blob: d34433ae791781b5799edfa838f058922443e78d [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/*
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600351 * Note: If a driver supplied the queue lock, it is disconnected
352 * by this function. The actual state of the lock doesn't matter
353 * here as the request_queue isn't accessible after this point
354 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500355 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100356void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500357{
Jens Axboee3335de92008-09-18 09:22:54 -0700358 /*
359 * We know we have process context here, so we can be a little
360 * cautious and ensure that pending block actions on this device
361 * are done before moving on. Going into this function, we should
362 * not have processes doing IO to this device.
363 */
364 blk_sync_queue(q);
365
Matthew Garrett31373d02010-04-06 14:25:14 +0200366 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500367 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200368 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500369 mutex_unlock(&q->sysfs_lock);
370
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600371 if (q->queue_lock != &q->__queue_lock)
372 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500373
Al Viro483f4af2006-03-18 18:34:37 -0500374 blk_put_queue(q);
375}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376EXPORT_SYMBOL(blk_cleanup_queue);
377
Jens Axboe165125e2007-07-24 09:28:11 +0200378static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
380 struct request_list *rl = &q->rq;
381
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400382 if (unlikely(rl->rq_pool))
383 return 0;
384
Jens Axboe1faa16d2009-04-06 14:48:01 +0200385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200387 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Christoph Lameter19460892005-06-23 00:08:19 -0700391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 if (!rl->rq_pool)
395 return -ENOMEM;
396
397 return 0;
398}
399
Jens Axboe165125e2007-07-24 09:28:11 +0200400struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Christoph Lameter19460892005-06-23 00:08:19 -0700402 return blk_alloc_queue_node(gfp_mask, -1);
403}
404EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Jens Axboe165125e2007-07-24 09:28:11 +0200406struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700407{
Jens Axboe165125e2007-07-24 09:28:11 +0200408 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700409 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700410
Jens Axboe8324aa92008-01-29 14:51:59 +0100411 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700412 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (!q)
414 return NULL;
415
Jens Axboe0989a022009-06-12 14:42:56 +0200416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200420 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200421
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700422 err = bdi_init(&q->backing_dev_info);
423 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100424 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700425 return NULL;
426 }
427
Vivek Goyale43473b2010-09-15 17:06:35 -0400428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
430 return NULL;
431 }
432
Matthew Garrett31373d02010-04-06 14:25:14 +0200433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500441
Jens Axboe8324aa92008-01-29 14:51:59 +0100442 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Al Viro483f4af2006-03-18 18:34:37 -0500444 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700445 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500446
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500447 /*
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
450 */
451 q->queue_lock = &q->__queue_lock;
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 return q;
454}
Christoph Lameter19460892005-06-23 00:08:19 -0700455EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457/**
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
462 *
463 * Description:
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
472 *
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
477 *
478 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200482 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 * it didn't succeed.
484 *
485 * Note:
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
488 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700489
Jens Axboe165125e2007-07-24 09:28:11 +0200490struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Christoph Lameter19460892005-06-23 00:08:19 -0700492 return blk_init_queue_node(rfn, lock, -1);
493}
494EXPORT_SYMBOL(blk_init_queue);
495
Jens Axboe165125e2007-07-24 09:28:11 +0200496struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700497blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
498{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600499 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
502 if (!uninit_q)
503 return NULL;
504
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
506 if (!q)
507 blk_cleanup_queue(uninit_q);
508
509 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200510}
511EXPORT_SYMBOL(blk_init_queue_node);
512
513struct request_queue *
514blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
515 spinlock_t *lock)
516{
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
518}
519EXPORT_SYMBOL(blk_init_allocated_queue);
520
521struct request_queue *
522blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
524{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (!q)
526 return NULL;
527
Christoph Lameter19460892005-06-23 00:08:19 -0700528 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600529 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500530 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
532 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900534 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100535 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500536
537 /* Override internal queue lock with supplied lock pointer */
538 if (lock)
539 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Jens Axboef3b144a2009-03-06 08:48:33 +0100541 /*
542 * This also sets hw/phys segments, boundary and size
543 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Alan Stern44ec9542007-02-20 11:01:57 -0500546 q->sg_reserved_size = INT_MAX;
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 /*
549 * all done
550 */
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
553 return q;
554 }
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return NULL;
557}
Mike Snitzer01effb02010-05-11 08:57:42 +0200558EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Jens Axboe165125e2007-07-24 09:28:11 +0200560int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500563 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return 0;
565 }
566
567 return 1;
568}
Jens Axboed86e0e82011-05-27 07:44:43 +0200569EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jens Axboe165125e2007-07-24 09:28:11 +0200571static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200573 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200574 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 mempool_free(rq, q->rq.rq_pool);
576}
577
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200578static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200579blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
581 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
582
583 if (!rq)
584 return NULL;
585
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200586 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200587
Jerome Marchand42dad762009-04-22 14:01:49 +0200588 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Tejun Heocb98fc82005-10-28 08:29:39 +0200590 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200591 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200592 mempool_free(rq, q->rq.rq_pool);
593 return NULL;
594 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200595 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Tejun Heocb98fc82005-10-28 08:29:39 +0200598 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601/*
602 * ioc_batching returns true if the ioc is a valid batching request and
603 * should be given priority access to a request.
604 */
Jens Axboe165125e2007-07-24 09:28:11 +0200605static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
607 if (!ioc)
608 return 0;
609
610 /*
611 * Make sure the process is able to allocate at least 1 request
612 * even if the batch times out, otherwise we could theoretically
613 * lose wakeups.
614 */
615 return ioc->nr_batch_requests == q->nr_batching ||
616 (ioc->nr_batch_requests > 0
617 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
618}
619
620/*
621 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
622 * will cause the process to be a "batcher" on all queues in the system. This
623 * is the behaviour we want though - once it gets a wakeup it should be given
624 * a nice run.
625 */
Jens Axboe165125e2007-07-24 09:28:11 +0200626static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
628 if (!ioc || ioc_batching(q, ioc))
629 return;
630
631 ioc->nr_batch_requests = q->nr_batching;
632 ioc->last_waited = jiffies;
633}
634
Jens Axboe1faa16d2009-04-06 14:48:01 +0200635static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
637 struct request_list *rl = &q->rq;
638
Jens Axboe1faa16d2009-04-06 14:48:01 +0200639 if (rl->count[sync] < queue_congestion_off_threshold(q))
640 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Jens Axboe1faa16d2009-04-06 14:48:01 +0200642 if (rl->count[sync] + 1 <= q->nr_requests) {
643 if (waitqueue_active(&rl->wait[sync]))
644 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Jens Axboe1faa16d2009-04-06 14:48:01 +0200646 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 }
648}
649
650/*
651 * A request has just been released. Account for it, update the full and
652 * congestion status, wake up any waiters. Called under q->queue_lock.
653 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200654static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
656 struct request_list *rl = &q->rq;
657
Jens Axboe1faa16d2009-04-06 14:48:01 +0200658 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200659 if (priv)
660 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Jens Axboe1faa16d2009-04-06 14:48:01 +0200662 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Jens Axboe1faa16d2009-04-06 14:48:01 +0200664 if (unlikely(rl->starved[sync ^ 1]))
665 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666}
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100669 * Determine if elevator data should be initialized when allocating the
670 * request associated with @bio.
671 */
672static bool blk_rq_should_init_elevator(struct bio *bio)
673{
674 if (!bio)
675 return true;
676
677 /*
678 * Flush requests do not use the elevator so skip initialization.
679 * This allows a request to share the flush and elevator data.
680 */
681 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
682 return false;
683
684 return true;
685}
686
687/*
Nick Piggind6344532005-06-28 20:45:14 -0700688 * Get a free request, queue_lock must be held.
689 * Returns NULL on failure, with queue_lock held.
690 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 */
Jens Axboe165125e2007-07-24 09:28:11 +0200692static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100693 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
695 struct request *rq = NULL;
696 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100697 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200698 const bool is_sync = rw_is_sync(rw_flags) != 0;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100699 int may_queue, priv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Jens Axboe7749a8d2006-12-13 13:02:26 +0100701 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100702 if (may_queue == ELV_MQUEUE_NO)
703 goto rq_starved;
704
Jens Axboe1faa16d2009-04-06 14:48:01 +0200705 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
706 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200707 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100708 /*
709 * The queue will fill after this allocation, so set
710 * it as full, and mark this process as "batching".
711 * This process will be allowed to complete a batch of
712 * requests, others will be blocked.
713 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200714 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100715 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200716 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100717 } else {
718 if (may_queue != ELV_MQUEUE_MUST
719 && !ioc_batching(q, ioc)) {
720 /*
721 * The queue is full and the allocating
722 * process is not a "batcher", and not
723 * exempted by the IO scheduler
724 */
725 goto out;
726 }
727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200729 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731
Jens Axboe082cf692005-06-28 16:35:11 +0200732 /*
733 * Only allow batching queuers to allocate up to 50% over the defined
734 * limit of requests, otherwise we could have thousands of requests
735 * allocated with any setting of ->nr_requests
736 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200737 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200738 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100739
Jens Axboe1faa16d2009-04-06 14:48:01 +0200740 rl->count[is_sync]++;
741 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200742
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100743 if (blk_rq_should_init_elevator(bio)) {
744 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
745 if (priv)
746 rl->elvpriv++;
747 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200748
Jens Axboef253b862010-10-24 22:06:02 +0200749 if (blk_queue_io_stat(q))
750 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 spin_unlock_irq(q->queue_lock);
752
Jens Axboe7749a8d2006-12-13 13:02:26 +0100753 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100754 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /*
756 * Allocation failed presumably due to memory. Undo anything
757 * we might have messed up.
758 *
759 * Allocating task should really be put onto the front of the
760 * wait queue, but this is pretty rare.
761 */
762 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200763 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 /*
766 * in the very unlikely event that allocation failed and no
767 * requests for this direction was pending, mark us starved
768 * so that freeing of a request in the other direction will
769 * notice us. another possible fix would be to split the
770 * rq mempool into READ and WRITE
771 */
772rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200773 if (unlikely(rl->count[is_sync] == 0))
774 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 goto out;
777 }
778
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100779 /*
780 * ioc may be NULL here, and ioc_batching will be false. That's
781 * OK, if the queue is under the request limit then requests need
782 * not count toward the nr_batch_requests limit. There will always
783 * be some limit enforced by BLK_BATCH_TIME.
784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (ioc_batching(q, ioc))
786 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100787
Jens Axboe1faa16d2009-04-06 14:48:01 +0200788 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return rq;
791}
792
793/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100794 * No available requests for this queue, wait for some requests to become
795 * available.
Nick Piggind6344532005-06-28 20:45:14 -0700796 *
797 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Jens Axboe165125e2007-07-24 09:28:11 +0200799static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200800 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200802 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 struct request *rq;
804
Jens Axboe7749a8d2006-12-13 13:02:26 +0100805 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700806 while (!rq) {
807 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200808 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 struct request_list *rl = &q->rq;
810
Jens Axboe1faa16d2009-04-06 14:48:01 +0200811 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 TASK_UNINTERRUPTIBLE);
813
Jens Axboe1faa16d2009-04-06 14:48:01 +0200814 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200816 spin_unlock_irq(q->queue_lock);
817 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200819 /*
820 * After sleeping, we become a "batching" process and
821 * will be able to allocate at least one request, and
822 * up to a big batch of them for a small period time.
823 * See ioc_batching, ioc_set_batching
824 */
825 ioc = current_io_context(GFP_NOIO, q->node);
826 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100827
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200828 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200829 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200830
831 rq = get_request(q, rw_flags, bio, GFP_NOIO);
832 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
834 return rq;
835}
836
Jens Axboe165125e2007-07-24 09:28:11 +0200837struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
839 struct request *rq;
840
James Bottomleybfe159a2011-07-07 15:45:40 -0500841 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
842 return NULL;
843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 BUG_ON(rw != READ && rw != WRITE);
845
Nick Piggind6344532005-06-28 20:45:14 -0700846 spin_lock_irq(q->queue_lock);
847 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200848 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700849 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200850 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700851 if (!rq)
852 spin_unlock_irq(q->queue_lock);
853 }
854 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 return rq;
857}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858EXPORT_SYMBOL(blk_get_request);
859
860/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300861 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700862 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300863 * @bio: The bio describing the memory mappings that will be submitted for IO.
864 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700865 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200866 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300867 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
868 * type commands. Where the struct request needs to be farther initialized by
869 * the caller. It is passed a &struct bio, which describes the memory info of
870 * the I/O transfer.
871 *
872 * The caller of blk_make_request must make sure that bi_io_vec
873 * are set to describe the memory buffers. That bio_data_dir() will return
874 * the needed direction of the request. (And all bio's in the passed bio-chain
875 * are properly set accordingly)
876 *
877 * If called under none-sleepable conditions, mapped bio buffers must not
878 * need bouncing, by calling the appropriate masked or flagged allocator,
879 * suitable for the target device. Otherwise the call to blk_queue_bounce will
880 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200881 *
882 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
883 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
884 * anything but the first bio in the chain. Otherwise you risk waiting for IO
885 * completion of a bio that hasn't been submitted yet, thus resulting in a
886 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
887 * of bio_alloc(), as that avoids the mempool deadlock.
888 * If possible a big IO should be split into smaller parts when allocation
889 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200890 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300891struct request *blk_make_request(struct request_queue *q, struct bio *bio,
892 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200893{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300894 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
895
896 if (unlikely(!rq))
897 return ERR_PTR(-ENOMEM);
898
899 for_each_bio(bio) {
900 struct bio *bounce_bio = bio;
901 int ret;
902
903 blk_queue_bounce(q, &bounce_bio);
904 ret = blk_rq_append_bio(q, rq, bounce_bio);
905 if (unlikely(ret)) {
906 blk_put_request(rq);
907 return ERR_PTR(ret);
908 }
909 }
910
911 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200912}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300913EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200914
915/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 * blk_requeue_request - put a request back on queue
917 * @q: request queue where request should be inserted
918 * @rq: request to be inserted
919 *
920 * Description:
921 * Drivers often keep queueing requests until the hardware cannot accept
922 * more, when that condition happens we need to put the request back
923 * on the queue. Must be called with queue lock held.
924 */
Jens Axboe165125e2007-07-24 09:28:11 +0200925void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700927 blk_delete_timer(rq);
928 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100929 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 if (blk_rq_tagged(rq))
932 blk_queue_end_tag(q, rq);
933
James Bottomleyba396a62009-05-27 14:17:08 +0200934 BUG_ON(blk_queued_rq(rq));
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 elv_requeue_request(q, rq);
937}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938EXPORT_SYMBOL(blk_requeue_request);
939
Jens Axboe73c10102011-03-08 13:19:51 +0100940static void add_acct_request(struct request_queue *q, struct request *rq,
941 int where)
942{
943 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100944 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +0100945}
946
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200948 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 * @q: request queue where request should be inserted
950 * @rq: request to be inserted
951 * @at_head: insert request at head or tail of queue
952 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 *
954 * Description:
955 * Many block devices need to execute commands asynchronously, so they don't
956 * block the whole kernel from preemption during request execution. This is
957 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200958 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
959 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 *
961 * We have the option of inserting the head or the tail of the queue.
962 * Typically we use the tail for new ioctls and so forth. We use the head
963 * of the queue for things like a QUEUE_FULL message from a device, or a
964 * host that is unable to accept a particular command.
965 */
Jens Axboe165125e2007-07-24 09:28:11 +0200966void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500967 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
Tejun Heo 867d1192005-04-24 02:06:05 -0500969 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 unsigned long flags;
971
972 /*
973 * tell I/O scheduler that this isn't a regular read/write (ie it
974 * must not attempt merges on this) and that it acts as a soft
975 * barrier
976 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200977 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 rq->special = data;
980
981 spin_lock_irqsave(q->queue_lock, flags);
982
983 /*
984 * If command is tagged, release the tag
985 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500986 if (blk_rq_tagged(rq))
987 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Jens Axboe73c10102011-03-08 13:19:51 +0100989 add_acct_request(q, rq, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200990 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 spin_unlock_irqrestore(q->queue_lock, flags);
992}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993EXPORT_SYMBOL(blk_insert_request);
994
Tejun Heo074a7ac2008-08-25 19:56:14 +0900995static void part_round_stats_single(int cpu, struct hd_struct *part,
996 unsigned long now)
997{
998 if (now == part->stamp)
999 return;
1000
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001001 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001002 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001003 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001004 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1005 }
1006 part->stamp = now;
1007}
1008
1009/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001010 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1011 * @cpu: cpu number for stats access
1012 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 *
1014 * The average IO queue length and utilisation statistics are maintained
1015 * by observing the current state of the queue length and the amount of
1016 * time it has been in this state for.
1017 *
1018 * Normally, that accounting is done on IO completion, but that can result
1019 * in more than a second's worth of IO being accounted for within any one
1020 * second, leading to >100% utilisation. To deal with that, we call this
1021 * function to do a round-off before returning the results when reading
1022 * /proc/diskstats. This accounts immediately for all queue usage up to
1023 * the current jiffies and restarts the counters again.
1024 */
Tejun Heoc9959052008-08-25 19:47:21 +09001025void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001026{
1027 unsigned long now = jiffies;
1028
Tejun Heo074a7ac2008-08-25 19:56:14 +09001029 if (part->partno)
1030 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1031 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001032}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001033EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035/*
1036 * queue lock must be held
1037 */
Jens Axboe165125e2007-07-24 09:28:11 +02001038void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 if (unlikely(!q))
1041 return;
1042 if (unlikely(--req->ref_count))
1043 return;
1044
Tejun Heo8922e162005-10-20 16:23:44 +02001045 elv_completed_request(q, req);
1046
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001047 /* this is a bio leak */
1048 WARN_ON(req->bio != NULL);
1049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /*
1051 * Request may not have originated from ll_rw_blk. if not,
1052 * it didn't come out of our reserved rq pools
1053 */
Jens Axboe49171e52006-08-10 08:59:11 +02001054 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001055 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001056 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001059 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
1061 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001062 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
1064}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001065EXPORT_SYMBOL_GPL(__blk_put_request);
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067void blk_put_request(struct request *req)
1068{
Tejun Heo8922e162005-10-20 16:23:44 +02001069 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001070 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001072 spin_lock_irqsave(q->queue_lock, flags);
1073 __blk_put_request(q, req);
1074 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076EXPORT_SYMBOL(blk_put_request);
1077
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001078/**
1079 * blk_add_request_payload - add a payload to a request
1080 * @rq: request to update
1081 * @page: page backing the payload
1082 * @len: length of the payload.
1083 *
1084 * This allows to later add a payload to an already submitted request by
1085 * a block driver. The driver needs to take care of freeing the payload
1086 * itself.
1087 *
1088 * Note that this is a quite horrible hack and nothing but handling of
1089 * discard requests should ever use it.
1090 */
1091void blk_add_request_payload(struct request *rq, struct page *page,
1092 unsigned int len)
1093{
1094 struct bio *bio = rq->bio;
1095
1096 bio->bi_io_vec->bv_page = page;
1097 bio->bi_io_vec->bv_offset = 0;
1098 bio->bi_io_vec->bv_len = len;
1099
1100 bio->bi_size = len;
1101 bio->bi_vcnt = 1;
1102 bio->bi_phys_segments = 1;
1103
1104 rq->__data_len = rq->resid_len = len;
1105 rq->nr_phys_segments = 1;
1106 rq->buffer = bio_data(bio);
1107}
1108EXPORT_SYMBOL_GPL(blk_add_request_payload);
1109
Jens Axboe73c10102011-03-08 13:19:51 +01001110static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1111 struct bio *bio)
1112{
1113 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1114
Jens Axboe73c10102011-03-08 13:19:51 +01001115 if (!ll_back_merge_fn(q, req, bio))
1116 return false;
1117
1118 trace_block_bio_backmerge(q, bio);
1119
1120 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1121 blk_rq_set_mixed_merge(req);
1122
1123 req->biotail->bi_next = bio;
1124 req->biotail = bio;
1125 req->__data_len += bio->bi_size;
1126 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1127
1128 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001129 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001130 return true;
1131}
1132
1133static bool bio_attempt_front_merge(struct request_queue *q,
1134 struct request *req, struct bio *bio)
1135{
1136 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001137
Jens Axboe73c10102011-03-08 13:19:51 +01001138 if (!ll_front_merge_fn(q, req, bio))
1139 return false;
1140
1141 trace_block_bio_frontmerge(q, bio);
1142
1143 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1144 blk_rq_set_mixed_merge(req);
1145
Jens Axboe73c10102011-03-08 13:19:51 +01001146 bio->bi_next = req->bio;
1147 req->bio = bio;
1148
1149 /*
1150 * may not be valid. if the low level driver said
1151 * it didn't need a bounce buffer then it better
1152 * not touch req->buffer either...
1153 */
1154 req->buffer = bio_data(bio);
1155 req->__sector = bio->bi_sector;
1156 req->__data_len += bio->bi_size;
1157 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1158
1159 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001160 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001161 return true;
1162}
1163
1164/*
1165 * Attempts to merge with the plugged list in the current process. Returns
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001166 * true if merge was successful, otherwise false.
Jens Axboe73c10102011-03-08 13:19:51 +01001167 */
1168static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001169 struct bio *bio, unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001170{
1171 struct blk_plug *plug;
1172 struct request *rq;
1173 bool ret = false;
1174
1175 plug = tsk->plug;
1176 if (!plug)
1177 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001178 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001179
1180 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1181 int el_ret;
1182
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001183 (*request_count)++;
1184
Jens Axboe73c10102011-03-08 13:19:51 +01001185 if (rq->q != q)
1186 continue;
1187
1188 el_ret = elv_try_merge(rq, bio);
1189 if (el_ret == ELEVATOR_BACK_MERGE) {
1190 ret = bio_attempt_back_merge(q, rq, bio);
1191 if (ret)
1192 break;
1193 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1194 ret = bio_attempt_front_merge(q, rq, bio);
1195 if (ret)
1196 break;
1197 }
1198 }
1199out:
1200 return ret;
1201}
1202
Jens Axboe86db1e22008-01-29 14:53:40 +01001203void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001204{
Jens Axboec7c22e42008-09-13 20:26:01 +02001205 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001206 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001207
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001208 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1209 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001210 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001211
Tejun Heo52d9e672006-01-06 09:49:58 +01001212 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001213 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001214 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001215 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001216}
1217
Jens Axboe165125e2007-07-24 09:28:11 +02001218static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001220 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001221 struct blk_plug *plug;
1222 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1223 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001224 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 /*
1227 * low level driver can indicate that it wants pages above a
1228 * certain limit bounced to low memory (ie for highmem, or even
1229 * ISA dma in theory)
1230 */
1231 blk_queue_bounce(q, &bio);
1232
Tejun Heo4fed9472010-09-03 11:56:17 +02001233 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001234 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001235 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001236 goto get_rq;
1237 }
1238
Jens Axboe73c10102011-03-08 13:19:51 +01001239 /*
1240 * Check if we can merge with the plugged list before grabbing
1241 * any locks.
1242 */
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001243 if (attempt_plug_merge(current, q, bio, &request_count))
Jens Axboe73c10102011-03-08 13:19:51 +01001244 goto out;
1245
1246 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
1248 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001249 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001250 if (bio_attempt_back_merge(q, req, bio)) {
1251 if (!attempt_back_merge(q, req))
1252 elv_merged_request(q, req, el_ret);
1253 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001254 }
Jens Axboe73c10102011-03-08 13:19:51 +01001255 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001256 if (bio_attempt_front_merge(q, req, bio)) {
1257 if (!attempt_front_merge(q, req))
1258 elv_merged_request(q, req, el_ret);
1259 goto out_unlock;
1260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 }
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001264 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001265 * This sync check and mask will be re-done in init_request_from_bio(),
1266 * but we need to set it earlier to expose the sync flag to the
1267 * rq allocator and io schedulers.
1268 */
1269 rw_flags = bio_data_dir(bio);
1270 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001271 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001272
1273 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001274 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001275 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001276 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001277 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001278
Nick Piggin450991b2005-06-28 20:45:13 -07001279 /*
1280 * After dropping the lock and possibly sleeping here, our request
1281 * may now be mergeable after it had proven unmergeable (above).
1282 * We don't worry about that case for efficiency. It won't happen
1283 * often, and the elevators are able to handle it.
1284 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001285 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Jens Axboec7c22e42008-09-13 20:26:01 +02001287 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Dan Williams5757a6d2011-07-23 20:44:25 +02001288 bio_flagged(bio, BIO_CPU_AFFINE))
Jens Axboe11ccf112011-07-26 15:01:15 +02001289 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001290
Jens Axboe73c10102011-03-08 13:19:51 +01001291 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001292 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001293 /*
1294 * If this is the first request added after a plug, fire
1295 * of a plug trace. If others have been added before, check
1296 * if we have multiple devices in this plug. If so, make a
1297 * note to sort the list before dispatch.
1298 */
1299 if (list_empty(&plug->list))
1300 trace_block_plug(q);
1301 else if (!plug->should_sort) {
Jens Axboe73c10102011-03-08 13:19:51 +01001302 struct request *__rq;
1303
1304 __rq = list_entry_rq(plug->list.prev);
1305 if (__rq->q != q)
1306 plug->should_sort = 1;
1307 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001308 if (request_count >= BLK_MAX_REQUEST_COUNT)
Shaohua Li55c022b2011-07-08 08:19:20 +02001309 blk_flush_plug_list(plug, false);
Shaohua Lia6327162011-08-24 16:04:32 +02001310 list_add_tail(&req->queuelist, &plug->list);
1311 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001312 } else {
1313 spin_lock_irq(q->queue_lock);
1314 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001315 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001316out_unlock:
1317 spin_unlock_irq(q->queue_lock);
1318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321}
1322
1323/*
1324 * If bio->bi_dev is a partition, remap the location
1325 */
1326static inline void blk_partition_remap(struct bio *bio)
1327{
1328 struct block_device *bdev = bio->bi_bdev;
1329
Jens Axboebf2de6f2007-09-27 13:01:25 +02001330 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 bio->bi_sector += p->start_sect;
1334 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001335
Mike Snitzerd07335e2010-11-16 12:52:38 +01001336 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1337 bdev->bd_dev,
1338 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340}
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342static void handle_bad_sector(struct bio *bio)
1343{
1344 char b[BDEVNAME_SIZE];
1345
1346 printk(KERN_INFO "attempt to access beyond end of device\n");
1347 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1348 bdevname(bio->bi_bdev, b),
1349 bio->bi_rw,
1350 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001351 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
1353 set_bit(BIO_EOF, &bio->bi_flags);
1354}
1355
Akinobu Mitac17bb492006-12-08 02:39:46 -08001356#ifdef CONFIG_FAIL_MAKE_REQUEST
1357
1358static DECLARE_FAULT_ATTR(fail_make_request);
1359
1360static int __init setup_fail_make_request(char *str)
1361{
1362 return setup_fault_attr(&fail_make_request, str);
1363}
1364__setup("fail_make_request=", setup_fail_make_request);
1365
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001366static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001367{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001368 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001369}
1370
1371static int __init fail_make_request_debugfs(void)
1372{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001373 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1374 NULL, &fail_make_request);
1375
1376 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001377}
1378
1379late_initcall(fail_make_request_debugfs);
1380
1381#else /* CONFIG_FAIL_MAKE_REQUEST */
1382
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001383static inline bool should_fail_request(struct hd_struct *part,
1384 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001385{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001386 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001387}
1388
1389#endif /* CONFIG_FAIL_MAKE_REQUEST */
1390
Jens Axboec07e2b42007-07-18 13:27:58 +02001391/*
1392 * Check whether this bio extends beyond the end of the device.
1393 */
1394static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1395{
1396 sector_t maxsector;
1397
1398 if (!nr_sectors)
1399 return 0;
1400
1401 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001402 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001403 if (maxsector) {
1404 sector_t sector = bio->bi_sector;
1405
1406 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1407 /*
1408 * This may well happen - the kernel calls bread()
1409 * without checking the size of the device, e.g., when
1410 * mounting a device.
1411 */
1412 handle_bad_sector(bio);
1413 return 1;
1414 }
1415 }
1416
1417 return 0;
1418}
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001421 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 * @bio: The bio describing the location in memory and on the device.
1423 *
1424 * generic_make_request() is used to make I/O requests of block
1425 * devices. It is passed a &struct bio, which describes the I/O that needs
1426 * to be done.
1427 *
1428 * generic_make_request() does not return any status. The
1429 * success/failure status of the request, along with notification of
1430 * completion, is delivered asynchronously through the bio->bi_end_io
1431 * function described (one day) else where.
1432 *
1433 * The caller of generic_make_request must make sure that bi_io_vec
1434 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1435 * set to describe the device address, and the
1436 * bi_end_io and optionally bi_private are set to describe how
1437 * completion notification should be signaled.
1438 *
1439 * generic_make_request and the drivers it calls may use bi_next if this
1440 * bio happens to be merged with someone else, and may change bi_dev and
1441 * bi_sector for remaps as it sees fit. So the values of these fields
1442 * should NOT be depended on after the call to generic_make_request.
1443 */
Neil Brownd89d8792007-05-01 09:53:42 +02001444static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
Jens Axboe165125e2007-07-24 09:28:11 +02001446 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001447 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001449 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001450 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
1452 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Jens Axboec07e2b42007-07-18 13:27:58 +02001454 if (bio_check_eod(bio, nr_sectors))
1455 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 /*
1458 * Resolve the mapping until finished. (drivers are
1459 * still free to implement/resolve their own stacking
1460 * by explicitly returning 0)
1461 *
1462 * NOTE: we don't repeat the blk_size check for each new device.
1463 * Stacking drivers are expected to know what they are doing.
1464 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001465 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001466 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 do {
1468 char b[BDEVNAME_SIZE];
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001469 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
1471 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001472 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 printk(KERN_ERR
1474 "generic_make_request: Trying to access "
1475 "nonexistent block-device %s (%Lu)\n",
1476 bdevname(bio->bi_bdev, b),
1477 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001478 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 }
1480
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001481 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001482 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001483 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001484 bdevname(bio->bi_bdev, b),
1485 bio_sectors(bio),
1486 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 goto end_io;
1488 }
1489
Nick Pigginfde6ad22005-06-23 00:08:53 -07001490 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 goto end_io;
1492
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001493 part = bio->bi_bdev->bd_part;
1494 if (should_fail_request(part, bio->bi_size) ||
1495 should_fail_request(&part_to_disk(part)->part0,
1496 bio->bi_size))
Akinobu Mitac17bb492006-12-08 02:39:46 -08001497 goto end_io;
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 /*
1500 * If this device has partitions, remap block n
1501 * of partition p to block n+start(p) of the disk.
1502 */
1503 blk_partition_remap(bio);
1504
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001505 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1506 goto end_io;
1507
NeilBrown5ddfe962006-10-30 22:07:21 -08001508 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001509 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001510
NeilBrown5ddfe962006-10-30 22:07:21 -08001511 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001512 old_dev = bio->bi_bdev->bd_dev;
1513
Jens Axboec07e2b42007-07-18 13:27:58 +02001514 if (bio_check_eod(bio, nr_sectors))
1515 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001516
Tejun Heo1e879012010-09-03 11:56:17 +02001517 /*
1518 * Filter flush bio's early so that make_request based
1519 * drivers without flush support don't have to worry
1520 * about them.
1521 */
1522 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1523 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1524 if (!nr_sectors) {
1525 err = 0;
1526 goto end_io;
1527 }
1528 }
1529
Adrian Hunter8d57a982010-08-11 14:17:49 -07001530 if ((bio->bi_rw & REQ_DISCARD) &&
1531 (!blk_queue_discard(q) ||
1532 ((bio->bi_rw & REQ_SECURE) &&
1533 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001534 err = -EOPNOTSUPP;
1535 goto end_io;
1536 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001537
Vivek Goyalf469a7b2011-05-19 15:38:23 -04001538 if (blk_throtl_bio(q, &bio))
1539 goto end_io;
Vivek Goyale43473b2010-09-15 17:06:35 -04001540
1541 /*
1542 * If bio = NULL, bio has been throttled and will be submitted
1543 * later.
1544 */
1545 if (!bio)
1546 break;
1547
Minchan Kim01edede2009-09-08 21:56:38 +02001548 trace_block_bio_queue(q, bio);
1549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 ret = q->make_request_fn(q, bio);
1551 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001552
1553 return;
1554
1555end_io:
1556 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557}
1558
Neil Brownd89d8792007-05-01 09:53:42 +02001559/*
1560 * We only want one ->make_request_fn to be active at a time,
1561 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001562 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001563 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001564 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001565 * generic_make_request is currently active in this task or not.
1566 * If it is NULL, then no make_request is active. If it is non-NULL,
1567 * then a make_request is active, and new requests should be added
1568 * at the tail
1569 */
1570void generic_make_request(struct bio *bio)
1571{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001572 struct bio_list bio_list_on_stack;
1573
1574 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001575 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001577 return;
1578 }
1579 /* following loop may be a bit non-obvious, and so deserves some
1580 * explanation.
1581 * Before entering the loop, bio->bi_next is NULL (as all callers
1582 * ensure that) so we have a list with a single bio.
1583 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001584 * we assign bio_list to a pointer to the bio_list_on_stack,
1585 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001586 * added. __generic_make_request may indeed add some more bios
1587 * through a recursive call to generic_make_request. If it
1588 * did, we find a non-NULL value in bio_list and re-enter the loop
1589 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001590 * of the top of the list (no pretending) and so remove it from
1591 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001592 *
1593 * The loop was structured like this to make only one call to
1594 * __generic_make_request (which is important as it is large and
1595 * inlined) and to keep the structure simple.
1596 */
1597 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001598 bio_list_init(&bio_list_on_stack);
1599 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001600 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001601 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001602 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001603 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001604 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001605}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606EXPORT_SYMBOL(generic_make_request);
1607
1608/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001609 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1611 * @bio: The &struct bio which describes the I/O
1612 *
1613 * submit_bio() is very similar in purpose to generic_make_request(), and
1614 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001615 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 *
1617 */
1618void submit_bio(int rw, struct bio *bio)
1619{
1620 int count = bio_sectors(bio);
1621
Jens Axboe22e2c502005-06-27 10:55:12 +02001622 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Jens Axboebf2de6f2007-09-27 13:01:25 +02001624 /*
1625 * If it's a regular read/write or a barrier with data attached,
1626 * go through the normal accounting stuff before submission.
1627 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001628 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001629 if (rw & WRITE) {
1630 count_vm_events(PGPGOUT, count);
1631 } else {
1632 task_io_account_read(bio->bi_size);
1633 count_vm_events(PGPGIN, count);
1634 }
1635
1636 if (unlikely(block_dump)) {
1637 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001638 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001639 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001640 (rw & WRITE) ? "WRITE" : "READ",
1641 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001642 bdevname(bio->bi_bdev, b),
1643 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 }
1646
1647 generic_make_request(bio);
1648}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649EXPORT_SYMBOL(submit_bio);
1650
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001651/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001652 * blk_rq_check_limits - Helper function to check a request for the queue limit
1653 * @q: the queue
1654 * @rq: the request being checked
1655 *
1656 * Description:
1657 * @rq may have been made based on weaker limitations of upper-level queues
1658 * in request stacking drivers, and it may violate the limitation of @q.
1659 * Since the block layer and the underlying device driver trust @rq
1660 * after it is inserted to @q, it should be checked against @q before
1661 * the insertion using this generic function.
1662 *
1663 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001664 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001665 * Request stacking drivers like request-based dm may change the queue
1666 * limits while requests are in the queue (e.g. dm's table swapping).
1667 * Such request stacking drivers should check those requests agaist
1668 * the new queue limits again when they dispatch those requests,
1669 * although such checkings are also done against the old queue limits
1670 * when submitting requests.
1671 */
1672int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1673{
ike Snitzer33839772010-08-08 12:11:33 -04001674 if (rq->cmd_flags & REQ_DISCARD)
1675 return 0;
1676
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001677 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1678 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001679 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1680 return -EIO;
1681 }
1682
1683 /*
1684 * queue's settings related to segment counting like q->bounce_pfn
1685 * may differ from that of other stacking queues.
1686 * Recalculate it to check the request correctly on this queue's
1687 * limitation.
1688 */
1689 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001690 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001691 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1692 return -EIO;
1693 }
1694
1695 return 0;
1696}
1697EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1698
1699/**
1700 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1701 * @q: the queue to submit the request
1702 * @rq: the request being queued
1703 */
1704int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1705{
1706 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001707 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001708
1709 if (blk_rq_check_limits(q, rq))
1710 return -EIO;
1711
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001712 if (rq->rq_disk &&
1713 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001714 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001715
1716 spin_lock_irqsave(q->queue_lock, flags);
1717
1718 /*
1719 * Submitting request must be dequeued before calling this function
1720 * because it will be linked to another request_queue
1721 */
1722 BUG_ON(blk_queued_rq(rq));
1723
Jeff Moyer4853aba2011-08-15 21:37:25 +02001724 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1725 where = ELEVATOR_INSERT_FLUSH;
1726
1727 add_acct_request(q, rq, where);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001728 spin_unlock_irqrestore(q->queue_lock, flags);
1729
1730 return 0;
1731}
1732EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1733
Tejun Heo80a761f2009-07-03 17:48:17 +09001734/**
1735 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1736 * @rq: request to examine
1737 *
1738 * Description:
1739 * A request could be merge of IOs which require different failure
1740 * handling. This function determines the number of bytes which
1741 * can be failed from the beginning of the request without
1742 * crossing into area which need to be retried further.
1743 *
1744 * Return:
1745 * The number of bytes to fail.
1746 *
1747 * Context:
1748 * queue_lock must be held.
1749 */
1750unsigned int blk_rq_err_bytes(const struct request *rq)
1751{
1752 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1753 unsigned int bytes = 0;
1754 struct bio *bio;
1755
1756 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1757 return blk_rq_bytes(rq);
1758
1759 /*
1760 * Currently the only 'mixing' which can happen is between
1761 * different fastfail types. We can safely fail portions
1762 * which have all the failfast bits that the first one has -
1763 * the ones which are at least as eager to fail as the first
1764 * one.
1765 */
1766 for (bio = rq->bio; bio; bio = bio->bi_next) {
1767 if ((bio->bi_rw & ff) != ff)
1768 break;
1769 bytes += bio->bi_size;
1770 }
1771
1772 /* this could lead to infinite loop */
1773 BUG_ON(blk_rq_bytes(rq) && !bytes);
1774 return bytes;
1775}
1776EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1777
Jens Axboebc58ba92009-01-23 10:54:44 +01001778static void blk_account_io_completion(struct request *req, unsigned int bytes)
1779{
Jens Axboec2553b52009-04-24 08:10:11 +02001780 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001781 const int rw = rq_data_dir(req);
1782 struct hd_struct *part;
1783 int cpu;
1784
1785 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001786 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001787 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1788 part_stat_unlock();
1789 }
1790}
1791
1792static void blk_account_io_done(struct request *req)
1793{
Jens Axboebc58ba92009-01-23 10:54:44 +01001794 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001795 * Account IO completion. flush_rq isn't accounted as a
1796 * normal IO on queueing nor completion. Accounting the
1797 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001798 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001799 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001800 unsigned long duration = jiffies - req->start_time;
1801 const int rw = rq_data_dir(req);
1802 struct hd_struct *part;
1803 int cpu;
1804
1805 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001806 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001807
1808 part_stat_inc(cpu, part, ios[rw]);
1809 part_stat_add(cpu, part, ticks[rw], duration);
1810 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001811 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001812
Jens Axboe6c23a962011-01-07 08:43:37 +01001813 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001814 part_stat_unlock();
1815 }
1816}
1817
Tejun Heo53a08802008-12-03 12:41:26 +01001818/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001819 * blk_peek_request - peek at the top of a request queue
1820 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001821 *
1822 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001823 * Return the request at the top of @q. The returned request
1824 * should be started using blk_start_request() before LLD starts
1825 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001826 *
1827 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001828 * Pointer to the request at the top of @q if available. Null
1829 * otherwise.
1830 *
1831 * Context:
1832 * queue_lock must be held.
1833 */
1834struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001835{
1836 struct request *rq;
1837 int ret;
1838
1839 while ((rq = __elv_next_request(q)) != NULL) {
1840 if (!(rq->cmd_flags & REQ_STARTED)) {
1841 /*
1842 * This is the first time the device driver
1843 * sees this request (possibly after
1844 * requeueing). Notify IO scheduler.
1845 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001846 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001847 elv_activate_rq(q, rq);
1848
1849 /*
1850 * just mark as started even if we don't start
1851 * it, a request that has been delayed should
1852 * not be passed by new incoming requests
1853 */
1854 rq->cmd_flags |= REQ_STARTED;
1855 trace_block_rq_issue(q, rq);
1856 }
1857
1858 if (!q->boundary_rq || q->boundary_rq == rq) {
1859 q->end_sector = rq_end_sector(rq);
1860 q->boundary_rq = NULL;
1861 }
1862
1863 if (rq->cmd_flags & REQ_DONTPREP)
1864 break;
1865
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001866 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001867 /*
1868 * make sure space for the drain appears we
1869 * know we can do this because max_hw_segments
1870 * has been adjusted to be one fewer than the
1871 * device can handle
1872 */
1873 rq->nr_phys_segments++;
1874 }
1875
1876 if (!q->prep_rq_fn)
1877 break;
1878
1879 ret = q->prep_rq_fn(q, rq);
1880 if (ret == BLKPREP_OK) {
1881 break;
1882 } else if (ret == BLKPREP_DEFER) {
1883 /*
1884 * the request may have been (partially) prepped.
1885 * we need to keep this request in the front to
1886 * avoid resource deadlock. REQ_STARTED will
1887 * prevent other fs requests from passing this one.
1888 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001889 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001890 !(rq->cmd_flags & REQ_DONTPREP)) {
1891 /*
1892 * remove the space for the drain we added
1893 * so that we don't add it again
1894 */
1895 --rq->nr_phys_segments;
1896 }
1897
1898 rq = NULL;
1899 break;
1900 } else if (ret == BLKPREP_KILL) {
1901 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001902 /*
1903 * Mark this request as started so we don't trigger
1904 * any debug logic in the end I/O path.
1905 */
1906 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001907 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001908 } else {
1909 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1910 break;
1911 }
1912 }
1913
1914 return rq;
1915}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001916EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001917
Tejun Heo9934c8c2009-05-08 11:54:16 +09001918void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001919{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001920 struct request_queue *q = rq->q;
1921
Tejun Heo158dbda2009-04-23 11:05:18 +09001922 BUG_ON(list_empty(&rq->queuelist));
1923 BUG_ON(ELV_ON_HASH(rq));
1924
1925 list_del_init(&rq->queuelist);
1926
1927 /*
1928 * the time frame between a request being removed from the lists
1929 * and to it is freed is accounted as io that is in progress at
1930 * the driver side.
1931 */
Divyesh Shah91952912010-04-01 15:01:41 -07001932 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001933 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001934 set_io_start_time_ns(rq);
1935 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001936}
1937
Tejun Heo5efccd12009-04-23 11:05:18 +09001938/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001939 * blk_start_request - start request processing on the driver
1940 * @req: request to dequeue
1941 *
1942 * Description:
1943 * Dequeue @req and start timeout timer on it. This hands off the
1944 * request to the driver.
1945 *
1946 * Block internal functions which don't want to start timer should
1947 * call blk_dequeue_request().
1948 *
1949 * Context:
1950 * queue_lock must be held.
1951 */
1952void blk_start_request(struct request *req)
1953{
1954 blk_dequeue_request(req);
1955
1956 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001957 * We are now handing the request to the hardware, initialize
1958 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001959 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001960 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001961 if (unlikely(blk_bidi_rq(req)))
1962 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1963
Tejun Heo9934c8c2009-05-08 11:54:16 +09001964 blk_add_timer(req);
1965}
1966EXPORT_SYMBOL(blk_start_request);
1967
1968/**
1969 * blk_fetch_request - fetch a request from a request queue
1970 * @q: request queue to fetch a request from
1971 *
1972 * Description:
1973 * Return the request at the top of @q. The request is started on
1974 * return and LLD can start processing it immediately.
1975 *
1976 * Return:
1977 * Pointer to the request at the top of @q if available. Null
1978 * otherwise.
1979 *
1980 * Context:
1981 * queue_lock must be held.
1982 */
1983struct request *blk_fetch_request(struct request_queue *q)
1984{
1985 struct request *rq;
1986
1987 rq = blk_peek_request(q);
1988 if (rq)
1989 blk_start_request(rq);
1990 return rq;
1991}
1992EXPORT_SYMBOL(blk_fetch_request);
1993
1994/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001995 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001996 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001997 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001998 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001999 *
2000 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002001 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2002 * the request structure even if @req doesn't have leftover.
2003 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002004 *
2005 * This special helper function is only for request stacking drivers
2006 * (e.g. request-based dm) so that they can handle partial completion.
2007 * Actual device drivers should use blk_end_request instead.
2008 *
2009 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2010 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002011 *
2012 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002013 * %false - this request doesn't have any more data
2014 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002015 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002016bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002018 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 struct bio *bio;
2020
Tejun Heo2e60e022009-04-23 11:05:18 +09002021 if (!req->bio)
2022 return false;
2023
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002024 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002025
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002027 * For fs requests, rq is just carrier of independent bio's
2028 * and each partial completion should be handled separately.
2029 * Reset per-request error on each partial completion.
2030 *
2031 * TODO: tj: This is too subtle. It would be better to let
2032 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002034 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 req->errors = 0;
2036
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002037 if (error && req->cmd_type == REQ_TYPE_FS &&
2038 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002039 char *error_type;
2040
2041 switch (error) {
2042 case -ENOLINK:
2043 error_type = "recoverable transport";
2044 break;
2045 case -EREMOTEIO:
2046 error_type = "critical target";
2047 break;
2048 case -EBADE:
2049 error_type = "critical nexus";
2050 break;
2051 case -EIO:
2052 default:
2053 error_type = "I/O";
2054 break;
2055 }
2056 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2057 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2058 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 }
2060
Jens Axboebc58ba92009-01-23 10:54:44 +01002061 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 total_bytes = bio_nbytes = 0;
2064 while ((bio = req->bio) != NULL) {
2065 int nbytes;
2066
2067 if (nr_bytes >= bio->bi_size) {
2068 req->bio = bio->bi_next;
2069 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002070 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 next_idx = 0;
2072 bio_nbytes = 0;
2073 } else {
2074 int idx = bio->bi_idx + next_idx;
2075
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002076 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002078 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002079 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 break;
2081 }
2082
2083 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2084 BIO_BUG_ON(nbytes > bio->bi_size);
2085
2086 /*
2087 * not a complete bvec done
2088 */
2089 if (unlikely(nbytes > nr_bytes)) {
2090 bio_nbytes += nr_bytes;
2091 total_bytes += nr_bytes;
2092 break;
2093 }
2094
2095 /*
2096 * advance to the next vector
2097 */
2098 next_idx++;
2099 bio_nbytes += nbytes;
2100 }
2101
2102 total_bytes += nbytes;
2103 nr_bytes -= nbytes;
2104
Jens Axboe6728cb02008-01-31 13:03:55 +01002105 bio = req->bio;
2106 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 /*
2108 * end more in this run, or just return 'not-done'
2109 */
2110 if (unlikely(nr_bytes <= 0))
2111 break;
2112 }
2113 }
2114
2115 /*
2116 * completely done
2117 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002118 if (!req->bio) {
2119 /*
2120 * Reset counters so that the request stacking driver
2121 * can find how many bytes remain in the request
2122 * later.
2123 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002124 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002125 return false;
2126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
2128 /*
2129 * if the request wasn't completed, update state
2130 */
2131 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002132 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 bio->bi_idx += next_idx;
2134 bio_iovec(bio)->bv_offset += nr_bytes;
2135 bio_iovec(bio)->bv_len -= nr_bytes;
2136 }
2137
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002138 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002139 req->buffer = bio_data(req->bio);
2140
2141 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002142 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002143 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002144
Tejun Heo80a761f2009-07-03 17:48:17 +09002145 /* mixed attributes always follow the first bio */
2146 if (req->cmd_flags & REQ_MIXED_MERGE) {
2147 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2148 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2149 }
2150
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002151 /*
2152 * If total number of sectors is less than the first segment
2153 * size, something has gone terribly wrong.
2154 */
2155 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002156 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002157 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002158 }
2159
2160 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002162
Tejun Heo2e60e022009-04-23 11:05:18 +09002163 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
Tejun Heo2e60e022009-04-23 11:05:18 +09002165EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Tejun Heo2e60e022009-04-23 11:05:18 +09002167static bool blk_update_bidi_request(struct request *rq, int error,
2168 unsigned int nr_bytes,
2169 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002170{
Tejun Heo2e60e022009-04-23 11:05:18 +09002171 if (blk_update_request(rq, error, nr_bytes))
2172 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002173
Tejun Heo2e60e022009-04-23 11:05:18 +09002174 /* Bidi request must be completed as a whole */
2175 if (unlikely(blk_bidi_rq(rq)) &&
2176 blk_update_request(rq->next_rq, error, bidi_bytes))
2177 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002178
Jens Axboee2e1a142010-06-09 10:42:09 +02002179 if (blk_queue_add_random(rq->q))
2180 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002181
2182 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183}
2184
James Bottomley28018c22010-07-01 19:49:17 +09002185/**
2186 * blk_unprep_request - unprepare a request
2187 * @req: the request
2188 *
2189 * This function makes a request ready for complete resubmission (or
2190 * completion). It happens only after all error handling is complete,
2191 * so represents the appropriate moment to deallocate any resources
2192 * that were allocated to the request in the prep_rq_fn. The queue
2193 * lock is held when calling this.
2194 */
2195void blk_unprep_request(struct request *req)
2196{
2197 struct request_queue *q = req->q;
2198
2199 req->cmd_flags &= ~REQ_DONTPREP;
2200 if (q->unprep_rq_fn)
2201 q->unprep_rq_fn(q, req);
2202}
2203EXPORT_SYMBOL_GPL(blk_unprep_request);
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205/*
2206 * queue lock must be held
2207 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002208static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002210 if (blk_rq_tagged(req))
2211 blk_queue_end_tag(req->q, req);
2212
James Bottomleyba396a62009-05-27 14:17:08 +02002213 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002215 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002216 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Mike Andersone78042e2008-10-30 02:16:20 -07002218 blk_delete_timer(req);
2219
James Bottomley28018c22010-07-01 19:49:17 +09002220 if (req->cmd_flags & REQ_DONTPREP)
2221 blk_unprep_request(req);
2222
2223
Jens Axboebc58ba92009-01-23 10:54:44 +01002224 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002227 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002228 else {
2229 if (blk_bidi_rq(req))
2230 __blk_put_request(req->next_rq->q, req->next_rq);
2231
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 }
2234}
2235
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002236/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002237 * blk_end_bidi_request - Complete a bidi request
2238 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002239 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002240 * @nr_bytes: number of bytes to complete @rq
2241 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002242 *
2243 * Description:
2244 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002245 * Drivers that supports bidi can safely call this member for any
2246 * type of request, bidi or uni. In the later case @bidi_bytes is
2247 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002248 *
2249 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002250 * %false - we are done with this request
2251 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002252 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002253static bool blk_end_bidi_request(struct request *rq, int error,
2254 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002255{
2256 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002257 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002258
Tejun Heo2e60e022009-04-23 11:05:18 +09002259 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2260 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002261
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002262 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002263 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002264 spin_unlock_irqrestore(q->queue_lock, flags);
2265
Tejun Heo2e60e022009-04-23 11:05:18 +09002266 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002267}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002268
2269/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002270 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2271 * @rq: the request to complete
2272 * @error: %0 for success, < %0 for error
2273 * @nr_bytes: number of bytes to complete @rq
2274 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002275 *
2276 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002277 * Identical to blk_end_bidi_request() except that queue lock is
2278 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002279 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002280 * Return:
2281 * %false - we are done with this request
2282 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002283 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002284bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002285 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002286{
Tejun Heo2e60e022009-04-23 11:05:18 +09002287 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2288 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002289
Tejun Heo2e60e022009-04-23 11:05:18 +09002290 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002291
Tejun Heo2e60e022009-04-23 11:05:18 +09002292 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002293}
2294
2295/**
2296 * blk_end_request - Helper function for drivers to complete the request.
2297 * @rq: the request being processed
2298 * @error: %0 for success, < %0 for error
2299 * @nr_bytes: number of bytes to complete
2300 *
2301 * Description:
2302 * Ends I/O on a number of bytes attached to @rq.
2303 * If @rq has leftover, sets it up for the next range of segments.
2304 *
2305 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002306 * %false - we are done with this request
2307 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002308 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002309bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002310{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002311 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002312}
Jens Axboe56ad1742009-07-28 22:11:24 +02002313EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002314
2315/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002316 * blk_end_request_all - Helper function for drives to finish the request.
2317 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002318 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002319 *
2320 * Description:
2321 * Completely finish @rq.
2322 */
2323void blk_end_request_all(struct request *rq, int error)
2324{
2325 bool pending;
2326 unsigned int bidi_bytes = 0;
2327
2328 if (unlikely(blk_bidi_rq(rq)))
2329 bidi_bytes = blk_rq_bytes(rq->next_rq);
2330
2331 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2332 BUG_ON(pending);
2333}
Jens Axboe56ad1742009-07-28 22:11:24 +02002334EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002335
2336/**
2337 * blk_end_request_cur - Helper function to finish the current request chunk.
2338 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002339 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002340 *
2341 * Description:
2342 * Complete the current consecutively mapped chunk from @rq.
2343 *
2344 * Return:
2345 * %false - we are done with this request
2346 * %true - still buffers pending for this request
2347 */
2348bool blk_end_request_cur(struct request *rq, int error)
2349{
2350 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2351}
Jens Axboe56ad1742009-07-28 22:11:24 +02002352EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002353
2354/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002355 * blk_end_request_err - Finish a request till the next failure boundary.
2356 * @rq: the request to finish till the next failure boundary for
2357 * @error: must be negative errno
2358 *
2359 * Description:
2360 * Complete @rq till the next failure boundary.
2361 *
2362 * Return:
2363 * %false - we are done with this request
2364 * %true - still buffers pending for this request
2365 */
2366bool blk_end_request_err(struct request *rq, int error)
2367{
2368 WARN_ON(error >= 0);
2369 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2370}
2371EXPORT_SYMBOL_GPL(blk_end_request_err);
2372
2373/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002374 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002375 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002376 * @error: %0 for success, < %0 for error
2377 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002378 *
2379 * Description:
2380 * Must be called with queue lock held unlike blk_end_request().
2381 *
2382 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002383 * %false - we are done with this request
2384 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002385 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002386bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002387{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002388 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002389}
Jens Axboe56ad1742009-07-28 22:11:24 +02002390EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002391
2392/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002393 * __blk_end_request_all - Helper function for drives to finish the request.
2394 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002395 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002396 *
2397 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002399 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002400void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002401{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002402 bool pending;
2403 unsigned int bidi_bytes = 0;
2404
2405 if (unlikely(blk_bidi_rq(rq)))
2406 bidi_bytes = blk_rq_bytes(rq->next_rq);
2407
2408 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2409 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002410}
Jens Axboe56ad1742009-07-28 22:11:24 +02002411EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002412
2413/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002414 * __blk_end_request_cur - Helper function to finish the current request chunk.
2415 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002416 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002417 *
2418 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002419 * Complete the current consecutively mapped chunk from @rq. Must
2420 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002421 *
2422 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002423 * %false - we are done with this request
2424 * %true - still buffers pending for this request
2425 */
2426bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002427{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002428 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002429}
Jens Axboe56ad1742009-07-28 22:11:24 +02002430EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002431
Tejun Heo80a761f2009-07-03 17:48:17 +09002432/**
2433 * __blk_end_request_err - Finish a request till the next failure boundary.
2434 * @rq: the request to finish till the next failure boundary for
2435 * @error: must be negative errno
2436 *
2437 * Description:
2438 * Complete @rq till the next failure boundary. Must be called
2439 * with queue lock held.
2440 *
2441 * Return:
2442 * %false - we are done with this request
2443 * %true - still buffers pending for this request
2444 */
2445bool __blk_end_request_err(struct request *rq, int error)
2446{
2447 WARN_ON(error >= 0);
2448 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2449}
2450EXPORT_SYMBOL_GPL(__blk_end_request_err);
2451
Jens Axboe86db1e22008-01-29 14:53:40 +01002452void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2453 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002455 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002456 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
David Woodhousefb2dce82008-08-05 18:01:53 +01002458 if (bio_has_data(bio)) {
2459 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002460 rq->buffer = bio_data(bio);
2461 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002462 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
NeilBrown66846572007-08-16 13:31:28 +02002465 if (bio->bi_bdev)
2466 rq->rq_disk = bio->bi_bdev->bd_disk;
2467}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002469#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2470/**
2471 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2472 * @rq: the request to be flushed
2473 *
2474 * Description:
2475 * Flush all pages in @rq.
2476 */
2477void rq_flush_dcache_pages(struct request *rq)
2478{
2479 struct req_iterator iter;
2480 struct bio_vec *bvec;
2481
2482 rq_for_each_segment(bvec, rq, iter)
2483 flush_dcache_page(bvec->bv_page);
2484}
2485EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2486#endif
2487
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002488/**
2489 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2490 * @q : the queue of the device being checked
2491 *
2492 * Description:
2493 * Check if underlying low-level drivers of a device are busy.
2494 * If the drivers want to export their busy state, they must set own
2495 * exporting function using blk_queue_lld_busy() first.
2496 *
2497 * Basically, this function is used only by request stacking drivers
2498 * to stop dispatching requests to underlying devices when underlying
2499 * devices are busy. This behavior helps more I/O merging on the queue
2500 * of the request stacking driver and prevents I/O throughput regression
2501 * on burst I/O load.
2502 *
2503 * Return:
2504 * 0 - Not busy (The request stacking driver should dispatch request)
2505 * 1 - Busy (The request stacking driver should stop dispatching request)
2506 */
2507int blk_lld_busy(struct request_queue *q)
2508{
2509 if (q->lld_busy_fn)
2510 return q->lld_busy_fn(q);
2511
2512 return 0;
2513}
2514EXPORT_SYMBOL_GPL(blk_lld_busy);
2515
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002516/**
2517 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2518 * @rq: the clone request to be cleaned up
2519 *
2520 * Description:
2521 * Free all bios in @rq for a cloned request.
2522 */
2523void blk_rq_unprep_clone(struct request *rq)
2524{
2525 struct bio *bio;
2526
2527 while ((bio = rq->bio) != NULL) {
2528 rq->bio = bio->bi_next;
2529
2530 bio_put(bio);
2531 }
2532}
2533EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2534
2535/*
2536 * Copy attributes of the original request to the clone request.
2537 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2538 */
2539static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2540{
2541 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002542 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002543 dst->cmd_type = src->cmd_type;
2544 dst->__sector = blk_rq_pos(src);
2545 dst->__data_len = blk_rq_bytes(src);
2546 dst->nr_phys_segments = src->nr_phys_segments;
2547 dst->ioprio = src->ioprio;
2548 dst->extra_len = src->extra_len;
2549}
2550
2551/**
2552 * blk_rq_prep_clone - Helper function to setup clone request
2553 * @rq: the request to be setup
2554 * @rq_src: original request to be cloned
2555 * @bs: bio_set that bios for clone are allocated from
2556 * @gfp_mask: memory allocation mask for bio
2557 * @bio_ctr: setup function to be called for each clone bio.
2558 * Returns %0 for success, non %0 for failure.
2559 * @data: private data to be passed to @bio_ctr
2560 *
2561 * Description:
2562 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2563 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2564 * are not copied, and copying such parts is the caller's responsibility.
2565 * Also, pages which the original bios are pointing to are not copied
2566 * and the cloned bios just point same pages.
2567 * So cloned bios must be completed before original bios, which means
2568 * the caller must complete @rq before @rq_src.
2569 */
2570int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2571 struct bio_set *bs, gfp_t gfp_mask,
2572 int (*bio_ctr)(struct bio *, struct bio *, void *),
2573 void *data)
2574{
2575 struct bio *bio, *bio_src;
2576
2577 if (!bs)
2578 bs = fs_bio_set;
2579
2580 blk_rq_init(NULL, rq);
2581
2582 __rq_for_each_bio(bio_src, rq_src) {
2583 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2584 if (!bio)
2585 goto free_and_out;
2586
2587 __bio_clone(bio, bio_src);
2588
2589 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002590 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002591 goto free_and_out;
2592
2593 if (bio_ctr && bio_ctr(bio, bio_src, data))
2594 goto free_and_out;
2595
2596 if (rq->bio) {
2597 rq->biotail->bi_next = bio;
2598 rq->biotail = bio;
2599 } else
2600 rq->bio = rq->biotail = bio;
2601 }
2602
2603 __blk_rq_prep_clone(rq, rq_src);
2604
2605 return 0;
2606
2607free_and_out:
2608 if (bio)
2609 bio_free(bio, bs);
2610 blk_rq_unprep_clone(rq);
2611
2612 return -ENOMEM;
2613}
2614EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2615
Jens Axboe18887ad2008-07-28 13:08:45 +02002616int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617{
2618 return queue_work(kblockd_workqueue, work);
2619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620EXPORT_SYMBOL(kblockd_schedule_work);
2621
Vivek Goyale43473b2010-09-15 17:06:35 -04002622int kblockd_schedule_delayed_work(struct request_queue *q,
2623 struct delayed_work *dwork, unsigned long delay)
2624{
2625 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2626}
2627EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2628
Jens Axboe73c10102011-03-08 13:19:51 +01002629#define PLUG_MAGIC 0x91827364
2630
2631void blk_start_plug(struct blk_plug *plug)
2632{
2633 struct task_struct *tsk = current;
2634
2635 plug->magic = PLUG_MAGIC;
2636 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002637 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002638 plug->should_sort = 0;
2639
2640 /*
2641 * If this is a nested plug, don't actually assign it. It will be
2642 * flushed on its own.
2643 */
2644 if (!tsk->plug) {
2645 /*
2646 * Store ordering should not be needed here, since a potential
2647 * preempt will imply a full memory barrier
2648 */
2649 tsk->plug = plug;
2650 }
2651}
2652EXPORT_SYMBOL(blk_start_plug);
2653
2654static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2655{
2656 struct request *rqa = container_of(a, struct request, queuelist);
2657 struct request *rqb = container_of(b, struct request, queuelist);
2658
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002659 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002660}
2661
Jens Axboe49cac012011-04-16 13:51:05 +02002662/*
2663 * If 'from_schedule' is true, then postpone the dispatch of requests
2664 * until a safe kblockd context. We due this to avoid accidental big
2665 * additional stack usage in driver dispatch, in places where the originally
2666 * plugger did not intend it.
2667 */
Jens Axboef6603782011-04-15 15:49:07 +02002668static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002669 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002670 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002671{
Jens Axboe49cac012011-04-16 13:51:05 +02002672 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002673
2674 /*
2675 * If we are punting this to kblockd, then we can safely drop
2676 * the queue_lock before waking kblockd (which needs to take
2677 * this lock).
2678 */
2679 if (from_schedule) {
2680 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002681 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002682 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002683 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002684 spin_unlock(q->queue_lock);
2685 }
2686
Jens Axboe94b5eb22011-04-12 10:12:19 +02002687}
2688
NeilBrown048c9372011-04-18 09:52:22 +02002689static void flush_plug_callbacks(struct blk_plug *plug)
2690{
2691 LIST_HEAD(callbacks);
2692
2693 if (list_empty(&plug->cb_list))
2694 return;
2695
2696 list_splice_init(&plug->cb_list, &callbacks);
2697
2698 while (!list_empty(&callbacks)) {
2699 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2700 struct blk_plug_cb,
2701 list);
2702 list_del(&cb->list);
2703 cb->callback(cb);
2704 }
2705}
2706
Jens Axboe49cac012011-04-16 13:51:05 +02002707void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002708{
2709 struct request_queue *q;
2710 unsigned long flags;
2711 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002712 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002713 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002714
2715 BUG_ON(plug->magic != PLUG_MAGIC);
2716
NeilBrown048c9372011-04-18 09:52:22 +02002717 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002718 if (list_empty(&plug->list))
2719 return;
2720
NeilBrown109b8122011-04-11 14:13:10 +02002721 list_splice_init(&plug->list, &list);
2722
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}