
427 }
428
Vivek Goyale43473b2010-09-15 17:06:35 -0400429 if (blk_throtl_init(q)) {
430 kmem_cache_free(blk_requestq_cachep, q);
431 return NULL;
432 }
433
Matthew Garrett31373d02010-04-06 14:25:14 +0200434 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
435 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700436 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
437 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100438 INIT_LIST_HEAD(&q->flush_queue[0]);
439 INIT_LIST_HEAD(&q->flush_queue[1]);
440 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500441 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500442
Jens Axboe8324aa92008-01-29 14:51:59 +0100443 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Al Viro483f4af2006-03-18 18:34:37 -0500445 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700446 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500447
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500448 /*
449 * By default initialize queue_lock to internal lock and driver can
450 * override it later if need be.
451 */
452 q->queue_lock = &q->__queue_lock;
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return q;
455}
Christoph Lameter19460892005-06-23 00:08:19 -0700456EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458/**
459 * blk_init_queue - prepare a request queue for use with a block device
460 * @rfn: The function to be called to process requests that have been
461 * placed on the queue.
462 * @lock: Request queue spin lock
463 *
464 * Description:
465 * If a block device wishes to use the standard request handling procedures,
466 * which sorts requests and coalesces adjacent requests, then it must
467 * call blk_init_queue(). The function @rfn will be called when there
468 * are requests on the queue that need to be processed. If the device
469 * supports plugging, then @rfn may not be called immediately when requests
470 * are available on the queue, but may be called at some time later instead.
471 * Plugged queues are generally unplugged when a buffer belonging to one
472 * of the requests on the queue is needed, or due to memory pressure.
473 *
474 * @rfn is not required, or even expected, to remove all requests off the
475 * queue, but only as many as it can handle at a time. If it does leave
476 * requests on the queue, it is responsible for arranging that the requests
477 * get dealt with eventually.
478 *
479 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200480 * request queue; this lock will be taken also from interrupt context, so irq
481 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200483 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 * it didn't succeed.
485 *
486 * Note:
487 * blk_init_queue() must be paired with a blk_cleanup_queue() call
488 * when the block device is deactivated (such as at module unload).
489 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700490
Jens Axboe165125e2007-07-24 09:28:11 +0200491struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Christoph Lameter19460892005-06-23 00:08:19 -0700493 return blk_init_queue_node(rfn, lock, -1);
494}
495EXPORT_SYMBOL(blk_init_queue);
496
Jens Axboe165125e2007-07-24 09:28:11 +0200497struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700498blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
499{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600500 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600502 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
503 if (!uninit_q)
504 return NULL;
505
506 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
507 if (!q)
508 blk_cleanup_queue(uninit_q);
509
510 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200511}
512EXPORT_SYMBOL(blk_init_queue_node);
513
514struct request_queue *
515blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
516 spinlock_t *lock)
517{
518 return blk_init_allocated_queue_node(q, rfn, lock, -1);
519}
520EXPORT_SYMBOL(blk_init_allocated_queue);
521
522struct request_queue *
523blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
524 spinlock_t *lock, int node_id)
525{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (!q)
527 return NULL;
528
Christoph Lameter19460892005-06-23 00:08:19 -0700529 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600530 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500531 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900535 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100536 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500537
538 /* Override internal queue lock with supplied lock pointer */
539 if (lock)
540 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Jens Axboef3b144a2009-03-06 08:48:33 +0100542 /*
543 * This also sets hw/phys segments, boundary and size
544 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Alan Stern44ec9542007-02-20 11:01:57 -0500547 q->sg_reserved_size = INT_MAX;
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 /*
550 * all done
551 */
552 if (!elevator_init(q, NULL)) {
553 blk_queue_congestion_threshold(q);
554 return q;
555 }
556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return NULL;
558}
Mike Snitzer01effb02010-05-11 08:57:42 +0200559EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Jens Axboe165125e2007-07-24 09:28:11 +0200561int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700563 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500564 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 return 0;
566 }
567
568 return 1;
569}
Jens Axboed86e0e82011-05-27 07:44:43 +0200570EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Jens Axboe165125e2007-07-24 09:28:11 +0200572static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200574 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200575 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 mempool_free(rq, q->rq.rq_pool);
577}
578
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200579static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200580blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
582 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
583
584 if (!rq)
585 return NULL;
586
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200587 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200588
Jerome Marchand42dad762009-04-22 14:01:49 +0200589 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Tejun Heocb98fc82005-10-28 08:29:39 +0200591 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200592 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200593 mempool_free(rq, q->rq.rq_pool);
594 return NULL;
595 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200596 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Tejun Heocb98fc82005-10-28 08:29:39 +0200599 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
602/*
603 * ioc_batching returns true if the ioc is a valid batching request and
604 * should be given priority access to a request.
605 */
Jens Axboe165125e2007-07-24 09:28:11 +0200606static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 if (!ioc)
609 return 0;
610
611 /*
612 * Make sure the process is able to allocate at least 1 request
613 * even if the batch times out, otherwise we could theoretically
614 * lose wakeups.
615 */
616 return ioc->nr_batch_requests == q->nr_batching ||
617 (ioc->nr_batch_requests > 0
618 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
619}
620
621/*
622 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
623 * will cause the process to be a "batcher" on all queues in the system. This
624 * is the behaviour we want though - once it gets a wakeup it should be given
625 * a nice run.
626 */
Jens Axboe165125e2007-07-24 09:28:11 +0200627static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 if (!ioc || ioc_batching(q, ioc))
630 return;
631
632 ioc->nr_batch_requests = q->nr_batching;
633 ioc->last_waited = jiffies;
634}
635
Jens Axboe1faa16d2009-04-06 14:48:01 +0200636static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 struct request_list *rl = &q->rq;
639
Jens Axboe1faa16d2009-04-06 14:48:01 +0200640 if (rl->count[sync] < queue_congestion_off_threshold(q))
641 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Jens Axboe1faa16d2009-04-06 14:48:01 +0200643 if (rl->count[sync] + 1 <= q->nr_requests) {
644 if (waitqueue_active(&rl->wait[sync]))
645 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Jens Axboe1faa16d2009-04-06 14:48:01 +0200647 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649}
650
651/*
652 * A request has just been released. Account for it, update the full and
653 * congestion status, wake up any waiters. Called under q->queue_lock.
654 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200655static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 struct request_list *rl = &q->rq;
658
Jens Axboe1faa16d2009-04-06 14:48:01 +0200659 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200660 if (priv)
661 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Jens Axboe1faa16d2009-04-06 14:48:01 +0200663 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Jens Axboe1faa16d2009-04-06 14:48:01 +0200665 if (unlikely(rl->starved[sync ^ 1]))
666 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100670 * Determine if elevator data should be initialized when allocating the
671 * request associated with @bio.
672 */
673static bool blk_rq_should_init_elevator(struct bio *bio)
674{
675 if (!bio)
676 return true;
677
678 /*
679 * Flush requests do not use the elevator so skip initialization.
680 * This allows a request to share the flush and elevator data.
681 */
682 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
683 return false;
684
685 return true;
686}
687
688/*
Nick Piggind6344532005-06-28 20:45:14 -0700689 * Get a free request, queue_lock must be held.
690 * Returns NULL on failure, with queue_lock held.
691 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 */
Jens Axboe165125e2007-07-24 09:28:11 +0200693static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100694 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct request *rq = NULL;
697 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100698 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200699 const bool is_sync = rw_is_sync(rw_flags) != 0;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100700 int may_queue, priv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Jens Axboe7749a8d2006-12-13 13:02:26 +0100702 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100703 if (may_queue == ELV_MQUEUE_NO)
704 goto rq_starved;
705
Jens Axboe1faa16d2009-04-06 14:48:01 +0200706 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
707 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200708 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100709 /*
710 * The queue will fill after this allocation, so set
711 * it as full, and mark this process as "batching".
712 * This process will be allowed to complete a batch of
713 * requests, others will be blocked.
714 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200715 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100716 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200717 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100718 } else {
719 if (may_queue != ELV_MQUEUE_MUST
720 && !ioc_batching(q, ioc)) {
721 /*
722 * The queue is full and the allocating
723 * process is not a "batcher", and not
724 * exempted by the IO scheduler
725 */
726 goto out;
727 }
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200730 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732
Jens Axboe082cf692005-06-28 16:35:11 +0200733 /*
734 * Only allow batching queuers to allocate up to 50% over the defined
735 * limit of requests, otherwise we could have thousands of requests
736 * allocated with any setting of ->nr_requests
737 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200738 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200739 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100740
Jens Axboe1faa16d2009-04-06 14:48:01 +0200741 rl->count[is_sync]++;
742 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200743
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100744 if (blk_rq_should_init_elevator(bio)) {
745 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
746 if (priv)
747 rl->elvpriv++;
748 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200749
Jens Axboef253b862010-10-24 22:06:02 +0200750 if (blk_queue_io_stat(q))
751 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 spin_unlock_irq(q->queue_lock);
753
Jens Axboe7749a8d2006-12-13 13:02:26 +0100754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100755 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /*
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
759 *
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
762 */
763 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200764 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /*
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
772 */
773rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200774 if (unlikely(rl->count[is_sync] == 0))
775 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto out;
778 }
779
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100780 /*
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100788
Jens Axboe1faa16d2009-04-06 14:48:01 +0200789 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return rq;
792}
793
794/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100795 * No available requests for this queue, wait for some requests to become
796 * available.
Nick Piggind6344532005-06-28 20:45:14 -0700797 *
798 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Jens Axboe165125e2007-07-24 09:28:11 +0200800static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200801 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200803 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 struct request *rq;
805
Jens Axboe7749a8d2006-12-13 13:02:26 +0100806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700807 while (!rq) {
808 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200809 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct request_list *rl = &q->rq;
811
Jens Axboe1faa16d2009-04-06 14:48:01 +0200812 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 TASK_UNINTERRUPTIBLE);
814
Jens Axboe1faa16d2009-04-06 14:48:01 +0200815 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200817 spin_unlock_irq(q->queue_lock);
818 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200820 /*
821 * After sleeping, we become a "batching" process and
822 * will be able to allocate at least one request, and
823 * up to a big batch of them for a small period time.
824 * See ioc_batching, ioc_set_batching
825 */
826 ioc = current_io_context(GFP_NOIO, q->node);
827 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100828
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200829 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200830 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200831
832 rq = get_request(q, rw_flags, bio, GFP_NOIO);
833 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 return rq;
836}
837
Jens Axboe165125e2007-07-24 09:28:11 +0200838struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
840 struct request *rq;
841
James Bottomleybfe159a2011-07-07 15:45:40 -0500842 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
843 return NULL;
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 BUG_ON(rw != READ && rw != WRITE);
846
Nick Piggind6344532005-06-28 20:45:14 -0700847 spin_lock_irq(q->queue_lock);
848 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200849 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700850 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200851 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700852 if (!rq)
853 spin_unlock_irq(q->queue_lock);
854 }
855 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 return rq;
858}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859EXPORT_SYMBOL(blk_get_request);
860
861/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300862 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700863 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300864 * @bio: The bio describing the memory mappings that will be submitted for IO.
865 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700866 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200867 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300868 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
869 * type commands. Where the struct request needs to be farther initialized by
870 * the caller. It is passed a &struct bio, which describes the memory info of
871 * the I/O transfer.
872 *
873 * The caller of blk_make_request must make sure that bi_io_vec
874 * are set to describe the memory buffers. That bio_data_dir() will return
875 * the needed direction of the request. (And all bio's in the passed bio-chain
876 * are properly set accordingly)
877 *
878 * If called under none-sleepable conditions, mapped bio buffers must not
879 * need bouncing, by calling the appropriate masked or flagged allocator,
880 * suitable for the target device. Otherwise the call to blk_queue_bounce will
881 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200882 *
883 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
884 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
885 * anything but the first bio in the chain. Otherwise you risk waiting for IO
886 * completion of a bio that hasn't been submitted yet, thus resulting in a
887 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
888 * of bio_alloc(), as that avoids the mempool deadlock.
889 * If possible a big IO should be split into smaller parts when allocation
890 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200891 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300892struct request *blk_make_request(struct request_queue *q, struct bio *bio,
893 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200894{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300895 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
896
897 if (unlikely(!rq))
898 return ERR_PTR(-ENOMEM);
899
900 for_each_bio(bio) {
901 struct bio *bounce_bio = bio;
902 int ret;
903
904 blk_queue_bounce(q, &bounce_bio);
905 ret = blk_rq_append_bio(q, rq, bounce_bio);
906 if (unlikely(ret)) {
907 blk_put_request(rq);
908 return ERR_PTR(ret);
909 }
910 }
911
912 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200913}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300914EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200915
916/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 * blk_requeue_request - put a request back on queue
918 * @q: request queue where request should be inserted
919 * @rq: request to be inserted
920 *
921 * Description:
922 * Drivers often keep queueing requests until the hardware cannot accept
923 * more, when that condition happens we need to put the request back
924 * on the queue. Must be called with queue lock held.
925 */
Jens Axboe165125e2007-07-24 09:28:11 +0200926void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700928 blk_delete_timer(rq);
929 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100930 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 if (blk_rq_tagged(rq))
933 blk_queue_end_tag(q, rq);
934
James Bottomleyba396a62009-05-27 14:17:08 +0200935 BUG_ON(blk_queued_rq(rq));
936
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 elv_requeue_request(q, rq);
938}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939EXPORT_SYMBOL(blk_requeue_request);
940
Jens Axboe73c10102011-03-08 13:19:51 +0100941static void add_acct_request(struct request_queue *q, struct request *rq,
942 int where)
943{
944 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100945 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +0100946}
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200949 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 * @q: request queue where request should be inserted
951 * @rq: request to be inserted
952 * @at_head: insert request at head or tail of queue
953 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 *
955 * Description:
956 * Many block devices need to execute commands asynchronously, so they don't
957 * block the whole kernel from preemption during request execution. This is
958 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200959 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
960 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 *
962 * We have the option of inserting the head or the tail of the queue.
963 * Typically we use the tail for new ioctls and so forth. We use the head
964 * of the queue for things like a QUEUE_FULL message from a device, or a
965 * host that is unable to accept a particular command.
966 */
Jens Axboe165125e2007-07-24 09:28:11 +0200967void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500968 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969{
Tejun Heo 867d1192005-04-24 02:06:05 -0500970 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 unsigned long flags;
972
973 /*
974 * tell I/O scheduler that this isn't a regular read/write (ie it
975 * must not attempt merges on this) and that it acts as a soft
976 * barrier
977 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200978 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 rq->special = data;
981
982 spin_lock_irqsave(q->queue_lock, flags);
983
984 /*
985 * If command is tagged, release the tag
986 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500987 if (blk_rq_tagged(rq))
988 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Jens Axboe73c10102011-03-08 13:19:51 +0100990 add_acct_request(q, rq, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200991 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 spin_unlock_irqrestore(q->queue_lock, flags);
993}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994EXPORT_SYMBOL(blk_insert_request);
995
Tejun Heo074a7ac2008-08-25 19:56:14 +0900996static void part_round_stats_single(int cpu, struct hd_struct *part,
997 unsigned long now)
998{
999 if (now == part->stamp)
1000 return;
1001
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001002 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001003 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001004 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001005 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1006 }
1007 part->stamp = now;
1008}
1009
1010/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001011 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1012 * @cpu: cpu number for stats access
1013 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 *
1015 * The average IO queue length and utilisation statistics are maintained
1016 * by observing the current state of the queue length and the amount of
1017 * time it has been in this state for.
1018 *
1019 * Normally, that accounting is done on IO completion, but that can result
1020 * in more than a second's worth of IO being accounted for within any one
1021 * second, leading to >100% utilisation. To deal with that, we call this
1022 * function to do a round-off before returning the results when reading
1023 * /proc/diskstats. This accounts immediately for all queue usage up to
1024 * the current jiffies and restarts the counters again.
1025 */
Tejun Heoc9959052008-08-25 19:47:21 +09001026void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001027{
1028 unsigned long now = jiffies;
1029
Tejun Heo074a7ac2008-08-25 19:56:14 +09001030 if (part->partno)
1031 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1032 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001033}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001034EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036/*
1037 * queue lock must be held
1038 */
Jens Axboe165125e2007-07-24 09:28:11 +02001039void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 if (unlikely(!q))
1042 return;
1043 if (unlikely(--req->ref_count))
1044 return;
1045
Tejun Heo8922e162005-10-20 16:23:44 +02001046 elv_completed_request(q, req);
1047
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001048 /* this is a bio leak */
1049 WARN_ON(req->bio != NULL);
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /*
1052 * Request may not have originated from ll_rw_blk. if not,
1053 * it didn't come out of our reserved rq pools
1054 */
Jens Axboe49171e52006-08-10 08:59:11 +02001055 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001056 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001057 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001060 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001063 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
1065}
Mike Christie6e39b692005-11-11 05:30:24 -06001066EXPORT_SYMBOL_GPL(__blk_put_request);
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068void blk_put_request(struct request *req)
1069{
Tejun Heo8922e162005-10-20 16:23:44 +02001070 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001071 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001073 spin_lock_irqsave(q->queue_lock, flags);
1074 __blk_put_request(q, req);
1075 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077EXPORT_SYMBOL(blk_put_request);
1078
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001079/**
1080 * blk_add_request_payload - add a payload to a request
1081 * @rq: request to update
1082 * @page: page backing the payload
1083 * @len: length of the payload.
1084 *
1085 * This allows to later add a payload to an already submitted request by
1086 * a block driver. The driver needs to take care of freeing the payload
1087 * itself.
1088 *
1089 * Note that this is a quite horrible hack and nothing but handling of
1090 * discard requests should ever use it.
1091 */
1092void blk_add_request_payload(struct request *rq, struct page *page,
1093 unsigned int len)
1094{
1095 struct bio *bio = rq->bio;
1096
1097 bio->bi_io_vec->bv_page = page;
1098 bio->bi_io_vec->bv_offset = 0;
1099 bio->bi_io_vec->bv_len = len;
1100
1101 bio->bi_size = len;
1102 bio->bi_vcnt = 1;
1103 bio->bi_phys_segments = 1;
1104
1105 rq->__data_len = rq->resid_len = len;
1106 rq->nr_phys_segments = 1;
1107 rq->buffer = bio_data(bio);
1108}
1109EXPORT_SYMBOL_GPL(blk_add_request_payload);
1110
Jens Axboe73c10102011-03-08 13:19:51 +01001111static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1112 struct bio *bio)
1113{
1114 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1115
Jens Axboe73c10102011-03-08 13:19:51 +01001116 if (!ll_back_merge_fn(q, req, bio))
1117 return false;
1118
1119 trace_block_bio_backmerge(q, bio);
1120
1121 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1122 blk_rq_set_mixed_merge(req);
1123
1124 req->biotail->bi_next = bio;
1125 req->biotail = bio;
1126 req->__data_len += bio->bi_size;
1127 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1128
1129 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001130 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001131 return true;
1132}
1133
1134static bool bio_attempt_front_merge(struct request_queue *q,
1135 struct request *req, struct bio *bio)
1136{
1137 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001138
Jens Axboe73c10102011-03-08 13:19:51 +01001139 if (!ll_front_merge_fn(q, req, bio))
1140 return false;
1141
1142 trace_block_bio_frontmerge(q, bio);
1143
1144 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1145 blk_rq_set_mixed_merge(req);
1146
Jens Axboe73c10102011-03-08 13:19:51 +01001147 bio->bi_next = req->bio;
1148 req->bio = bio;
1149
1150 /*
1151 * may not be valid. if the low level driver said
1152 * it didn't need a bounce buffer then it better
1153 * not touch req->buffer either...
1154 */
1155 req->buffer = bio_data(bio);
1156 req->__sector = bio->bi_sector;
1157 req->__data_len += bio->bi_size;
1158 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1159
1160 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001161 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001162 return true;
1163}
1164
1165/*
1166 * Attempts to merge with the plugged list in the current process. Returns
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001167 * true if merge was successful, otherwise false.
Jens Axboe73c10102011-03-08 13:19:51 +01001168 */
1169static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1170 struct bio *bio)
1171{
1172 struct blk_plug *plug;
1173 struct request *rq;
1174 bool ret = false;
1175
1176 plug = tsk->plug;
1177 if (!plug)
1178 goto out;
1179
1180 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1181 int el_ret;
1182
1183 if (rq->q != q)
1184 continue;
1185
1186 el_ret = elv_try_merge(rq, bio);
1187 if (el_ret == ELEVATOR_BACK_MERGE) {
1188 ret = bio_attempt_back_merge(q, rq, bio);
1189 if (ret)
1190 break;
1191 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1192 ret = bio_attempt_front_merge(q, rq, bio);
1193 if (ret)
1194 break;
1195 }
1196 }
1197out:
1198 return ret;
1199}
1200
Jens Axboe86db1e22008-01-29 14:53:40 +01001201void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001202{
Jens Axboec7c22e42008-09-13 20:26:01 +02001203 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001204 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001205
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001206 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1207 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001208 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001209
Tejun Heo52d9e672006-01-06 09:49:58 +01001210 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001211 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001212 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001213 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001214}
1215
Jens Axboe165125e2007-07-24 09:28:11 +02001216static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001218 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001219 struct blk_plug *plug;
1220 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1221 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 /*
1224 * low level driver can indicate that it wants pages above a
1225 * certain limit bounced to low memory (ie for highmem, or even
1226 * ISA dma in theory)
1227 */
1228 blk_queue_bounce(q, &bio);
1229
Tejun Heo4fed9472010-09-03 11:56:17 +02001230 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001231 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001232 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001233 goto get_rq;
1234 }
1235
Jens Axboe73c10102011-03-08 13:19:51 +01001236 /*
1237 * Check if we can merge with the plugged list before grabbing
1238 * any locks.
1239 */
1240 if (attempt_plug_merge(current, q, bio))
1241 goto out;
1242
1243 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
1245 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001246 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001247 if (bio_attempt_back_merge(q, req, bio)) {
1248 if (!attempt_back_merge(q, req))
1249 elv_merged_request(q, req, el_ret);
1250 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001251 }
Jens Axboe73c10102011-03-08 13:19:51 +01001252 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001253 if (bio_attempt_front_merge(q, req, bio)) {
1254 if (!attempt_front_merge(q, req))
1255 elv_merged_request(q, req, el_ret);
1256 goto out_unlock;
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 }
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001261 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001262 * This sync check and mask will be re-done in init_request_from_bio(),
1263 * but we need to set it earlier to expose the sync flag to the
1264 * rq allocator and io schedulers.
1265 */
1266 rw_flags = bio_data_dir(bio);
1267 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001268 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001269
1270 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001271 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001272 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001273 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001274 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001275
Nick Piggin450991b2005-06-28 20:45:13 -07001276 /*
1277 * After dropping the lock and possibly sleeping here, our request
1278 * may now be mergeable after it had proven unmergeable (above).
1279 * We don't worry about that case for efficiency. It won't happen
1280 * often, and the elevators are able to handle it.
1281 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001282 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Jens Axboec7c22e42008-09-13 20:26:01 +02001284 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Dan Williams5757a6d2011-07-23 20:44:25 +02001285 bio_flagged(bio, BIO_CPU_AFFINE))
Jens Axboe11ccf112011-07-26 15:01:15 +02001286 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001287
Jens Axboe73c10102011-03-08 13:19:51 +01001288 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001289 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001290 /*
1291 * If this is the first request added after a plug, fire
1292 * of a plug trace. If others have been added before, check
1293 * if we have multiple devices in this plug. If so, make a
1294 * note to sort the list before dispatch.
1295 */
1296 if (list_empty(&plug->list))
1297 trace_block_plug(q);
1298 else if (!plug->should_sort) {
Jens Axboe73c10102011-03-08 13:19:51 +01001299 struct request *__rq;
1300
1301 __rq = list_entry_rq(plug->list.prev);
1302 if (__rq->q != q)
1303 plug->should_sort = 1;
1304 }
Jens Axboe73c10102011-03-08 13:19:51 +01001305 list_add_tail(&req->queuelist, &plug->list);
Shaohua Li55c022b2011-07-08 08:19:20 +02001306 plug->count++;
Jens Axboe73c10102011-03-08 13:19:51 +01001307 drive_stat_acct(req, 1);
Shaohua Li55c022b2011-07-08 08:19:20 +02001308 if (plug->count >= BLK_MAX_REQUEST_COUNT)
1309 blk_flush_plug_list(plug, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001310 } else {
1311 spin_lock_irq(q->queue_lock);
1312 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001313 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001314out_unlock:
1315 spin_unlock_irq(q->queue_lock);
1316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320
1321/*
1322 * If bio->bi_dev is a partition, remap the location
1323 */
1324static inline void blk_partition_remap(struct bio *bio)
1325{
1326 struct block_device *bdev = bio->bi_bdev;
1327
Jens Axboebf2de6f2007-09-27 13:01:25 +02001328 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 bio->bi_sector += p->start_sect;
1332 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001333
Mike Snitzerd07335e2010-11-16 12:52:38 +01001334 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1335 bdev->bd_dev,
1336 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
1338}
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340static void handle_bad_sector(struct bio *bio)
1341{
1342 char b[BDEVNAME_SIZE];
1343
1344 printk(KERN_INFO "attempt to access beyond end of device\n");
1345 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1346 bdevname(bio->bi_bdev, b),
1347 bio->bi_rw,
1348 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001349 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
1351 set_bit(BIO_EOF, &bio->bi_flags);
1352}
1353
Akinobu Mitac17bb492006-12-08 02:39:46 -08001354#ifdef CONFIG_FAIL_MAKE_REQUEST
1355
1356static DECLARE_FAULT_ATTR(fail_make_request);
1357
1358static int __init setup_fail_make_request(char *str)
1359{
1360 return setup_fault_attr(&fail_make_request, str);
1361}
1362__setup("fail_make_request=", setup_fail_make_request);
1363
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001364static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001365{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001366 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001367}
1368
1369static int __init fail_make_request_debugfs(void)
1370{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001371 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1372 NULL, &fail_make_request);
1373
1374 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001375}
1376
1377late_initcall(fail_make_request_debugfs);
1378
1379#else /* CONFIG_FAIL_MAKE_REQUEST */
1380
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001381static inline bool should_fail_request(struct hd_struct *part,
1382 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001383{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001384 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001385}
1386
1387#endif /* CONFIG_FAIL_MAKE_REQUEST */
1388
Jens Axboec07e2b42007-07-18 13:27:58 +02001389/*
1390 * Check whether this bio extends beyond the end of the device.
1391 */
1392static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1393{
1394 sector_t maxsector;
1395
1396 if (!nr_sectors)
1397 return 0;
1398
1399 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001400 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001401 if (maxsector) {
1402 sector_t sector = bio->bi_sector;
1403
1404 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1405 /*
1406 * This may well happen - the kernel calls bread()
1407 * without checking the size of the device, e.g., when
1408 * mounting a device.
1409 */
1410 handle_bad_sector(bio);
1411 return 1;
1412 }
1413 }
1414
1415 return 0;
1416}
1417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001419 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 * @bio: The bio describing the location in memory and on the device.
1421 *
1422 * generic_make_request() is used to make I/O requests of block
1423 * devices. It is passed a &struct bio, which describes the I/O that needs
1424 * to be done.
1425 *
1426 * generic_make_request() does not return any status. The
1427 * success/failure status of the request, along with notification of
1428 * completion, is delivered asynchronously through the bio->bi_end_io
1429 * function described (one day) else where.
1430 *
1431 * The caller of generic_make_request must make sure that bi_io_vec
1432 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1433 * set to describe the device address, and the
1434 * bi_end_io and optionally bi_private are set to describe how
1435 * completion notification should be signaled.
1436 *
1437 * generic_make_request and the drivers it calls may use bi_next if this
1438 * bio happens to be merged with someone else, and may change bi_dev and
1439 * bi_sector for remaps as it sees fit. So the values of these fields
1440 * should NOT be depended on after the call to generic_make_request.
1441 */
Neil Brownd89d8792007-05-01 09:53:42 +02001442static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443{
Jens Axboe165125e2007-07-24 09:28:11 +02001444 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001445 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001447 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001448 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
Jens Axboec07e2b42007-07-18 13:27:58 +02001452 if (bio_check_eod(bio, nr_sectors))
1453 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455 /*
1456 * Resolve the mapping until finished. (drivers are
1457 * still free to implement/resolve their own stacking
1458 * by explicitly returning 0)
1459 *
1460 * NOTE: we don't repeat the blk_size check for each new device.
1461 * Stacking drivers are expected to know what they are doing.
1462 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001463 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001464 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 do {
1466 char b[BDEVNAME_SIZE];
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001467 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001470 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 printk(KERN_ERR
1472 "generic_make_request: Trying to access "
1473 "nonexistent block-device %s (%Lu)\n",
1474 bdevname(bio->bi_bdev, b),
1475 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001476 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 }
1478
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001479 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001480 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001481 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001482 bdevname(bio->bi_bdev, b),
1483 bio_sectors(bio),
1484 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 goto end_io;
1486 }
1487
Nick Pigginfde6ad22005-06-23 00:08:53 -07001488 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 goto end_io;
1490
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001491 part = bio->bi_bdev->bd_part;
1492 if (should_fail_request(part, bio->bi_size) ||
1493 should_fail_request(&part_to_disk(part)->part0,
1494 bio->bi_size))
Akinobu Mitac17bb492006-12-08 02:39:46 -08001495 goto end_io;
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 /*
1498 * If this device has partitions, remap block n
1499 * of partition p to block n+start(p) of the disk.
1500 */
1501 blk_partition_remap(bio);
1502
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001503 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1504 goto end_io;
1505
NeilBrown5ddfe962006-10-30 22:07:21 -08001506 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001507 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001508
NeilBrown5ddfe962006-10-30 22:07:21 -08001509 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001510 old_dev = bio->bi_bdev->bd_dev;
1511
Jens Axboec07e2b42007-07-18 13:27:58 +02001512 if (bio_check_eod(bio, nr_sectors))
1513 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001514
Tejun Heo1e879012010-09-03 11:56:17 +02001515 /*
1516 * Filter flush bio's early so that make_request based
1517 * drivers without flush support don't have to worry
1518 * about them.
1519 */
1520 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1521 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1522 if (!nr_sectors) {
1523 err = 0;
1524 goto end_io;
1525 }
1526 }
1527
Adrian Hunter8d57a982010-08-11 14:17:49 -07001528 if ((bio->bi_rw & REQ_DISCARD) &&
1529 (!blk_queue_discard(q) ||
1530 ((bio->bi_rw & REQ_SECURE) &&
1531 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001532 err = -EOPNOTSUPP;
1533 goto end_io;
1534 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001535
Vivek Goyalf469a7b2011-05-19 15:38:23 -04001536 if (blk_throtl_bio(q, &bio))
1537 goto end_io;
Vivek Goyale43473b2010-09-15 17:06:35 -04001538
1539 /*
1540 * If bio = NULL, bio has been throttled and will be submitted
1541 * later.
1542 */
1543 if (!bio)
1544 break;
1545
Minchan Kim01edede2009-09-08 21:56:38 +02001546 trace_block_bio_queue(q, bio);
1547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 ret = q->make_request_fn(q, bio);
1549 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001550
1551 return;
1552
1553end_io:
1554 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555}
1556
Neil Brownd89d8792007-05-01 09:53:42 +02001557/*
1558 * We only want one ->make_request_fn to be active at a time,
1559 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001560 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001561 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001562 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001563 * generic_make_request is currently active in this task or not.
1564 * If it is NULL, then no make_request is active. If it is non-NULL,
1565 * then a make_request is active, and new requests should be added
1566 * at the tail
1567 */
1568void generic_make_request(struct bio *bio)
1569{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001570 struct bio_list bio_list_on_stack;
1571
1572 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001573 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001574 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001575 return;
1576 }
1577 /* following loop may be a bit non-obvious, and so deserves some
1578 * explanation.
1579 * Before entering the loop, bio->bi_next is NULL (as all callers
1580 * ensure that) so we have a list with a single bio.
1581 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001582 * we assign bio_list to a pointer to the bio_list_on_stack,
1583 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001584 * added. __generic_make_request may indeed add some more bios
1585 * through a recursive call to generic_make_request. If it
1586 * did, we find a non-NULL value in bio_list and re-enter the loop
1587 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001588 * of the top of the list (no pretending) and so remove it from
1589 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001590 *
1591 * The loop was structured like this to make only one call to
1592 * __generic_make_request (which is important as it is large and
1593 * inlined) and to keep the structure simple.
1594 */
1595 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001596 bio_list_init(&bio_list_on_stack);
1597 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001598 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001599 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001600 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001601 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001602 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001603}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604EXPORT_SYMBOL(generic_make_request);
1605
1606/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001607 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1609 * @bio: The &struct bio which describes the I/O
1610 *
1611 * submit_bio() is very similar in purpose to generic_make_request(), and
1612 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001613 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 *
1615 */
1616void submit_bio(int rw, struct bio *bio)
1617{
1618 int count = bio_sectors(bio);
1619
Jens Axboe22e2c502005-06-27 10:55:12 +02001620 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Jens Axboebf2de6f2007-09-27 13:01:25 +02001622 /*
1623 * If it's a regular read/write or a barrier with data attached,
1624 * go through the normal accounting stuff before submission.
1625 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001626 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001627 if (rw & WRITE) {
1628 count_vm_events(PGPGOUT, count);
1629 } else {
1630 task_io_account_read(bio->bi_size);
1631 count_vm_events(PGPGIN, count);
1632 }
1633
1634 if (unlikely(block_dump)) {
1635 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001636 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001637 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001638 (rw & WRITE) ? "WRITE" : "READ",
1639 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001640 bdevname(bio->bi_bdev, b),
1641 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
1644
1645 generic_make_request(bio);
1646}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647EXPORT_SYMBOL(submit_bio);
1648
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001649/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001650 * blk_rq_check_limits - Helper function to check a request for the queue limit
1651 * @q: the queue
1652 * @rq: the request being checked
1653 *
1654 * Description:
1655 * @rq may have been made based on weaker limitations of upper-level queues
1656 * in request stacking drivers, and it may violate the limitation of @q.
1657 * Since the block layer and the underlying device driver trust @rq
1658 * after it is inserted to @q, it should be checked against @q before
1659 * the insertion using this generic function.
1660 *
1661 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001662 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001663 * Request stacking drivers like request-based dm may change the queue
1664 * limits while requests are in the queue (e.g. dm's table swapping).
1665 * Such request stacking drivers should check those requests agaist
1666 * the new queue limits again when they dispatch those requests,
1667 * although such checkings are also done against the old queue limits
1668 * when submitting requests.
1669 */
1670int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1671{
ike Snitzer33839772010-08-08 12:11:33 -04001672 if (rq->cmd_flags & REQ_DISCARD)
1673 return 0;
1674
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001675 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1676 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001677 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1678 return -EIO;
1679 }
1680
1681 /*
1682 * queue's settings related to segment counting like q->bounce_pfn
1683 * may differ from that of other stacking queues.
1684 * Recalculate it to check the request correctly on this queue's
1685 * limitation.
1686 */
1687 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001688 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001689 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1690 return -EIO;
1691 }
1692
1693 return 0;
1694}
1695EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1696
1697/**
1698 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1699 * @q: the queue to submit the request
1700 * @rq: the request being queued
1701 */
1702int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1703{
1704 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001705 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001706
1707 if (blk_rq_check_limits(q, rq))
1708 return -EIO;
1709
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001710 if (rq->rq_disk &&
1711 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001712 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001713
1714 spin_lock_irqsave(q->queue_lock, flags);
1715
1716 /*
1717 * Submitting request must be dequeued before calling this function
1718 * because it will be linked to another request_queue
1719 */
1720 BUG_ON(blk_queued_rq(rq));
1721
Jeff Moyer4853aba2011-08-15 21:37:25 +02001722 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1723 where = ELEVATOR_INSERT_FLUSH;
1724
1725 add_acct_request(q, rq, where);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001726 spin_unlock_irqrestore(q->queue_lock, flags);
1727
1728 return 0;
1729}
1730EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1731
Tejun Heo80a761f2009-07-03 17:48:17 +09001732/**
1733 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1734 * @rq: request to examine
1735 *
1736 * Description:
1737 * A request could be merge of IOs which require different failure
1738 * handling. This function determines the number of bytes which
1739 * can be failed from the beginning of the request without
1740 * crossing into area which need to be retried further.
1741 *
1742 * Return:
1743 * The number of bytes to fail.
1744 *
1745 * Context:
1746 * queue_lock must be held.
1747 */
1748unsigned int blk_rq_err_bytes(const struct request *rq)
1749{
1750 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1751 unsigned int bytes = 0;
1752 struct bio *bio;
1753
1754 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1755 return blk_rq_bytes(rq);
1756
1757 /*
1758 * Currently the only 'mixing' which can happen is between
1759 * different fastfail types. We can safely fail portions
1760 * which have all the failfast bits that the first one has -
1761 * the ones which are at least as eager to fail as the first
1762 * one.
1763 */
1764 for (bio = rq->bio; bio; bio = bio->bi_next) {
1765 if ((bio->bi_rw & ff) != ff)
1766 break;
1767 bytes += bio->bi_size;
1768 }
1769
1770 /* this could lead to infinite loop */
1771 BUG_ON(blk_rq_bytes(rq) && !bytes);
1772 return bytes;
1773}
1774EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1775
Jens Axboebc58ba92009-01-23 10:54:44 +01001776static void blk_account_io_completion(struct request *req, unsigned int bytes)
1777{
Jens Axboec2553b52009-04-24 08:10:11 +02001778 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001779 const int rw = rq_data_dir(req);
1780 struct hd_struct *part;
1781 int cpu;
1782
1783 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001784 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001785 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1786 part_stat_unlock();
1787 }
1788}
1789
1790static void blk_account_io_done(struct request *req)
1791{
Jens Axboebc58ba92009-01-23 10:54:44 +01001792 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001793 * Account IO completion. flush_rq isn't accounted as a
1794 * normal IO on queueing nor completion. Accounting the
1795 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001796 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001797 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001798 unsigned long duration = jiffies - req->start_time;
1799 const int rw = rq_data_dir(req);
1800 struct hd_struct *part;
1801 int cpu;
1802
1803 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001804 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001805
1806 part_stat_inc(cpu, part, ios[rw]);
1807 part_stat_add(cpu, part, ticks[rw], duration);
1808 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001809 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001810
Jens Axboe6c23a962011-01-07 08:43:37 +01001811 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001812 part_stat_unlock();
1813 }
1814}
1815
Tejun Heo53a08802008-12-03 12:41:26 +01001816/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001817 * blk_peek_request - peek at the top of a request queue
1818 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001819 *
1820 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001821 * Return the request at the top of @q. The returned request
1822 * should be started using blk_start_request() before LLD starts
1823 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001824 *
1825 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001826 * Pointer to the request at the top of @q if available. Null
1827 * otherwise.
1828 *
1829 * Context:
1830 * queue_lock must be held.
1831 */
1832struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001833{
1834 struct request *rq;
1835 int ret;
1836
1837 while ((rq = __elv_next_request(q)) != NULL) {
1838 if (!(rq->cmd_flags & REQ_STARTED)) {
1839 /*
1840 * This is the first time the device driver
1841 * sees this request (possibly after
1842 * requeueing). Notify IO scheduler.
1843 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001844 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001845 elv_activate_rq(q, rq);
1846
1847 /*
1848 * just mark as started even if we don't start
1849 * it, a request that has been delayed should
1850 * not be passed by new incoming requests
1851 */
1852 rq->cmd_flags |= REQ_STARTED;
1853 trace_block_rq_issue(q, rq);
1854 }
1855
1856 if (!q->boundary_rq || q->boundary_rq == rq) {
1857 q->end_sector = rq_end_sector(rq);
1858 q->boundary_rq = NULL;
1859 }
1860
1861 if (rq->cmd_flags & REQ_DONTPREP)
1862 break;
1863
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001864 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001865 /*
1866 * make sure space for the drain appears we
1867 * know we can do this because max_hw_segments
1868 * has been adjusted to be one fewer than the
1869 * device can handle
1870 */
1871 rq->nr_phys_segments++;
1872 }
1873
1874 if (!q->prep_rq_fn)
1875 break;
1876
1877 ret = q->prep_rq_fn(q, rq);
1878 if (ret == BLKPREP_OK) {
1879 break;
1880 } else if (ret == BLKPREP_DEFER) {
1881 /*
1882 * the request may have been (partially) prepped.
1883 * we need to keep this request in the front to
1884 * avoid resource deadlock. REQ_STARTED will
1885 * prevent other fs requests from passing this one.
1886 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001887 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001888 !(rq->cmd_flags & REQ_DONTPREP)) {
1889 /*
1890 * remove the space for the drain we added
1891 * so that we don't add it again
1892 */
1893 --rq->nr_phys_segments;
1894 }
1895
1896 rq = NULL;
1897 break;
1898 } else if (ret == BLKPREP_KILL) {
1899 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001900 /*
1901 * Mark this request as started so we don't trigger
1902 * any debug logic in the end I/O path.
1903 */
1904 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001905 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001906 } else {
1907 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1908 break;
1909 }
1910 }
1911
1912 return rq;
1913}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001914EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001915
Tejun Heo9934c8c2009-05-08 11:54:16 +09001916void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001917{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001918 struct request_queue *q = rq->q;
1919
Tejun Heo158dbda2009-04-23 11:05:18 +09001920 BUG_ON(list_empty(&rq->queuelist));
1921 BUG_ON(ELV_ON_HASH(rq));
1922
1923 list_del_init(&rq->queuelist);
1924
1925 /*
1926 * the time frame between a request being removed from the lists
1927 * and to it is freed is accounted as io that is in progress at
1928 * the driver side.
1929 */
Divyesh Shah91952912010-04-01 15:01:41 -07001930 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001931 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001932 set_io_start_time_ns(rq);
1933 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001934}
1935
Tejun Heo5efccd12009-04-23 11:05:18 +09001936/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001937 * blk_start_request - start request processing on the driver
1938 * @req: request to dequeue
1939 *
1940 * Description:
1941 * Dequeue @req and start timeout timer on it. This hands off the
1942 * request to the driver.
1943 *
1944 * Block internal functions which don't want to start timer should
1945 * call blk_dequeue_request().
1946 *
1947 * Context:
1948 * queue_lock must be held.
1949 */
1950void blk_start_request(struct request *req)
1951{
1952 blk_dequeue_request(req);
1953
1954 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001955 * We are now handing the request to the hardware, initialize
1956 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001957 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001958 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001959 if (unlikely(blk_bidi_rq(req)))
1960 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1961
Tejun Heo9934c8c2009-05-08 11:54:16 +09001962 blk_add_timer(req);
1963}
1964EXPORT_SYMBOL(blk_start_request);
1965
1966/**
1967 * blk_fetch_request - fetch a request from a request queue
1968 * @q: request queue to fetch a request from
1969 *
1970 * Description:
1971 * Return the request at the top of @q. The request is started on
1972 * return and LLD can start processing it immediately.
1973 *
1974 * Return:
1975 * Pointer to the request at the top of @q if available. Null
1976 * otherwise.
1977 *
1978 * Context:
1979 * queue_lock must be held.
1980 */
1981struct request *blk_fetch_request(struct request_queue *q)
1982{
1983 struct request *rq;
1984
1985 rq = blk_peek_request(q);
1986 if (rq)
1987 blk_start_request(rq);
1988 return rq;
1989}
1990EXPORT_SYMBOL(blk_fetch_request);
1991
1992/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001993 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001994 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001995 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001996 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001997 *
1998 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001999 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2000 * the request structure even if @req doesn't have leftover.
2001 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002002 *
2003 * This special helper function is only for request stacking drivers
2004 * (e.g. request-based dm) so that they can handle partial completion.
2005 * Actual device drivers should use blk_end_request instead.
2006 *
2007 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2008 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002009 *
2010 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002011 * %false - this request doesn't have any more data
2012 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002013 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002014bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002016 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 struct bio *bio;
2018
Tejun Heo2e60e022009-04-23 11:05:18 +09002019 if (!req->bio)
2020 return false;
2021
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002022 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002023
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002025 * For fs requests, rq is just carrier of independent bio's
2026 * and each partial completion should be handled separately.
2027 * Reset per-request error on each partial completion.
2028 *
2029 * TODO: tj: This is too subtle. It would be better to let
2030 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002032 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 req->errors = 0;
2034
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002035 if (error && req->cmd_type == REQ_TYPE_FS &&
2036 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002037 char *error_type;
2038
2039 switch (error) {
2040 case -ENOLINK:
2041 error_type = "recoverable transport";
2042 break;
2043 case -EREMOTEIO:
2044 error_type = "critical target";
2045 break;
2046 case -EBADE:
2047 error_type = "critical nexus";
2048 break;
2049 case -EIO:
2050 default:
2051 error_type = "I/O";
2052 break;
2053 }
2054 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2055 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2056 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 }
2058
Jens Axboebc58ba92009-01-23 10:54:44 +01002059 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 total_bytes = bio_nbytes = 0;
2062 while ((bio = req->bio) != NULL) {
2063 int nbytes;
2064
2065 if (nr_bytes >= bio->bi_size) {
2066 req->bio = bio->bi_next;
2067 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002068 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 next_idx = 0;
2070 bio_nbytes = 0;
2071 } else {
2072 int idx = bio->bi_idx + next_idx;
2073
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002074 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002076 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002077 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 break;
2079 }
2080
2081 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2082 BIO_BUG_ON(nbytes > bio->bi_size);
2083
2084 /*
2085 * not a complete bvec done
2086 */
2087 if (unlikely(nbytes > nr_bytes)) {
2088 bio_nbytes += nr_bytes;
2089 total_bytes += nr_bytes;
2090 break;
2091 }
2092
2093 /*
2094 * advance to the next vector
2095 */
2096 next_idx++;
2097 bio_nbytes += nbytes;
2098 }
2099
2100 total_bytes += nbytes;
2101 nr_bytes -= nbytes;
2102
Jens Axboe6728cb02008-01-31 13:03:55 +01002103 bio = req->bio;
2104 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 /*
2106 * end more in this run, or just return 'not-done'
2107 */
2108 if (unlikely(nr_bytes <= 0))
2109 break;
2110 }
2111 }
2112
2113 /*
2114 * completely done
2115 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002116 if (!req->bio) {
2117 /*
2118 * Reset counters so that the request stacking driver
2119 * can find how many bytes remain in the request
2120 * later.
2121 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002122 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002123 return false;
2124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
2126 /*
2127 * if the request wasn't completed, update state
2128 */
2129 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002130 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 bio->bi_idx += next_idx;
2132 bio_iovec(bio)->bv_offset += nr_bytes;
2133 bio_iovec(bio)->bv_len -= nr_bytes;
2134 }
2135
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002136 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002137 req->buffer = bio_data(req->bio);
2138
2139 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002140 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002141 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002142
Tejun Heo80a761f2009-07-03 17:48:17 +09002143 /* mixed attributes always follow the first bio */
2144 if (req->cmd_flags & REQ_MIXED_MERGE) {
2145 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2146 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2147 }
2148
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002149 /*
2150 * If total number of sectors is less than the first segment
2151 * size, something has gone terribly wrong.
2152 */
2153 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002154 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002155 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002156 }
2157
2158 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002160
Tejun Heo2e60e022009-04-23 11:05:18 +09002161 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162}
Tejun Heo2e60e022009-04-23 11:05:18 +09002163EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
Tejun Heo2e60e022009-04-23 11:05:18 +09002165static bool blk_update_bidi_request(struct request *rq, int error,
2166 unsigned int nr_bytes,
2167 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002168{
Tejun Heo2e60e022009-04-23 11:05:18 +09002169 if (blk_update_request(rq, error, nr_bytes))
2170 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002171
Tejun Heo2e60e022009-04-23 11:05:18 +09002172 /* Bidi request must be completed as a whole */
2173 if (unlikely(blk_bidi_rq(rq)) &&
2174 blk_update_request(rq->next_rq, error, bidi_bytes))
2175 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002176
Jens Axboee2e1a142010-06-09 10:42:09 +02002177 if (blk_queue_add_random(rq->q))
2178 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002179
2180 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181}
2182
James Bottomley28018c22010-07-01 19:49:17 +09002183/**
2184 * blk_unprep_request - unprepare a request
2185 * @req: the request
2186 *
2187 * This function makes a request ready for complete resubmission (or
2188 * completion). It happens only after all error handling is complete,
2189 * so represents the appropriate moment to deallocate any resources
2190 * that were allocated to the request in the prep_rq_fn. The queue
2191 * lock is held when calling this.
2192 */
2193void blk_unprep_request(struct request *req)
2194{
2195 struct request_queue *q = req->q;
2196
2197 req->cmd_flags &= ~REQ_DONTPREP;
2198 if (q->unprep_rq_fn)
2199 q->unprep_rq_fn(q, req);
2200}
2201EXPORT_SYMBOL_GPL(blk_unprep_request);
2202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203/*
2204 * queue lock must be held
2205 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002206static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002208 if (blk_rq_tagged(req))
2209 blk_queue_end_tag(req->q, req);
2210
James Bottomleyba396a62009-05-27 14:17:08 +02002211 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002213 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002214 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
Mike Andersone78042e2008-10-30 02:16:20 -07002216 blk_delete_timer(req);
2217
James Bottomley28018c22010-07-01 19:49:17 +09002218 if (req->cmd_flags & REQ_DONTPREP)
2219 blk_unprep_request(req);
2220
2221
Jens Axboebc58ba92009-01-23 10:54:44 +01002222 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002223
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002225 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002226 else {
2227 if (blk_bidi_rq(req))
2228 __blk_put_request(req->next_rq->q, req->next_rq);
2229
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 }
2232}
2233
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002234/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002235 * blk_end_bidi_request - Complete a bidi request
2236 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002237 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002238 * @nr_bytes: number of bytes to complete @rq
2239 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002240 *
2241 * Description:
2242 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002243 * Drivers that supports bidi can safely call this member for any
2244 * type of request, bidi or uni. In the later case @bidi_bytes is
2245 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002246 *
2247 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002248 * %false - we are done with this request
2249 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002250 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002251static bool blk_end_bidi_request(struct request *rq, int error,
2252 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002253{
2254 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002255 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002256
Tejun Heo2e60e022009-04-23 11:05:18 +09002257 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2258 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002259
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002260 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002261 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002262 spin_unlock_irqrestore(q->queue_lock, flags);
2263
Tejun Heo2e60e022009-04-23 11:05:18 +09002264 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002265}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002266
2267/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002268 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2269 * @rq: the request to complete
2270 * @error: %0 for success, < %0 for error
2271 * @nr_bytes: number of bytes to complete @rq
2272 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002273 *
2274 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002275 * Identical to blk_end_bidi_request() except that queue lock is
2276 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002277 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002278 * Return:
2279 * %false - we are done with this request
2280 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002281 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002282bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002283 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002284{
Tejun Heo2e60e022009-04-23 11:05:18 +09002285 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2286 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002287
Tejun Heo2e60e022009-04-23 11:05:18 +09002288 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002289
Tejun Heo2e60e022009-04-23 11:05:18 +09002290 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002291}
2292
2293/**
2294 * blk_end_request - Helper function for drivers to complete the request.
2295 * @rq: the request being processed
2296 * @error: %0 for success, < %0 for error
2297 * @nr_bytes: number of bytes to complete
2298 *
2299 * Description:
2300 * Ends I/O on a number of bytes attached to @rq.
2301 * If @rq has leftover, sets it up for the next range of segments.
2302 *
2303 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002304 * %false - we are done with this request
2305 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002306 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002307bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002308{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002309 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002310}
Jens Axboe56ad1742009-07-28 22:11:24 +02002311EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002312
2313/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002314 * blk_end_request_all - Helper function for drives to finish the request.
2315 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002316 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002317 *
2318 * Description:
2319 * Completely finish @rq.
2320 */
2321void blk_end_request_all(struct request *rq, int error)
2322{
2323 bool pending;
2324 unsigned int bidi_bytes = 0;
2325
2326 if (unlikely(blk_bidi_rq(rq)))
2327 bidi_bytes = blk_rq_bytes(rq->next_rq);
2328
2329 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2330 BUG_ON(pending);
2331}
Jens Axboe56ad1742009-07-28 22:11:24 +02002332EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002333
2334/**
2335 * blk_end_request_cur - Helper function to finish the current request chunk.
2336 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002337 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002338 *
2339 * Description:
2340 * Complete the current consecutively mapped chunk from @rq.
2341 *
2342 * Return:
2343 * %false - we are done with this request
2344 * %true - still buffers pending for this request
2345 */
2346bool blk_end_request_cur(struct request *rq, int error)
2347{
2348 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2349}
Jens Axboe56ad1742009-07-28 22:11:24 +02002350EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002351
2352/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002353 * blk_end_request_err - Finish a request till the next failure boundary.
2354 * @rq: the request to finish till the next failure boundary for
2355 * @error: must be negative errno
2356 *
2357 * Description:
2358 * Complete @rq till the next failure boundary.
2359 *
2360 * Return:
2361 * %false - we are done with this request
2362 * %true - still buffers pending for this request
2363 */
2364bool blk_end_request_err(struct request *rq, int error)
2365{
2366 WARN_ON(error >= 0);
2367 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2368}
2369EXPORT_SYMBOL_GPL(blk_end_request_err);
2370
2371/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002372 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002373 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002374 * @error: %0 for success, < %0 for error
2375 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002376 *
2377 * Description:
2378 * Must be called with queue lock held unlike blk_end_request().
2379 *
2380 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002381 * %false - we are done with this request
2382 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002383 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002384bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002385{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002386 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002387}
Jens Axboe56ad1742009-07-28 22:11:24 +02002388EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002389
2390/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002391 * __blk_end_request_all - Helper function for drives to finish the request.
2392 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002393 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002394 *
2395 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002396 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002397 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002399{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002400 bool pending;
2401 unsigned int bidi_bytes = 0;
2402
2403 if (unlikely(blk_bidi_rq(rq)))
2404 bidi_bytes = blk_rq_bytes(rq->next_rq);
2405
2406 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2407 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002408}
Jens Axboe56ad1742009-07-28 22:11:24 +02002409EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002410
2411/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002412 * __blk_end_request_cur - Helper function to finish the current request chunk.
2413 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002414 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002415 *
2416 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002417 * Complete the current consecutively mapped chunk from @rq. Must
2418 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002419 *
2420 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002421 * %false - we are done with this request
2422 * %true - still buffers pending for this request
2423 */
2424bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002425{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002426 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002427}
Jens Axboe56ad1742009-07-28 22:11:24 +02002428EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002429
Tejun Heo80a761f2009-07-03 17:48:17 +09002430/**
2431 * __blk_end_request_err - Finish a request till the next failure boundary.
2432 * @rq: the request to finish till the next failure boundary for
2433 * @error: must be negative errno
2434 *
2435 * Description:
2436 * Complete @rq till the next failure boundary. Must be called
2437 * with queue lock held.
2438 *
2439 * Return:
2440 * %false - we are done with this request
2441 * %true - still buffers pending for this request
2442 */
2443bool __blk_end_request_err(struct request *rq, int error)
2444{
2445 WARN_ON(error >= 0);
2446 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2447}
2448EXPORT_SYMBOL_GPL(__blk_end_request_err);
2449
Jens Axboe86db1e22008-01-29 14:53:40 +01002450void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2451 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002453 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002454 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
David Woodhousefb2dce82008-08-05 18:01:53 +01002456 if (bio_has_data(bio)) {
2457 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002458 rq->buffer = bio_data(bio);
2459 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002460 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462
NeilBrown66846572007-08-16 13:31:28 +02002463 if (bio->bi_bdev)
2464 rq->rq_disk = bio->bi_bdev->bd_disk;
2465}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002467#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2468/**
2469 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2470 * @rq: the request to be flushed
2471 *
2472 * Description:
2473 * Flush all pages in @rq.
2474 */
2475void rq_flush_dcache_pages(struct request *rq)
2476{
2477 struct req_iterator iter;
2478 struct bio_vec *bvec;
2479
2480 rq_for_each_segment(bvec, rq, iter)
2481 flush_dcache_page(bvec->bv_page);
2482}
2483EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2484#endif
2485
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002486/**
2487 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2488 * @q : the queue of the device being checked
2489 *
2490 * Description:
2491 * Check if underlying low-level drivers of a device are busy.
2492 * If the drivers want to export their busy state, they must set own
2493 * exporting function using blk_queue_lld_busy() first.
2494 *
2495 * Basically, this function is used only by request stacking drivers
2496 * to stop dispatching requests to underlying devices when underlying
2497 * devices are busy. This behavior helps more I/O merging on the queue
2498 * of the request stacking driver and prevents I/O throughput regression
2499 * on burst I/O load.
2500 *
2501 * Return:
2502 * 0 - Not busy (The request stacking driver should dispatch request)
2503 * 1 - Busy (The request stacking driver should stop dispatching request)
2504 */
2505int blk_lld_busy(struct request_queue *q)
2506{
2507 if (q->lld_busy_fn)
2508 return q->lld_busy_fn(q);
2509
2510 return 0;
2511}
2512EXPORT_SYMBOL_GPL(blk_lld_busy);
2513
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002514/**
2515 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2516 * @rq: the clone request to be cleaned up
2517 *
2518 * Description:
2519 * Free all bios in @rq for a cloned request.
2520 */
2521void blk_rq_unprep_clone(struct request *rq)
2522{
2523 struct bio *bio;
2524
2525 while ((bio = rq->bio) != NULL) {
2526 rq->bio = bio->bi_next;
2527
2528 bio_put(bio);
2529 }
2530}
2531EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2532
2533/*
2534 * Copy attributes of the original request to the clone request.
2535 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2536 */
2537static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2538{
2539 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002540 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002541 dst->cmd_type = src->cmd_type;
2542 dst->__sector = blk_rq_pos(src);
2543 dst->__data_len = blk_rq_bytes(src);
2544 dst->nr_phys_segments = src->nr_phys_segments;
2545 dst->ioprio = src->ioprio;
2546 dst->extra_len = src->extra_len;
2547}
2548
2549/**
2550 * blk_rq_prep_clone - Helper function to setup clone request
2551 * @rq: the request to be setup
2552 * @rq_src: original request to be cloned
2553 * @bs: bio_set that bios for clone are allocated from
2554 * @gfp_mask: memory allocation mask for bio
2555 * @bio_ctr: setup function to be called for each clone bio.
2556 * Returns %0 for success, non %0 for failure.
2557 * @data: private data to be passed to @bio_ctr
2558 *
2559 * Description:
2560 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2561 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2562 * are not copied, and copying such parts is the caller's responsibility.
2563 * Also, pages which the original bios are pointing to are not copied
2564 * and the cloned bios just point same pages.
2565 * So cloned bios must be completed before original bios, which means
2566 * the caller must complete @rq before @rq_src.
2567 */
2568int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2569 struct bio_set *bs, gfp_t gfp_mask,
2570 int (*bio_ctr)(struct bio *, struct bio *, void *),
2571 void *data)
2572{
2573 struct bio *bio, *bio_src;
2574
2575 if (!bs)
2576 bs = fs_bio_set;
2577
2578 blk_rq_init(NULL, rq);
2579
2580 __rq_for_each_bio(bio_src, rq_src) {
2581 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2582 if (!bio)
2583 goto free_and_out;
2584
2585 __bio_clone(bio, bio_src);
2586
2587 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002588 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002589 goto free_and_out;
2590
2591 if (bio_ctr && bio_ctr(bio, bio_src, data))
2592 goto free_and_out;
2593
2594 if (rq->bio) {
2595 rq->biotail->bi_next = bio;
2596 rq->biotail = bio;
2597 } else
2598 rq->bio = rq->biotail = bio;
2599 }
2600
2601 __blk_rq_prep_clone(rq, rq_src);
2602
2603 return 0;
2604
2605free_and_out:
2606 if (bio)
2607 bio_free(bio, bs);
2608 blk_rq_unprep_clone(rq);
2609
2610 return -ENOMEM;
2611}
2612EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2613
Jens Axboe18887ad2008-07-28 13:08:45 +02002614int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615{
2616 return queue_work(kblockd_workqueue, work);
2617}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618EXPORT_SYMBOL(kblockd_schedule_work);
2619
Vivek Goyale43473b2010-09-15 17:06:35 -04002620int kblockd_schedule_delayed_work(struct request_queue *q,
2621 struct delayed_work *dwork, unsigned long delay)
2622{
2623 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2624}
2625EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2626
Jens Axboe73c10102011-03-08 13:19:51 +01002627#define PLUG_MAGIC 0x91827364
2628
2629void blk_start_plug(struct blk_plug *plug)
2630{
2631 struct task_struct *tsk = current;
2632
2633 plug->magic = PLUG_MAGIC;
2634 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002635 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002636 plug->should_sort = 0;
Shaohua Li55c022b2011-07-08 08:19:20 +02002637 plug->count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002638
2639 /*
2640 * If this is a nested plug, don't actually assign it. It will be
2641 * flushed on its own.
2642 */
2643 if (!tsk->plug) {
2644 /*
2645 * Store ordering should not be needed here, since a potential
2646 * preempt will imply a full memory barrier
2647 */
2648 tsk->plug = plug;
2649 }
2650}
2651EXPORT_SYMBOL(blk_start_plug);
2652
2653static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2654{
2655 struct request *rqa = container_of(a, struct request, queuelist);
2656 struct request *rqb = container_of(b, struct request, queuelist);
2657
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002658 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002659}
2660
Jens Axboe49cac012011-04-16 13:51:05 +02002661/*
2662 * If 'from_schedule' is true, then postpone the dispatch of requests
2663 * until a safe kblockd context. We due this to avoid accidental big
2664 * additional stack usage in driver dispatch, in places where the originally
2665 * plugger did not intend it.
2666 */
Jens Axboef6603782011-04-15 15:49:07 +02002667static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002668 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002669 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002670{
Jens Axboe49cac012011-04-16 13:51:05 +02002671 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002672
2673 /*
2674 * If we are punting this to kblockd, then we can safely drop
2675 * the queue_lock before waking kblockd (which needs to take
2676 * this lock).
2677 */
2678 if (from_schedule) {
2679 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002680 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002681 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002682 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002683 spin_unlock(q->queue_lock);
2684 }
2685
Jens Axboe94b5eb22011-04-12 10:12:19 +02002686}
2687
NeilBrown048c9372011-04-18 09:52:22 +02002688static void flush_plug_callbacks(struct blk_plug *plug)
2689{
2690 LIST_HEAD(callbacks);
2691
2692 if (list_empty(&plug->cb_list))
2693 return;
2694
2695 list_splice_init(&plug->cb_list, &callbacks);
2696
2697 while (!list_empty(&callbacks)) {
2698 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2699 struct blk_plug_cb,
2700 list);
2701 list_del(&cb->list);
2702 cb->callback(cb);
2703 }
2704}
2705
Jens Axboe49cac012011-04-16 13:51:05 +02002706void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002707{
2708 struct request_queue *q;
2709 unsigned long flags;
2710 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002711 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002712 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002713
2714 BUG_ON(plug->magic != PLUG_MAGIC);
2715
NeilBrown048c9372011-04-18 09:52:22 +02002716 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002717 if (list_empty(&plug->list))
2718 return;
2719
NeilBrown109b8122011-04-11 14:13:10 +02002720 list_splice_init(&plug->list, &list);
Shaohua Li55c022b2011-07-08 08:19:20 +02002721 plug->count = 0;
NeilBrown109b8122011-04-11 14:13:10 +02002722
2723 if (plug->should_sort) {
2724 list_sort(NULL, &list, plug_rq_cmp);
2725 plug->should_sort = 0;
2726 }
Jens Axboe73c10102011-03-08 13:19:51 +01002727
2728 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002729 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002730
2731 /*
2732 * Save and disable interrupts here, to avoid doing it for every
2733 * queue lock we have to take.
2734 */
Jens Axboe73c10102011-03-08 13:19:51 +01002735 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002736 while (!list_empty(&list)) {
2737 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002738 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002739 BUG_ON(!rq->q);
2740 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002741 /*
2742 * This drops the queue lock
2743 */
2744 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002745 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002746 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002747 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002748 spin_lock(q->queue_lock);
2749 }
Jens Axboe73c10102011-03-08 13:19:51 +01002750 /*
2751 * rq is already accounted, so use raw insert
2752 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002753 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2754 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2755 else
2756 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002757
2758 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002759 }
2760
Jens Axboe99e22592011-04-18 09:59:55 +02002761 /*
2762 * This drops the queue lock
2763 */
2764 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002765 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002766
Jens Axboe73c10102011-03-08 13:19:51 +01002767 local_irq_restore(flags);
2768}
Jens Axboe73c10102011-03-08 13:19:51 +01002769
2770void blk_finish_plug(struct blk_plug *plug)
2771{
Jens Axboef6603782011-04-15 15:49:07 +02002772 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002773
2774 if (plug == current->plug)
2775 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002776}
2777EXPORT_SYMBOL(blk_finish_plug);
2778
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779int __init blk_dev_init(void)
2780{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002781 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2782 sizeof(((struct request *)0)->cmd_flags));
2783
Tejun Heo89b90be2011-01-03 15:01:47 +01002784 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2785 kblockd_workqueue = alloc_workqueue("kblockd",
2786 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 if (!kblockd_workqueue)
2788 panic("Failed to create kblockd\n");
2789
2790 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002791 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792
Jens Axboe8324aa92008-01-29 14:51:59 +01002793 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002794 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 return 0;
2797}