blob: d2f8f4049abddbf76d3414cd85ef36d2d4027d07 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Li Zefan55782132009-06-09 13:43:05 +080031
32#define CREATE_TRACE_POINTS
33#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Jens Axboe8324aa92008-01-29 14:51:59 +010035#include "blk.h"
36
Mike Snitzerd07335e2010-11-16 12:52:38 +010037EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020038EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080039EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010040
Jens Axboe165125e2007-07-24 09:28:11 +020041static int __make_request(struct request_queue *q, struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For the allocated request tables
45 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010046static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
49 * For queue allocation
50 */
Jens Axboe6728cb02008-01-31 13:03:55 +010051struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * Controlling structure to kblockd
55 */
Jens Axboeff856ba2006-01-09 16:02:34 +010056static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboe26b82562008-01-29 13:54:41 +010058static void drive_stat_acct(struct request *rq, int new_io)
59{
Jens Axboe28f13702008-05-07 10:15:46 +020060 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010061 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090062 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010063
Jens Axboec2553b52009-04-24 08:10:11 +020064 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010065 return;
66
Tejun Heo074a7ac2008-08-25 19:56:14 +090067 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090068
Jerome Marchand09e099d2011-01-05 16:57:38 +010069 if (!new_io) {
70 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090071 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010072 } else {
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010074 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010075 /*
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
78 *
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
82 */
83 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010084 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010085 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090086 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020087 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010088 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010089 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020090
Tejun Heo074a7ac2008-08-25 19:56:14 +090091 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010092}
93
Jens Axboe8324aa92008-01-29 14:51:59 +010094void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095{
96 int nr;
97
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
107}
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109/**
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * @bdev: device
112 *
113 * Locates the passed device's request queue and returns the address of its
114 * backing_dev_info
115 *
116 * Will return NULL if the request queue cannot be located.
117 */
118struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119{
120 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200121 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123 if (q)
124 ret = &q->backing_dev_info;
125 return ret;
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127EXPORT_SYMBOL(blk_get_backing_dev_info);
128
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200129void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200131 memset(rq, 0, sizeof(*rq));
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700134 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200135 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100136 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900137 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200140 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800141 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100142 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100143 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900144 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700145 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100146 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200148EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
NeilBrown5bb23a62007-09-27 12:46:13 +0200150static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100152{
Tejun Heo143a87f2011-01-25 12:43:52 +0100153 if (error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100157
Tejun Heo143a87f2011-01-25 12:43:52 +0100158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200162 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100163
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
166
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
169
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
172
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100176}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
Jens Axboe6728cb02008-01-31 13:03:55 +0100182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Tejun Heo83096eb2009-05-07 22:24:39 +0900186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100193 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 printk("%02x ", rq->cmd[bit]);
196 printk("\n");
197 }
198}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199EXPORT_SYMBOL(blk_dump_rq_flags);
200
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500201static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200202{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500203 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200204
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200207 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500208 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 *
216 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
220 */
221void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Jens Axboe4521cc42011-04-18 11:36:39 +0200223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500226EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/**
229 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200230 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 *
232 * Description:
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200237void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200239 WARN_ON(!irqs_disabled());
240
Nick Piggin75ad23b2008-04-29 14:48:33 +0200241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200242 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244EXPORT_SYMBOL(blk_start_queue);
245
246/**
247 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200248 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 *
250 * Description:
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200260void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
Jens Axboead3d9d72011-03-25 16:58:59 +0100262 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265EXPORT_SYMBOL(blk_stop_queue);
266
267/**
268 * blk_sync_queue - cancel any pending callbacks on a queue
269 * @q: the queue
270 *
271 * Description:
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200276 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * that its ->make_request_fn will not re-add plugging prior to calling
278 * this function.
279 *
Vivek Goyalda527772011-03-02 19:05:33 -0500280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
283 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
285void blk_sync_queue(struct request_queue *q)
286{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100287 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500288 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290EXPORT_SYMBOL(blk_sync_queue);
291
292/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200293 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200295 *
296 * Description:
297 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200298 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200300void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Tejun Heoa538cd02009-04-23 11:05:17 +0900302 if (unlikely(blk_queue_stopped(q)))
303 return;
304
Jens Axboec21e6be2011-04-19 13:32:46 +0200305 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200306}
307EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200308
Nick Piggin75ad23b2008-04-29 14:48:33 +0200309/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
312 *
313 * Description:
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us.
316 */
317void blk_run_queue_async(struct request_queue *q)
318{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200322 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200323}
Jens Axboec21e6be2011-04-19 13:32:46 +0200324EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200325
326/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200329 *
330 * Description:
331 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900332 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200333 */
334void blk_run_queue(struct request_queue *q)
335{
336 unsigned long flags;
337
338 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200339 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 spin_unlock_irqrestore(q->queue_lock, flags);
341}
342EXPORT_SYMBOL(blk_run_queue);
343
Jens Axboe165125e2007-07-24 09:28:11 +0200344void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500345{
346 kobject_put(&q->kobj);
347}
Jens Axboed86e0e82011-05-27 07:44:43 +0200348EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500349
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500350/*
351 * Note: If a driver supplied the queue lock, it should not zap that lock
352 * unexpectedly as some queue cleanup components like elevator_exit() and
353 * blk_throtl_exit() need queue lock.
354 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100355void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500356{
Jens Axboee3335de2008-09-18 09:22:54 -0700357 /*
358 * We know we have process context here, so we can be a little
359 * cautious and ensure that pending block actions on this device
360 * are done before moving on. Going into this function, we should
361 * not have processes doing IO to this device.
362 */
363 blk_sync_queue(q);
364
Matthew Garrett31373d02010-04-06 14:25:14 +0200365 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500366 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200367 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500368 mutex_unlock(&q->sysfs_lock);
369
370 if (q->elevator)
371 elevator_exit(q->elevator);
372
Vivek Goyalda527772011-03-02 19:05:33 -0500373 blk_throtl_exit(q);
374
Al Viro483f4af2006-03-18 18:34:37 -0500375 blk_put_queue(q);
376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377EXPORT_SYMBOL(blk_cleanup_queue);
378
Jens Axboe165125e2007-07-24 09:28:11 +0200379static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
381 struct request_list *rl = &q->rq;
382
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400383 if (unlikely(rl->rq_pool))
384 return 0;
385
Jens Axboe1faa16d2009-04-06 14:48:01 +0200386 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
387 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200388 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200389 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
390 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Christoph Lameter19460892005-06-23 00:08:19 -0700392 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
393 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 if (!rl->rq_pool)
396 return -ENOMEM;
397
398 return 0;
399}
400
Jens Axboe165125e2007-07-24 09:28:11 +0200401struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
Christoph Lameter19460892005-06-23 00:08:19 -0700403 return blk_alloc_queue_node(gfp_mask, -1);
404}
405EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Jens Axboe165125e2007-07-24 09:28:11 +0200407struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700408{
Jens Axboe165125e2007-07-24 09:28:11 +0200409 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700410 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700411
Jens Axboe8324aa92008-01-29 14:51:59 +0100412 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700413 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (!q)
415 return NULL;
416
Jens Axboe0989a022009-06-12 14:42:56 +0200417 q->backing_dev_info.ra_pages =
418 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
419 q->backing_dev_info.state = 0;
420 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200421 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200422
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700423 err = bdi_init(&q->backing_dev_info);
424 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100425 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700426 return NULL;
427 }
428
Vivek Goyale43473b2010-09-15 17:06:35 -0400429 if (blk_throtl_init(q)) {
430 kmem_cache_free(blk_requestq_cachep, q);
431 return NULL;
432 }
433
Matthew Garrett31373d02010-04-06 14:25:14 +0200434 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
435 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700436 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
437 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100438 INIT_LIST_HEAD(&q->flush_queue[0]);
439 INIT_LIST_HEAD(&q->flush_queue[1]);
440 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500441 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500442
Jens Axboe8324aa92008-01-29 14:51:59 +0100443 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Al Viro483f4af2006-03-18 18:34:37 -0500445 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700446 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500447
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500448 /*
449 * By default initialize queue_lock to internal lock and driver can
450 * override it later if need be.
451 */
452 q->queue_lock = &q->__queue_lock;
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return q;
455}
Christoph Lameter19460892005-06-23 00:08:19 -0700456EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458/**
459 * blk_init_queue - prepare a request queue for use with a block device
460 * @rfn: The function to be called to process requests that have been
461 * placed on the queue.
462 * @lock: Request queue spin lock
463 *
464 * Description:
465 * If a block device wishes to use the standard request handling procedures,
466 * which sorts requests and coalesces adjacent requests, then it must
467 * call blk_init_queue(). The function @rfn will be called when there
468 * are requests on the queue that need to be processed. If the device
469 * supports plugging, then @rfn may not be called immediately when requests
470 * are available on the queue, but may be called at some time later instead.
471 * Plugged queues are generally unplugged when a buffer belonging to one
472 * of the requests on the queue is needed, or due to memory pressure.
473 *
474 * @rfn is not required, or even expected, to remove all requests off the
475 * queue, but only as many as it can handle at a time. If it does leave
476 * requests on the queue, it is responsible for arranging that the requests
477 * get dealt with eventually.
478 *
479 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200480 * request queue; this lock will be taken also from interrupt context, so irq
481 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200483 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 * it didn't succeed.
485 *
486 * Note:
487 * blk_init_queue() must be paired with a blk_cleanup_queue() call
488 * when the block device is deactivated (such as at module unload).
489 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700490
Jens Axboe165125e2007-07-24 09:28:11 +0200491struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Christoph Lameter19460892005-06-23 00:08:19 -0700493 return blk_init_queue_node(rfn, lock, -1);
494}
495EXPORT_SYMBOL(blk_init_queue);
496
Jens Axboe165125e2007-07-24 09:28:11 +0200497struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700498blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
499{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600500 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600502 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
503 if (!uninit_q)
504 return NULL;
505
506 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
507 if (!q)
508 blk_cleanup_queue(uninit_q);
509
510 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200511}
512EXPORT_SYMBOL(blk_init_queue_node);
513
514struct request_queue *
515blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
516 spinlock_t *lock)
517{
518 return blk_init_allocated_queue_node(q, rfn, lock, -1);
519}
520EXPORT_SYMBOL(blk_init_allocated_queue);
521
522struct request_queue *
523blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
524 spinlock_t *lock, int node_id)
525{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (!q)
527 return NULL;
528
Christoph Lameter19460892005-06-23 00:08:19 -0700529 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600530 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500531 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900535 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100536 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500537
538 /* Override internal queue lock with supplied lock pointer */
539 if (lock)
540 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Jens Axboef3b144a2009-03-06 08:48:33 +0100542 /*
543 * This also sets hw/phys segments, boundary and size
544 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Alan Stern44ec9542007-02-20 11:01:57 -0500547 q->sg_reserved_size = INT_MAX;
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 /*
550 * all done
551 */
552 if (!elevator_init(q, NULL)) {
553 blk_queue_congestion_threshold(q);
554 return q;
555 }
556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return NULL;
558}
Mike Snitzer01effb02010-05-11 08:57:42 +0200559EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Jens Axboe165125e2007-07-24 09:28:11 +0200561int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700563 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500564 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 return 0;
566 }
567
568 return 1;
569}
Jens Axboed86e0e82011-05-27 07:44:43 +0200570EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Jens Axboe165125e2007-07-24 09:28:11 +0200572static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
Jens Axboe4aff5e22006-08-10 08:44:47 +0200574 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200575 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 mempool_free(rq, q->rq.rq_pool);
577}
578
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200579static struct request *
Jerome Marchand42dad762009-04-22 14:01:49 +0200580blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
582 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
583
584 if (!rq)
585 return NULL;
586
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200587 blk_rq_init(q, rq);
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200588
Jerome Marchand42dad762009-04-22 14:01:49 +0200589 rq->cmd_flags = flags | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Tejun Heocb98fc82005-10-28 08:29:39 +0200591 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +0200592 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200593 mempool_free(rq, q->rq.rq_pool);
594 return NULL;
595 }
Jens Axboe4aff5e22006-08-10 08:44:47 +0200596 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +0200597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Tejun Heocb98fc82005-10-28 08:29:39 +0200599 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
602/*
603 * ioc_batching returns true if the ioc is a valid batching request and
604 * should be given priority access to a request.
605 */
Jens Axboe165125e2007-07-24 09:28:11 +0200606static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 if (!ioc)
609 return 0;
610
611 /*
612 * Make sure the process is able to allocate at least 1 request
613 * even if the batch times out, otherwise we could theoretically
614 * lose wakeups.
615 */
616 return ioc->nr_batch_requests == q->nr_batching ||
617 (ioc->nr_batch_requests > 0
618 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
619}
620
621/*
622 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
623 * will cause the process to be a "batcher" on all queues in the system. This
624 * is the behaviour we want though - once it gets a wakeup it should be given
625 * a nice run.
626 */
Jens Axboe165125e2007-07-24 09:28:11 +0200627static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 if (!ioc || ioc_batching(q, ioc))
630 return;
631
632 ioc->nr_batch_requests = q->nr_batching;
633 ioc->last_waited = jiffies;
634}
635
Jens Axboe1faa16d2009-04-06 14:48:01 +0200636static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 struct request_list *rl = &q->rq;
639
Jens Axboe1faa16d2009-04-06 14:48:01 +0200640 if (rl->count[sync] < queue_congestion_off_threshold(q))
641 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Jens Axboe1faa16d2009-04-06 14:48:01 +0200643 if (rl->count[sync] + 1 <= q->nr_requests) {
644 if (waitqueue_active(&rl->wait[sync]))
645 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Jens Axboe1faa16d2009-04-06 14:48:01 +0200647 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649}
650
651/*
652 * A request has just been released. Account for it, update the full and
653 * congestion status, wake up any waiters. Called under q->queue_lock.
654 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200655static void freed_request(struct request_queue *q, int sync, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 struct request_list *rl = &q->rq;
658
Jens Axboe1faa16d2009-04-06 14:48:01 +0200659 rl->count[sync]--;
Tejun Heocb98fc82005-10-28 08:29:39 +0200660 if (priv)
661 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Jens Axboe1faa16d2009-04-06 14:48:01 +0200663 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Jens Axboe1faa16d2009-04-06 14:48:01 +0200665 if (unlikely(rl->starved[sync ^ 1]))
666 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100670 * Determine if elevator data should be initialized when allocating the
671 * request associated with @bio.
672 */
673static bool blk_rq_should_init_elevator(struct bio *bio)
674{
675 if (!bio)
676 return true;
677
678 /*
679 * Flush requests do not use the elevator so skip initialization.
680 * This allows a request to share the flush and elevator data.
681 */
682 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
683 return false;
684
685 return true;
686}
687
688/*
Nick Piggind6344532005-06-28 20:45:14 -0700689 * Get a free request, queue_lock must be held.
690 * Returns NULL on failure, with queue_lock held.
691 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 */
Jens Axboe165125e2007-07-24 09:28:11 +0200693static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100694 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct request *rq = NULL;
697 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100698 struct io_context *ioc = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200699 const bool is_sync = rw_is_sync(rw_flags) != 0;
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100700 int may_queue, priv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Jens Axboe7749a8d2006-12-13 13:02:26 +0100702 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100703 if (may_queue == ELV_MQUEUE_NO)
704 goto rq_starved;
705
Jens Axboe1faa16d2009-04-06 14:48:01 +0200706 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
707 if (rl->count[is_sync]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +0200708 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100709 /*
710 * The queue will fill after this allocation, so set
711 * it as full, and mark this process as "batching".
712 * This process will be allowed to complete a batch of
713 * requests, others will be blocked.
714 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200715 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100716 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200717 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100718 } else {
719 if (may_queue != ELV_MQUEUE_MUST
720 && !ioc_batching(q, ioc)) {
721 /*
722 * The queue is full and the allocating
723 * process is not a "batcher", and not
724 * exempted by the IO scheduler
725 */
726 goto out;
727 }
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200730 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732
Jens Axboe082cf692005-06-28 16:35:11 +0200733 /*
734 * Only allow batching queuers to allocate up to 50% over the defined
735 * limit of requests, otherwise we could have thousands of requests
736 * allocated with any setting of ->nr_requests
737 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200738 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +0200739 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100740
Jens Axboe1faa16d2009-04-06 14:48:01 +0200741 rl->count[is_sync]++;
742 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200743
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100744 if (blk_rq_should_init_elevator(bio)) {
745 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
746 if (priv)
747 rl->elvpriv++;
748 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200749
Jens Axboef253b862010-10-24 22:06:02 +0200750 if (blk_queue_io_stat(q))
751 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 spin_unlock_irq(q->queue_lock);
753
Jens Axboe7749a8d2006-12-13 13:02:26 +0100754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100755 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /*
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
759 *
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
762 */
763 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200764 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /*
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
772 */
773rq_starved:
Jens Axboe1faa16d2009-04-06 14:48:01 +0200774 if (unlikely(rl->count[is_sync] == 0))
775 rl->starved[is_sync] = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto out;
778 }
779
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100780 /*
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
785 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100788
Jens Axboe1faa16d2009-04-06 14:48:01 +0200789 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return rq;
792}
793
794/*
Jens Axboe7eaceac2011-03-10 08:52:07 +0100795 * No available requests for this queue, wait for some requests to become
796 * available.
Nick Piggind6344532005-06-28 20:45:14 -0700797 *
798 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Jens Axboe165125e2007-07-24 09:28:11 +0200800static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +0200801 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200803 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 struct request *rq;
805
Jens Axboe7749a8d2006-12-13 13:02:26 +0100806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -0700807 while (!rq) {
808 DEFINE_WAIT(wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200809 struct io_context *ioc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct request_list *rl = &q->rq;
811
Jens Axboe1faa16d2009-04-06 14:48:01 +0200812 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 TASK_UNINTERRUPTIBLE);
814
Jens Axboe1faa16d2009-04-06 14:48:01 +0200815 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200817 spin_unlock_irq(q->queue_lock);
818 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200820 /*
821 * After sleeping, we become a "batching" process and
822 * will be able to allocate at least one request, and
823 * up to a big batch of them for a small period time.
824 * See ioc_batching, ioc_set_batching
825 */
826 ioc = current_io_context(GFP_NOIO, q->node);
827 ioc_set_batching(q, ioc);
Jens Axboe2056a782006-03-23 20:00:26 +0100828
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200829 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200830 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +0200831
832 rq = get_request(q, rw_flags, bio, GFP_NOIO);
833 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 return rq;
836}
837
Jens Axboe165125e2007-07-24 09:28:11 +0200838struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
840 struct request *rq;
841
842 BUG_ON(rw != READ && rw != WRITE);
843
Nick Piggind6344532005-06-28 20:45:14 -0700844 spin_lock_irq(q->queue_lock);
845 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200846 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -0700847 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +0200848 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -0700849 if (!rq)
850 spin_unlock_irq(q->queue_lock);
851 }
852 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 return rq;
855}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856EXPORT_SYMBOL(blk_get_request);
857
858/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300859 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700860 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300861 * @bio: The bio describing the memory mappings that will be submitted for IO.
862 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700863 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +0200864 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300865 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
866 * type commands. Where the struct request needs to be farther initialized by
867 * the caller. It is passed a &struct bio, which describes the memory info of
868 * the I/O transfer.
869 *
870 * The caller of blk_make_request must make sure that bi_io_vec
871 * are set to describe the memory buffers. That bio_data_dir() will return
872 * the needed direction of the request. (And all bio's in the passed bio-chain
873 * are properly set accordingly)
874 *
875 * If called under none-sleepable conditions, mapped bio buffers must not
876 * need bouncing, by calling the appropriate masked or flagged allocator,
877 * suitable for the target device. Otherwise the call to blk_queue_bounce will
878 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +0200879 *
880 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
881 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
882 * anything but the first bio in the chain. Otherwise you risk waiting for IO
883 * completion of a bio that hasn't been submitted yet, thus resulting in a
884 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
885 * of bio_alloc(), as that avoids the mempool deadlock.
886 * If possible a big IO should be split into smaller parts when allocation
887 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +0200888 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300889struct request *blk_make_request(struct request_queue *q, struct bio *bio,
890 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +0200891{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300892 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
893
894 if (unlikely(!rq))
895 return ERR_PTR(-ENOMEM);
896
897 for_each_bio(bio) {
898 struct bio *bounce_bio = bio;
899 int ret;
900
901 blk_queue_bounce(q, &bounce_bio);
902 ret = blk_rq_append_bio(q, rq, bounce_bio);
903 if (unlikely(ret)) {
904 blk_put_request(rq);
905 return ERR_PTR(ret);
906 }
907 }
908
909 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +0200910}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +0300911EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +0200912
913/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 * blk_requeue_request - put a request back on queue
915 * @q: request queue where request should be inserted
916 * @rq: request to be inserted
917 *
918 * Description:
919 * Drivers often keep queueing requests until the hardware cannot accept
920 * more, when that condition happens we need to put the request back
921 * on the queue. Must be called with queue lock held.
922 */
Jens Axboe165125e2007-07-24 09:28:11 +0200923void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700925 blk_delete_timer(rq);
926 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100927 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +0100928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 if (blk_rq_tagged(rq))
930 blk_queue_end_tag(q, rq);
931
James Bottomleyba396a62009-05-27 14:17:08 +0200932 BUG_ON(blk_queued_rq(rq));
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 elv_requeue_request(q, rq);
935}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936EXPORT_SYMBOL(blk_requeue_request);
937
Jens Axboe73c10102011-03-08 13:19:51 +0100938static void add_acct_request(struct request_queue *q, struct request *rq,
939 int where)
940{
941 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100942 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +0100943}
944
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200946 * blk_insert_request - insert a special request into a request queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 * @q: request queue where request should be inserted
948 * @rq: request to be inserted
949 * @at_head: insert request at head or tail of queue
950 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 *
952 * Description:
953 * Many block devices need to execute commands asynchronously, so they don't
954 * block the whole kernel from preemption during request execution. This is
955 * accomplished normally by inserting aritficial requests tagged as
Randy Dunlap710027a2008-08-19 20:13:11 +0200956 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
957 * be scheduled for actual execution by the request queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 *
959 * We have the option of inserting the head or the tail of the queue.
960 * Typically we use the tail for new ioctls and so forth. We use the head
961 * of the queue for things like a QUEUE_FULL message from a device, or a
962 * host that is unable to accept a particular command.
963 */
Jens Axboe165125e2007-07-24 09:28:11 +0200964void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -0500965 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Tejun Heo 867d1192005-04-24 02:06:05 -0500967 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 unsigned long flags;
969
970 /*
971 * tell I/O scheduler that this isn't a regular read/write (ie it
972 * must not attempt merges on this) and that it acts as a soft
973 * barrier
974 */
Jens Axboe4aff5e22006-08-10 08:44:47 +0200975 rq->cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 rq->special = data;
978
979 spin_lock_irqsave(q->queue_lock, flags);
980
981 /*
982 * If command is tagged, release the tag
983 */
Tejun Heo 867d1192005-04-24 02:06:05 -0500984 if (blk_rq_tagged(rq))
985 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Jens Axboe73c10102011-03-08 13:19:51 +0100987 add_acct_request(q, rq, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200988 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 spin_unlock_irqrestore(q->queue_lock, flags);
990}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991EXPORT_SYMBOL(blk_insert_request);
992
Tejun Heo074a7ac2008-08-25 19:56:14 +0900993static void part_round_stats_single(int cpu, struct hd_struct *part,
994 unsigned long now)
995{
996 if (now == part->stamp)
997 return;
998
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200999 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001000 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001001 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001002 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1003 }
1004 part->stamp = now;
1005}
1006
1007/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001008 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1009 * @cpu: cpu number for stats access
1010 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 *
1012 * The average IO queue length and utilisation statistics are maintained
1013 * by observing the current state of the queue length and the amount of
1014 * time it has been in this state for.
1015 *
1016 * Normally, that accounting is done on IO completion, but that can result
1017 * in more than a second's worth of IO being accounted for within any one
1018 * second, leading to >100% utilisation. To deal with that, we call this
1019 * function to do a round-off before returning the results when reading
1020 * /proc/diskstats. This accounts immediately for all queue usage up to
1021 * the current jiffies and restarts the counters again.
1022 */
Tejun Heoc9959052008-08-25 19:47:21 +09001023void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001024{
1025 unsigned long now = jiffies;
1026
Tejun Heo074a7ac2008-08-25 19:56:14 +09001027 if (part->partno)
1028 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1029 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001030}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001031EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033/*
1034 * queue lock must be held
1035 */
Jens Axboe165125e2007-07-24 09:28:11 +02001036void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (unlikely(!q))
1039 return;
1040 if (unlikely(--req->ref_count))
1041 return;
1042
Tejun Heo8922e162005-10-20 16:23:44 +02001043 elv_completed_request(q, req);
1044
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001045 /* this is a bio leak */
1046 WARN_ON(req->bio != NULL);
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /*
1049 * Request may not have originated from ll_rw_blk. if not,
1050 * it didn't come out of our reserved rq pools
1051 */
Jens Axboe49171e52006-08-10 08:59:11 +02001052 if (req->cmd_flags & REQ_ALLOCED) {
Jens Axboe1faa16d2009-04-06 14:48:01 +02001053 int is_sync = rq_is_sync(req) != 0;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001054 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001057 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 blk_free_request(q, req);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001060 freed_request(q, is_sync, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062}
Mike Christie6e39b692005-11-11 05:30:24 -06001063EXPORT_SYMBOL_GPL(__blk_put_request);
1064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065void blk_put_request(struct request *req)
1066{
Tejun Heo8922e162005-10-20 16:23:44 +02001067 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001068 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001070 spin_lock_irqsave(q->queue_lock, flags);
1071 __blk_put_request(q, req);
1072 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074EXPORT_SYMBOL(blk_put_request);
1075
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001076/**
1077 * blk_add_request_payload - add a payload to a request
1078 * @rq: request to update
1079 * @page: page backing the payload
1080 * @len: length of the payload.
1081 *
1082 * This allows to later add a payload to an already submitted request by
1083 * a block driver. The driver needs to take care of freeing the payload
1084 * itself.
1085 *
1086 * Note that this is a quite horrible hack and nothing but handling of
1087 * discard requests should ever use it.
1088 */
1089void blk_add_request_payload(struct request *rq, struct page *page,
1090 unsigned int len)
1091{
1092 struct bio *bio = rq->bio;
1093
1094 bio->bi_io_vec->bv_page = page;
1095 bio->bi_io_vec->bv_offset = 0;
1096 bio->bi_io_vec->bv_len = len;
1097
1098 bio->bi_size = len;
1099 bio->bi_vcnt = 1;
1100 bio->bi_phys_segments = 1;
1101
1102 rq->__data_len = rq->resid_len = len;
1103 rq->nr_phys_segments = 1;
1104 rq->buffer = bio_data(bio);
1105}
1106EXPORT_SYMBOL_GPL(blk_add_request_payload);
1107
Jens Axboe73c10102011-03-08 13:19:51 +01001108static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1109 struct bio *bio)
1110{
1111 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1112
Jens Axboe73c10102011-03-08 13:19:51 +01001113 if (!ll_back_merge_fn(q, req, bio))
1114 return false;
1115
1116 trace_block_bio_backmerge(q, bio);
1117
1118 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1119 blk_rq_set_mixed_merge(req);
1120
1121 req->biotail->bi_next = bio;
1122 req->biotail = bio;
1123 req->__data_len += bio->bi_size;
1124 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1125
1126 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001127 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001128 return true;
1129}
1130
1131static bool bio_attempt_front_merge(struct request_queue *q,
1132 struct request *req, struct bio *bio)
1133{
1134 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001135
Jens Axboe73c10102011-03-08 13:19:51 +01001136 if (!ll_front_merge_fn(q, req, bio))
1137 return false;
1138
1139 trace_block_bio_frontmerge(q, bio);
1140
1141 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1142 blk_rq_set_mixed_merge(req);
1143
Jens Axboe73c10102011-03-08 13:19:51 +01001144 bio->bi_next = req->bio;
1145 req->bio = bio;
1146
1147 /*
1148 * may not be valid. if the low level driver said
1149 * it didn't need a bounce buffer then it better
1150 * not touch req->buffer either...
1151 */
1152 req->buffer = bio_data(bio);
1153 req->__sector = bio->bi_sector;
1154 req->__data_len += bio->bi_size;
1155 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1156
1157 drive_stat_acct(req, 0);
Vivek Goyal95cf3dd2011-05-23 10:02:19 +02001158 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001159 return true;
1160}
1161
1162/*
1163 * Attempts to merge with the plugged list in the current process. Returns
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001164 * true if merge was successful, otherwise false.
Jens Axboe73c10102011-03-08 13:19:51 +01001165 */
1166static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1167 struct bio *bio)
1168{
1169 struct blk_plug *plug;
1170 struct request *rq;
1171 bool ret = false;
1172
1173 plug = tsk->plug;
1174 if (!plug)
1175 goto out;
1176
1177 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1178 int el_ret;
1179
1180 if (rq->q != q)
1181 continue;
1182
1183 el_ret = elv_try_merge(rq, bio);
1184 if (el_ret == ELEVATOR_BACK_MERGE) {
1185 ret = bio_attempt_back_merge(q, rq, bio);
1186 if (ret)
1187 break;
1188 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1189 ret = bio_attempt_front_merge(q, rq, bio);
1190 if (ret)
1191 break;
1192 }
1193 }
1194out:
1195 return ret;
1196}
1197
Jens Axboe86db1e22008-01-29 14:53:40 +01001198void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001199{
Jens Axboec7c22e42008-09-13 20:26:01 +02001200 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001201 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001202
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001203 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1204 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001205 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001206
Tejun Heo52d9e672006-01-06 09:49:58 +01001207 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001208 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001209 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001210 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001211}
1212
Jens Axboe165125e2007-07-24 09:28:11 +02001213static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001215 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001216 struct blk_plug *plug;
1217 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1218 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 /*
1221 * low level driver can indicate that it wants pages above a
1222 * certain limit bounced to low memory (ie for highmem, or even
1223 * ISA dma in theory)
1224 */
1225 blk_queue_bounce(q, &bio);
1226
Tejun Heo4fed9472010-09-03 11:56:17 +02001227 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001228 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001229 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001230 goto get_rq;
1231 }
1232
Jens Axboe73c10102011-03-08 13:19:51 +01001233 /*
1234 * Check if we can merge with the plugged list before grabbing
1235 * any locks.
1236 */
1237 if (attempt_plug_merge(current, q, bio))
1238 goto out;
1239
1240 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001243 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001244 if (bio_attempt_back_merge(q, req, bio)) {
1245 if (!attempt_back_merge(q, req))
1246 elv_merged_request(q, req, el_ret);
1247 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001248 }
Jens Axboe73c10102011-03-08 13:19:51 +01001249 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001250 if (bio_attempt_front_merge(q, req, bio)) {
1251 if (!attempt_front_merge(q, req))
1252 elv_merged_request(q, req, el_ret);
1253 goto out_unlock;
1254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001258 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001259 * This sync check and mask will be re-done in init_request_from_bio(),
1260 * but we need to set it earlier to expose the sync flag to the
1261 * rq allocator and io schedulers.
1262 */
1263 rw_flags = bio_data_dir(bio);
1264 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001265 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001266
1267 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001268 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001269 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001270 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001271 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001272
Nick Piggin450991b2005-06-28 20:45:13 -07001273 /*
1274 * After dropping the lock and possibly sleeping here, our request
1275 * may now be mergeable after it had proven unmergeable (above).
1276 * We don't worry about that case for efficiency. It won't happen
1277 * often, and the elevators are able to handle it.
1278 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001279 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Jens Axboec7c22e42008-09-13 20:26:01 +02001281 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Jens Axboe73c10102011-03-08 13:19:51 +01001282 bio_flagged(bio, BIO_CPU_AFFINE)) {
1283 req->cpu = blk_cpu_to_group(get_cpu());
1284 put_cpu();
1285 }
Tejun Heodd831002010-09-03 11:56:16 +02001286
Jens Axboe73c10102011-03-08 13:19:51 +01001287 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001288 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001289 /*
1290 * If this is the first request added after a plug, fire
1291 * of a plug trace. If others have been added before, check
1292 * if we have multiple devices in this plug. If so, make a
1293 * note to sort the list before dispatch.
1294 */
1295 if (list_empty(&plug->list))
1296 trace_block_plug(q);
1297 else if (!plug->should_sort) {
Jens Axboe73c10102011-03-08 13:19:51 +01001298 struct request *__rq;
1299
1300 __rq = list_entry_rq(plug->list.prev);
1301 if (__rq->q != q)
1302 plug->should_sort = 1;
1303 }
Jens Axboe73c10102011-03-08 13:19:51 +01001304 list_add_tail(&req->queuelist, &plug->list);
1305 drive_stat_acct(req, 1);
1306 } else {
1307 spin_lock_irq(q->queue_lock);
1308 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001309 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001310out_unlock:
1311 spin_unlock_irq(q->queue_lock);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315}
1316
1317/*
1318 * If bio->bi_dev is a partition, remap the location
1319 */
1320static inline void blk_partition_remap(struct bio *bio)
1321{
1322 struct block_device *bdev = bio->bi_bdev;
1323
Jens Axboebf2de6f2007-09-27 13:01:25 +02001324 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 bio->bi_sector += p->start_sect;
1328 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001329
Mike Snitzerd07335e2010-11-16 12:52:38 +01001330 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1331 bdev->bd_dev,
1332 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 }
1334}
1335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336static void handle_bad_sector(struct bio *bio)
1337{
1338 char b[BDEVNAME_SIZE];
1339
1340 printk(KERN_INFO "attempt to access beyond end of device\n");
1341 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1342 bdevname(bio->bi_bdev, b),
1343 bio->bi_rw,
1344 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001345 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 set_bit(BIO_EOF, &bio->bi_flags);
1348}
1349
Akinobu Mitac17bb492006-12-08 02:39:46 -08001350#ifdef CONFIG_FAIL_MAKE_REQUEST
1351
1352static DECLARE_FAULT_ATTR(fail_make_request);
1353
1354static int __init setup_fail_make_request(char *str)
1355{
1356 return setup_fault_attr(&fail_make_request, str);
1357}
1358__setup("fail_make_request=", setup_fail_make_request);
1359
1360static int should_fail_request(struct bio *bio)
1361{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001362 struct hd_struct *part = bio->bi_bdev->bd_part;
1363
1364 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001365 return should_fail(&fail_make_request, bio->bi_size);
1366
1367 return 0;
1368}
1369
1370static int __init fail_make_request_debugfs(void)
1371{
1372 return init_fault_attr_dentries(&fail_make_request,
1373 "fail_make_request");
1374}
1375
1376late_initcall(fail_make_request_debugfs);
1377
1378#else /* CONFIG_FAIL_MAKE_REQUEST */
1379
1380static inline int should_fail_request(struct bio *bio)
1381{
1382 return 0;
1383}
1384
1385#endif /* CONFIG_FAIL_MAKE_REQUEST */
1386
Jens Axboec07e2b42007-07-18 13:27:58 +02001387/*
1388 * Check whether this bio extends beyond the end of the device.
1389 */
1390static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1391{
1392 sector_t maxsector;
1393
1394 if (!nr_sectors)
1395 return 0;
1396
1397 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001398 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001399 if (maxsector) {
1400 sector_t sector = bio->bi_sector;
1401
1402 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1403 /*
1404 * This may well happen - the kernel calls bread()
1405 * without checking the size of the device, e.g., when
1406 * mounting a device.
1407 */
1408 handle_bad_sector(bio);
1409 return 1;
1410 }
1411 }
1412
1413 return 0;
1414}
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001417 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 * @bio: The bio describing the location in memory and on the device.
1419 *
1420 * generic_make_request() is used to make I/O requests of block
1421 * devices. It is passed a &struct bio, which describes the I/O that needs
1422 * to be done.
1423 *
1424 * generic_make_request() does not return any status. The
1425 * success/failure status of the request, along with notification of
1426 * completion, is delivered asynchronously through the bio->bi_end_io
1427 * function described (one day) else where.
1428 *
1429 * The caller of generic_make_request must make sure that bi_io_vec
1430 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1431 * set to describe the device address, and the
1432 * bi_end_io and optionally bi_private are set to describe how
1433 * completion notification should be signaled.
1434 *
1435 * generic_make_request and the drivers it calls may use bi_next if this
1436 * bio happens to be merged with someone else, and may change bi_dev and
1437 * bi_sector for remaps as it sees fit. So the values of these fields
1438 * should NOT be depended on after the call to generic_make_request.
1439 */
Neil Brownd89d8792007-05-01 09:53:42 +02001440static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441{
Jens Axboe165125e2007-07-24 09:28:11 +02001442 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001443 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001445 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001446 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
1448 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Jens Axboec07e2b42007-07-18 13:27:58 +02001450 if (bio_check_eod(bio, nr_sectors))
1451 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 /*
1454 * Resolve the mapping until finished. (drivers are
1455 * still free to implement/resolve their own stacking
1456 * by explicitly returning 0)
1457 *
1458 * NOTE: we don't repeat the blk_size check for each new device.
1459 * Stacking drivers are expected to know what they are doing.
1460 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001461 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001462 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 do {
1464 char b[BDEVNAME_SIZE];
1465
1466 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001467 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 printk(KERN_ERR
1469 "generic_make_request: Trying to access "
1470 "nonexistent block-device %s (%Lu)\n",
1471 bdevname(bio->bi_bdev, b),
1472 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001473 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
1475
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001476 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001477 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001478 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001479 bdevname(bio->bi_bdev, b),
1480 bio_sectors(bio),
1481 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 goto end_io;
1483 }
1484
Nick Pigginfde6ad22005-06-23 00:08:53 -07001485 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 goto end_io;
1487
Akinobu Mitac17bb492006-12-08 02:39:46 -08001488 if (should_fail_request(bio))
1489 goto end_io;
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 /*
1492 * If this device has partitions, remap block n
1493 * of partition p to block n+start(p) of the disk.
1494 */
1495 blk_partition_remap(bio);
1496
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001497 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1498 goto end_io;
1499
NeilBrown5ddfe962006-10-30 22:07:21 -08001500 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001501 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001502
NeilBrown5ddfe962006-10-30 22:07:21 -08001503 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001504 old_dev = bio->bi_bdev->bd_dev;
1505
Jens Axboec07e2b42007-07-18 13:27:58 +02001506 if (bio_check_eod(bio, nr_sectors))
1507 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001508
Tejun Heo1e879012010-09-03 11:56:17 +02001509 /*
1510 * Filter flush bio's early so that make_request based
1511 * drivers without flush support don't have to worry
1512 * about them.
1513 */
1514 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1515 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1516 if (!nr_sectors) {
1517 err = 0;
1518 goto end_io;
1519 }
1520 }
1521
Adrian Hunter8d57a982010-08-11 14:17:49 -07001522 if ((bio->bi_rw & REQ_DISCARD) &&
1523 (!blk_queue_discard(q) ||
1524 ((bio->bi_rw & REQ_SECURE) &&
1525 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001526 err = -EOPNOTSUPP;
1527 goto end_io;
1528 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001529
Vivek Goyalf469a7b2011-05-19 15:38:23 -04001530 if (blk_throtl_bio(q, &bio))
1531 goto end_io;
Vivek Goyale43473b2010-09-15 17:06:35 -04001532
1533 /*
1534 * If bio = NULL, bio has been throttled and will be submitted
1535 * later.
1536 */
1537 if (!bio)
1538 break;
1539
Minchan Kim01edede2009-09-08 21:56:38 +02001540 trace_block_bio_queue(q, bio);
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 ret = q->make_request_fn(q, bio);
1543 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001544
1545 return;
1546
1547end_io:
1548 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
Neil Brownd89d8792007-05-01 09:53:42 +02001551/*
1552 * We only want one ->make_request_fn to be active at a time,
1553 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001554 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001555 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001556 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001557 * generic_make_request is currently active in this task or not.
1558 * If it is NULL, then no make_request is active. If it is non-NULL,
1559 * then a make_request is active, and new requests should be added
1560 * at the tail
1561 */
1562void generic_make_request(struct bio *bio)
1563{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001564 struct bio_list bio_list_on_stack;
1565
1566 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001567 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001568 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001569 return;
1570 }
1571 /* following loop may be a bit non-obvious, and so deserves some
1572 * explanation.
1573 * Before entering the loop, bio->bi_next is NULL (as all callers
1574 * ensure that) so we have a list with a single bio.
1575 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 * we assign bio_list to a pointer to the bio_list_on_stack,
1577 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001578 * added. __generic_make_request may indeed add some more bios
1579 * through a recursive call to generic_make_request. If it
1580 * did, we find a non-NULL value in bio_list and re-enter the loop
1581 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001582 * of the top of the list (no pretending) and so remove it from
1583 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001584 *
1585 * The loop was structured like this to make only one call to
1586 * __generic_make_request (which is important as it is large and
1587 * inlined) and to keep the structure simple.
1588 */
1589 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001590 bio_list_init(&bio_list_on_stack);
1591 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001592 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001593 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001594 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001595 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001596 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001597}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598EXPORT_SYMBOL(generic_make_request);
1599
1600/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001601 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1603 * @bio: The &struct bio which describes the I/O
1604 *
1605 * submit_bio() is very similar in purpose to generic_make_request(), and
1606 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001607 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 *
1609 */
1610void submit_bio(int rw, struct bio *bio)
1611{
1612 int count = bio_sectors(bio);
1613
Jens Axboe22e2c502005-06-27 10:55:12 +02001614 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
Jens Axboebf2de6f2007-09-27 13:01:25 +02001616 /*
1617 * If it's a regular read/write or a barrier with data attached,
1618 * go through the normal accounting stuff before submission.
1619 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001620 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001621 if (rw & WRITE) {
1622 count_vm_events(PGPGOUT, count);
1623 } else {
1624 task_io_account_read(bio->bi_size);
1625 count_vm_events(PGPGIN, count);
1626 }
1627
1628 if (unlikely(block_dump)) {
1629 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001630 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001631 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001632 (rw & WRITE) ? "WRITE" : "READ",
1633 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001634 bdevname(bio->bi_bdev, b),
1635 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 }
1638
1639 generic_make_request(bio);
1640}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641EXPORT_SYMBOL(submit_bio);
1642
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001643/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001644 * blk_rq_check_limits - Helper function to check a request for the queue limit
1645 * @q: the queue
1646 * @rq: the request being checked
1647 *
1648 * Description:
1649 * @rq may have been made based on weaker limitations of upper-level queues
1650 * in request stacking drivers, and it may violate the limitation of @q.
1651 * Since the block layer and the underlying device driver trust @rq
1652 * after it is inserted to @q, it should be checked against @q before
1653 * the insertion using this generic function.
1654 *
1655 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001656 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001657 * Request stacking drivers like request-based dm may change the queue
1658 * limits while requests are in the queue (e.g. dm's table swapping).
1659 * Such request stacking drivers should check those requests agaist
1660 * the new queue limits again when they dispatch those requests,
1661 * although such checkings are also done against the old queue limits
1662 * when submitting requests.
1663 */
1664int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1665{
ike Snitzer33839772010-08-08 12:11:33 -04001666 if (rq->cmd_flags & REQ_DISCARD)
1667 return 0;
1668
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001669 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1670 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001671 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1672 return -EIO;
1673 }
1674
1675 /*
1676 * queue's settings related to segment counting like q->bounce_pfn
1677 * may differ from that of other stacking queues.
1678 * Recalculate it to check the request correctly on this queue's
1679 * limitation.
1680 */
1681 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001682 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001683 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1684 return -EIO;
1685 }
1686
1687 return 0;
1688}
1689EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1690
1691/**
1692 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1693 * @q: the queue to submit the request
1694 * @rq: the request being queued
1695 */
1696int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1697{
1698 unsigned long flags;
1699
1700 if (blk_rq_check_limits(q, rq))
1701 return -EIO;
1702
1703#ifdef CONFIG_FAIL_MAKE_REQUEST
1704 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1705 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1706 return -EIO;
1707#endif
1708
1709 spin_lock_irqsave(q->queue_lock, flags);
1710
1711 /*
1712 * Submitting request must be dequeued before calling this function
1713 * because it will be linked to another request_queue
1714 */
1715 BUG_ON(blk_queued_rq(rq));
1716
Jens Axboe73c10102011-03-08 13:19:51 +01001717 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001718 spin_unlock_irqrestore(q->queue_lock, flags);
1719
1720 return 0;
1721}
1722EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1723
Tejun Heo80a761f2009-07-03 17:48:17 +09001724/**
1725 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1726 * @rq: request to examine
1727 *
1728 * Description:
1729 * A request could be merge of IOs which require different failure
1730 * handling. This function determines the number of bytes which
1731 * can be failed from the beginning of the request without
1732 * crossing into area which need to be retried further.
1733 *
1734 * Return:
1735 * The number of bytes to fail.
1736 *
1737 * Context:
1738 * queue_lock must be held.
1739 */
1740unsigned int blk_rq_err_bytes(const struct request *rq)
1741{
1742 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1743 unsigned int bytes = 0;
1744 struct bio *bio;
1745
1746 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1747 return blk_rq_bytes(rq);
1748
1749 /*
1750 * Currently the only 'mixing' which can happen is between
1751 * different fastfail types. We can safely fail portions
1752 * which have all the failfast bits that the first one has -
1753 * the ones which are at least as eager to fail as the first
1754 * one.
1755 */
1756 for (bio = rq->bio; bio; bio = bio->bi_next) {
1757 if ((bio->bi_rw & ff) != ff)
1758 break;
1759 bytes += bio->bi_size;
1760 }
1761
1762 /* this could lead to infinite loop */
1763 BUG_ON(blk_rq_bytes(rq) && !bytes);
1764 return bytes;
1765}
1766EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1767
Jens Axboebc58ba92009-01-23 10:54:44 +01001768static void blk_account_io_completion(struct request *req, unsigned int bytes)
1769{
Jens Axboec2553b52009-04-24 08:10:11 +02001770 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001771 const int rw = rq_data_dir(req);
1772 struct hd_struct *part;
1773 int cpu;
1774
1775 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001776 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001777 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1778 part_stat_unlock();
1779 }
1780}
1781
1782static void blk_account_io_done(struct request *req)
1783{
Jens Axboebc58ba92009-01-23 10:54:44 +01001784 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001785 * Account IO completion. flush_rq isn't accounted as a
1786 * normal IO on queueing nor completion. Accounting the
1787 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001788 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001789 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001790 unsigned long duration = jiffies - req->start_time;
1791 const int rw = rq_data_dir(req);
1792 struct hd_struct *part;
1793 int cpu;
1794
1795 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001796 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001797
1798 part_stat_inc(cpu, part, ios[rw]);
1799 part_stat_add(cpu, part, ticks[rw], duration);
1800 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001801 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001802
Jens Axboe6c23a962011-01-07 08:43:37 +01001803 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001804 part_stat_unlock();
1805 }
1806}
1807
Tejun Heo53a08802008-12-03 12:41:26 +01001808/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001809 * blk_peek_request - peek at the top of a request queue
1810 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001811 *
1812 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001813 * Return the request at the top of @q. The returned request
1814 * should be started using blk_start_request() before LLD starts
1815 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001816 *
1817 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001818 * Pointer to the request at the top of @q if available. Null
1819 * otherwise.
1820 *
1821 * Context:
1822 * queue_lock must be held.
1823 */
1824struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001825{
1826 struct request *rq;
1827 int ret;
1828
1829 while ((rq = __elv_next_request(q)) != NULL) {
1830 if (!(rq->cmd_flags & REQ_STARTED)) {
1831 /*
1832 * This is the first time the device driver
1833 * sees this request (possibly after
1834 * requeueing). Notify IO scheduler.
1835 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001836 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001837 elv_activate_rq(q, rq);
1838
1839 /*
1840 * just mark as started even if we don't start
1841 * it, a request that has been delayed should
1842 * not be passed by new incoming requests
1843 */
1844 rq->cmd_flags |= REQ_STARTED;
1845 trace_block_rq_issue(q, rq);
1846 }
1847
1848 if (!q->boundary_rq || q->boundary_rq == rq) {
1849 q->end_sector = rq_end_sector(rq);
1850 q->boundary_rq = NULL;
1851 }
1852
1853 if (rq->cmd_flags & REQ_DONTPREP)
1854 break;
1855
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001856 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001857 /*
1858 * make sure space for the drain appears we
1859 * know we can do this because max_hw_segments
1860 * has been adjusted to be one fewer than the
1861 * device can handle
1862 */
1863 rq->nr_phys_segments++;
1864 }
1865
1866 if (!q->prep_rq_fn)
1867 break;
1868
1869 ret = q->prep_rq_fn(q, rq);
1870 if (ret == BLKPREP_OK) {
1871 break;
1872 } else if (ret == BLKPREP_DEFER) {
1873 /*
1874 * the request may have been (partially) prepped.
1875 * we need to keep this request in the front to
1876 * avoid resource deadlock. REQ_STARTED will
1877 * prevent other fs requests from passing this one.
1878 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001879 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001880 !(rq->cmd_flags & REQ_DONTPREP)) {
1881 /*
1882 * remove the space for the drain we added
1883 * so that we don't add it again
1884 */
1885 --rq->nr_phys_segments;
1886 }
1887
1888 rq = NULL;
1889 break;
1890 } else if (ret == BLKPREP_KILL) {
1891 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001892 /*
1893 * Mark this request as started so we don't trigger
1894 * any debug logic in the end I/O path.
1895 */
1896 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001897 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001898 } else {
1899 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1900 break;
1901 }
1902 }
1903
1904 return rq;
1905}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001906EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001907
Tejun Heo9934c8c2009-05-08 11:54:16 +09001908void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001909{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001910 struct request_queue *q = rq->q;
1911
Tejun Heo158dbda2009-04-23 11:05:18 +09001912 BUG_ON(list_empty(&rq->queuelist));
1913 BUG_ON(ELV_ON_HASH(rq));
1914
1915 list_del_init(&rq->queuelist);
1916
1917 /*
1918 * the time frame between a request being removed from the lists
1919 * and to it is freed is accounted as io that is in progress at
1920 * the driver side.
1921 */
Divyesh Shah91952912010-04-01 15:01:41 -07001922 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001923 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001924 set_io_start_time_ns(rq);
1925 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001926}
1927
Tejun Heo5efccd12009-04-23 11:05:18 +09001928/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001929 * blk_start_request - start request processing on the driver
1930 * @req: request to dequeue
1931 *
1932 * Description:
1933 * Dequeue @req and start timeout timer on it. This hands off the
1934 * request to the driver.
1935 *
1936 * Block internal functions which don't want to start timer should
1937 * call blk_dequeue_request().
1938 *
1939 * Context:
1940 * queue_lock must be held.
1941 */
1942void blk_start_request(struct request *req)
1943{
1944 blk_dequeue_request(req);
1945
1946 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001947 * We are now handing the request to the hardware, initialize
1948 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001949 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001950 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001951 if (unlikely(blk_bidi_rq(req)))
1952 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1953
Tejun Heo9934c8c2009-05-08 11:54:16 +09001954 blk_add_timer(req);
1955}
1956EXPORT_SYMBOL(blk_start_request);
1957
1958/**
1959 * blk_fetch_request - fetch a request from a request queue
1960 * @q: request queue to fetch a request from
1961 *
1962 * Description:
1963 * Return the request at the top of @q. The request is started on
1964 * return and LLD can start processing it immediately.
1965 *
1966 * Return:
1967 * Pointer to the request at the top of @q if available. Null
1968 * otherwise.
1969 *
1970 * Context:
1971 * queue_lock must be held.
1972 */
1973struct request *blk_fetch_request(struct request_queue *q)
1974{
1975 struct request *rq;
1976
1977 rq = blk_peek_request(q);
1978 if (rq)
1979 blk_start_request(rq);
1980 return rq;
1981}
1982EXPORT_SYMBOL(blk_fetch_request);
1983
1984/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001985 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001986 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001987 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001988 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001989 *
1990 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001991 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1992 * the request structure even if @req doesn't have leftover.
1993 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001994 *
1995 * This special helper function is only for request stacking drivers
1996 * (e.g. request-based dm) so that they can handle partial completion.
1997 * Actual device drivers should use blk_end_request instead.
1998 *
1999 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2000 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002001 *
2002 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002003 * %false - this request doesn't have any more data
2004 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002005 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002006bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002008 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 struct bio *bio;
2010
Tejun Heo2e60e022009-04-23 11:05:18 +09002011 if (!req->bio)
2012 return false;
2013
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002014 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002017 * For fs requests, rq is just carrier of independent bio's
2018 * and each partial completion should be handled separately.
2019 * Reset per-request error on each partial completion.
2020 *
2021 * TODO: tj: This is too subtle. It would be better to let
2022 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002024 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 req->errors = 0;
2026
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002027 if (error && req->cmd_type == REQ_TYPE_FS &&
2028 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002029 char *error_type;
2030
2031 switch (error) {
2032 case -ENOLINK:
2033 error_type = "recoverable transport";
2034 break;
2035 case -EREMOTEIO:
2036 error_type = "critical target";
2037 break;
2038 case -EBADE:
2039 error_type = "critical nexus";
2040 break;
2041 case -EIO:
2042 default:
2043 error_type = "I/O";
2044 break;
2045 }
2046 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2047 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2048 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050
Jens Axboebc58ba92009-01-23 10:54:44 +01002051 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 total_bytes = bio_nbytes = 0;
2054 while ((bio = req->bio) != NULL) {
2055 int nbytes;
2056
2057 if (nr_bytes >= bio->bi_size) {
2058 req->bio = bio->bi_next;
2059 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002060 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 next_idx = 0;
2062 bio_nbytes = 0;
2063 } else {
2064 int idx = bio->bi_idx + next_idx;
2065
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002066 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002068 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002069 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 break;
2071 }
2072
2073 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2074 BIO_BUG_ON(nbytes > bio->bi_size);
2075
2076 /*
2077 * not a complete bvec done
2078 */
2079 if (unlikely(nbytes > nr_bytes)) {
2080 bio_nbytes += nr_bytes;
2081 total_bytes += nr_bytes;
2082 break;
2083 }
2084
2085 /*
2086 * advance to the next vector
2087 */
2088 next_idx++;
2089 bio_nbytes += nbytes;
2090 }
2091
2092 total_bytes += nbytes;
2093 nr_bytes -= nbytes;
2094
Jens Axboe6728cb02008-01-31 13:03:55 +01002095 bio = req->bio;
2096 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 /*
2098 * end more in this run, or just return 'not-done'
2099 */
2100 if (unlikely(nr_bytes <= 0))
2101 break;
2102 }
2103 }
2104
2105 /*
2106 * completely done
2107 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002108 if (!req->bio) {
2109 /*
2110 * Reset counters so that the request stacking driver
2111 * can find how many bytes remain in the request
2112 * later.
2113 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002114 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002115 return false;
2116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
2118 /*
2119 * if the request wasn't completed, update state
2120 */
2121 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002122 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 bio->bi_idx += next_idx;
2124 bio_iovec(bio)->bv_offset += nr_bytes;
2125 bio_iovec(bio)->bv_len -= nr_bytes;
2126 }
2127
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002128 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002129 req->buffer = bio_data(req->bio);
2130
2131 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002132 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002133 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002134
Tejun Heo80a761f2009-07-03 17:48:17 +09002135 /* mixed attributes always follow the first bio */
2136 if (req->cmd_flags & REQ_MIXED_MERGE) {
2137 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2138 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2139 }
2140
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002141 /*
2142 * If total number of sectors is less than the first segment
2143 * size, something has gone terribly wrong.
2144 */
2145 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002146 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002147 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002148 }
2149
2150 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002152
Tejun Heo2e60e022009-04-23 11:05:18 +09002153 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154}
Tejun Heo2e60e022009-04-23 11:05:18 +09002155EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
Tejun Heo2e60e022009-04-23 11:05:18 +09002157static bool blk_update_bidi_request(struct request *rq, int error,
2158 unsigned int nr_bytes,
2159 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002160{
Tejun Heo2e60e022009-04-23 11:05:18 +09002161 if (blk_update_request(rq, error, nr_bytes))
2162 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002163
Tejun Heo2e60e022009-04-23 11:05:18 +09002164 /* Bidi request must be completed as a whole */
2165 if (unlikely(blk_bidi_rq(rq)) &&
2166 blk_update_request(rq->next_rq, error, bidi_bytes))
2167 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002168
Jens Axboee2e1a142010-06-09 10:42:09 +02002169 if (blk_queue_add_random(rq->q))
2170 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002171
2172 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173}
2174
James Bottomley28018c22010-07-01 19:49:17 +09002175/**
2176 * blk_unprep_request - unprepare a request
2177 * @req: the request
2178 *
2179 * This function makes a request ready for complete resubmission (or
2180 * completion). It happens only after all error handling is complete,
2181 * so represents the appropriate moment to deallocate any resources
2182 * that were allocated to the request in the prep_rq_fn. The queue
2183 * lock is held when calling this.
2184 */
2185void blk_unprep_request(struct request *req)
2186{
2187 struct request_queue *q = req->q;
2188
2189 req->cmd_flags &= ~REQ_DONTPREP;
2190 if (q->unprep_rq_fn)
2191 q->unprep_rq_fn(q, req);
2192}
2193EXPORT_SYMBOL_GPL(blk_unprep_request);
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195/*
2196 * queue lock must be held
2197 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002198static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002200 if (blk_rq_tagged(req))
2201 blk_queue_end_tag(req->q, req);
2202
James Bottomleyba396a62009-05-27 14:17:08 +02002203 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002205 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002206 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Mike Andersone78042e2008-10-30 02:16:20 -07002208 blk_delete_timer(req);
2209
James Bottomley28018c22010-07-01 19:49:17 +09002210 if (req->cmd_flags & REQ_DONTPREP)
2211 blk_unprep_request(req);
2212
2213
Jens Axboebc58ba92009-01-23 10:54:44 +01002214 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002215
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002217 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002218 else {
2219 if (blk_bidi_rq(req))
2220 __blk_put_request(req->next_rq->q, req->next_rq);
2221
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 }
2224}
2225
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002226/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002227 * blk_end_bidi_request - Complete a bidi request
2228 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002229 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002230 * @nr_bytes: number of bytes to complete @rq
2231 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002232 *
2233 * Description:
2234 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002235 * Drivers that supports bidi can safely call this member for any
2236 * type of request, bidi or uni. In the later case @bidi_bytes is
2237 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002238 *
2239 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002240 * %false - we are done with this request
2241 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002242 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002243static bool blk_end_bidi_request(struct request *rq, int error,
2244 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002245{
2246 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002247 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002248
Tejun Heo2e60e022009-04-23 11:05:18 +09002249 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2250 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002251
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002252 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002253 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002254 spin_unlock_irqrestore(q->queue_lock, flags);
2255
Tejun Heo2e60e022009-04-23 11:05:18 +09002256 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002257}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002258
2259/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002260 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2261 * @rq: the request to complete
2262 * @error: %0 for success, < %0 for error
2263 * @nr_bytes: number of bytes to complete @rq
2264 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002265 *
2266 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002267 * Identical to blk_end_bidi_request() except that queue lock is
2268 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002269 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002270 * Return:
2271 * %false - we are done with this request
2272 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002273 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002274static bool __blk_end_bidi_request(struct request *rq, int error,
2275 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002276{
Tejun Heo2e60e022009-04-23 11:05:18 +09002277 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2278 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002279
Tejun Heo2e60e022009-04-23 11:05:18 +09002280 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002281
Tejun Heo2e60e022009-04-23 11:05:18 +09002282 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002283}
2284
2285/**
2286 * blk_end_request - Helper function for drivers to complete the request.
2287 * @rq: the request being processed
2288 * @error: %0 for success, < %0 for error
2289 * @nr_bytes: number of bytes to complete
2290 *
2291 * Description:
2292 * Ends I/O on a number of bytes attached to @rq.
2293 * If @rq has leftover, sets it up for the next range of segments.
2294 *
2295 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002296 * %false - we are done with this request
2297 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002298 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002299bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002300{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002301 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002302}
Jens Axboe56ad1742009-07-28 22:11:24 +02002303EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002304
2305/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002306 * blk_end_request_all - Helper function for drives to finish the request.
2307 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002308 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002309 *
2310 * Description:
2311 * Completely finish @rq.
2312 */
2313void blk_end_request_all(struct request *rq, int error)
2314{
2315 bool pending;
2316 unsigned int bidi_bytes = 0;
2317
2318 if (unlikely(blk_bidi_rq(rq)))
2319 bidi_bytes = blk_rq_bytes(rq->next_rq);
2320
2321 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2322 BUG_ON(pending);
2323}
Jens Axboe56ad1742009-07-28 22:11:24 +02002324EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002325
2326/**
2327 * blk_end_request_cur - Helper function to finish the current request chunk.
2328 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002329 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002330 *
2331 * Description:
2332 * Complete the current consecutively mapped chunk from @rq.
2333 *
2334 * Return:
2335 * %false - we are done with this request
2336 * %true - still buffers pending for this request
2337 */
2338bool blk_end_request_cur(struct request *rq, int error)
2339{
2340 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2341}
Jens Axboe56ad1742009-07-28 22:11:24 +02002342EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002343
2344/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002345 * blk_end_request_err - Finish a request till the next failure boundary.
2346 * @rq: the request to finish till the next failure boundary for
2347 * @error: must be negative errno
2348 *
2349 * Description:
2350 * Complete @rq till the next failure boundary.
2351 *
2352 * Return:
2353 * %false - we are done with this request
2354 * %true - still buffers pending for this request
2355 */
2356bool blk_end_request_err(struct request *rq, int error)
2357{
2358 WARN_ON(error >= 0);
2359 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2360}
2361EXPORT_SYMBOL_GPL(blk_end_request_err);
2362
2363/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002364 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002365 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002366 * @error: %0 for success, < %0 for error
2367 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002368 *
2369 * Description:
2370 * Must be called with queue lock held unlike blk_end_request().
2371 *
2372 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002373 * %false - we are done with this request
2374 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002375 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002376bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002377{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002378 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002379}
Jens Axboe56ad1742009-07-28 22:11:24 +02002380EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002381
2382/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002383 * __blk_end_request_all - Helper function for drives to finish the request.
2384 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002385 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002386 *
2387 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002388 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002389 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002390void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002391{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002392 bool pending;
2393 unsigned int bidi_bytes = 0;
2394
2395 if (unlikely(blk_bidi_rq(rq)))
2396 bidi_bytes = blk_rq_bytes(rq->next_rq);
2397
2398 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2399 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002400}
Jens Axboe56ad1742009-07-28 22:11:24 +02002401EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002402
2403/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002404 * __blk_end_request_cur - Helper function to finish the current request chunk.
2405 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002406 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002407 *
2408 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002409 * Complete the current consecutively mapped chunk from @rq. Must
2410 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002411 *
2412 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002413 * %false - we are done with this request
2414 * %true - still buffers pending for this request
2415 */
2416bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002417{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002418 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002419}
Jens Axboe56ad1742009-07-28 22:11:24 +02002420EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002421
Tejun Heo80a761f2009-07-03 17:48:17 +09002422/**
2423 * __blk_end_request_err - Finish a request till the next failure boundary.
2424 * @rq: the request to finish till the next failure boundary for
2425 * @error: must be negative errno
2426 *
2427 * Description:
2428 * Complete @rq till the next failure boundary. Must be called
2429 * with queue lock held.
2430 *
2431 * Return:
2432 * %false - we are done with this request
2433 * %true - still buffers pending for this request
2434 */
2435bool __blk_end_request_err(struct request *rq, int error)
2436{
2437 WARN_ON(error >= 0);
2438 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2439}
2440EXPORT_SYMBOL_GPL(__blk_end_request_err);
2441
Jens Axboe86db1e22008-01-29 14:53:40 +01002442void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2443 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002445 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002446 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
David Woodhousefb2dce82008-08-05 18:01:53 +01002448 if (bio_has_data(bio)) {
2449 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002450 rq->buffer = bio_data(bio);
2451 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002452 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
NeilBrown66846572007-08-16 13:31:28 +02002455 if (bio->bi_bdev)
2456 rq->rq_disk = bio->bi_bdev->bd_disk;
2457}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002459#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2460/**
2461 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2462 * @rq: the request to be flushed
2463 *
2464 * Description:
2465 * Flush all pages in @rq.
2466 */
2467void rq_flush_dcache_pages(struct request *rq)
2468{
2469 struct req_iterator iter;
2470 struct bio_vec *bvec;
2471
2472 rq_for_each_segment(bvec, rq, iter)
2473 flush_dcache_page(bvec->bv_page);
2474}
2475EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2476#endif
2477
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002478/**
2479 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2480 * @q : the queue of the device being checked
2481 *
2482 * Description:
2483 * Check if underlying low-level drivers of a device are busy.
2484 * If the drivers want to export their busy state, they must set own
2485 * exporting function using blk_queue_lld_busy() first.
2486 *
2487 * Basically, this function is used only by request stacking drivers
2488 * to stop dispatching requests to underlying devices when underlying
2489 * devices are busy. This behavior helps more I/O merging on the queue
2490 * of the request stacking driver and prevents I/O throughput regression
2491 * on burst I/O load.
2492 *
2493 * Return:
2494 * 0 - Not busy (The request stacking driver should dispatch request)
2495 * 1 - Busy (The request stacking driver should stop dispatching request)
2496 */
2497int blk_lld_busy(struct request_queue *q)
2498{
2499 if (q->lld_busy_fn)
2500 return q->lld_busy_fn(q);
2501
2502 return 0;
2503}
2504EXPORT_SYMBOL_GPL(blk_lld_busy);
2505
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002506/**
2507 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2508 * @rq: the clone request to be cleaned up
2509 *
2510 * Description:
2511 * Free all bios in @rq for a cloned request.
2512 */
2513void blk_rq_unprep_clone(struct request *rq)
2514{
2515 struct bio *bio;
2516
2517 while ((bio = rq->bio) != NULL) {
2518 rq->bio = bio->bi_next;
2519
2520 bio_put(bio);
2521 }
2522}
2523EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2524
2525/*
2526 * Copy attributes of the original request to the clone request.
2527 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2528 */
2529static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2530{
2531 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002532 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002533 dst->cmd_type = src->cmd_type;
2534 dst->__sector = blk_rq_pos(src);
2535 dst->__data_len = blk_rq_bytes(src);
2536 dst->nr_phys_segments = src->nr_phys_segments;
2537 dst->ioprio = src->ioprio;
2538 dst->extra_len = src->extra_len;
2539}
2540
2541/**
2542 * blk_rq_prep_clone - Helper function to setup clone request
2543 * @rq: the request to be setup
2544 * @rq_src: original request to be cloned
2545 * @bs: bio_set that bios for clone are allocated from
2546 * @gfp_mask: memory allocation mask for bio
2547 * @bio_ctr: setup function to be called for each clone bio.
2548 * Returns %0 for success, non %0 for failure.
2549 * @data: private data to be passed to @bio_ctr
2550 *
2551 * Description:
2552 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2553 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2554 * are not copied, and copying such parts is the caller's responsibility.
2555 * Also, pages which the original bios are pointing to are not copied
2556 * and the cloned bios just point same pages.
2557 * So cloned bios must be completed before original bios, which means
2558 * the caller must complete @rq before @rq_src.
2559 */
2560int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2561 struct bio_set *bs, gfp_t gfp_mask,
2562 int (*bio_ctr)(struct bio *, struct bio *, void *),
2563 void *data)
2564{
2565 struct bio *bio, *bio_src;
2566
2567 if (!bs)
2568 bs = fs_bio_set;
2569
2570 blk_rq_init(NULL, rq);
2571
2572 __rq_for_each_bio(bio_src, rq_src) {
2573 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2574 if (!bio)
2575 goto free_and_out;
2576
2577 __bio_clone(bio, bio_src);
2578
2579 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002580 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002581 goto free_and_out;
2582
2583 if (bio_ctr && bio_ctr(bio, bio_src, data))
2584 goto free_and_out;
2585
2586 if (rq->bio) {
2587 rq->biotail->bi_next = bio;
2588 rq->biotail = bio;
2589 } else
2590 rq->bio = rq->biotail = bio;
2591 }
2592
2593 __blk_rq_prep_clone(rq, rq_src);
2594
2595 return 0;
2596
2597free_and_out:
2598 if (bio)
2599 bio_free(bio, bs);
2600 blk_rq_unprep_clone(rq);
2601
2602 return -ENOMEM;
2603}
2604EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2605
Jens Axboe18887ad2008-07-28 13:08:45 +02002606int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
2608 return queue_work(kblockd_workqueue, work);
2609}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610EXPORT_SYMBOL(kblockd_schedule_work);
2611
Vivek Goyale43473b2010-09-15 17:06:35 -04002612int kblockd_schedule_delayed_work(struct request_queue *q,
2613 struct delayed_work *dwork, unsigned long delay)
2614{
2615 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2616}
2617EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2618
Jens Axboe73c10102011-03-08 13:19:51 +01002619#define PLUG_MAGIC 0x91827364
2620
2621void blk_start_plug(struct blk_plug *plug)
2622{
2623 struct task_struct *tsk = current;
2624
2625 plug->magic = PLUG_MAGIC;
2626 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002627 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002628 plug->should_sort = 0;
2629
2630 /*
2631 * If this is a nested plug, don't actually assign it. It will be
2632 * flushed on its own.
2633 */
2634 if (!tsk->plug) {
2635 /*
2636 * Store ordering should not be needed here, since a potential
2637 * preempt will imply a full memory barrier
2638 */
2639 tsk->plug = plug;
2640 }
2641}
2642EXPORT_SYMBOL(blk_start_plug);
2643
2644static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2645{
2646 struct request *rqa = container_of(a, struct request, queuelist);
2647 struct request *rqb = container_of(b, struct request, queuelist);
2648
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002649 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002650}
2651
Jens Axboe49cac012011-04-16 13:51:05 +02002652/*
2653 * If 'from_schedule' is true, then postpone the dispatch of requests
2654 * until a safe kblockd context. We due this to avoid accidental big
2655 * additional stack usage in driver dispatch, in places where the originally
2656 * plugger did not intend it.
2657 */
Jens Axboef6603782011-04-15 15:49:07 +02002658static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002659 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002660 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002661{
Jens Axboe49cac012011-04-16 13:51:05 +02002662 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002663
2664 /*
2665 * If we are punting this to kblockd, then we can safely drop
2666 * the queue_lock before waking kblockd (which needs to take
2667 * this lock).
2668 */
2669 if (from_schedule) {
2670 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002671 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002672 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002673 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002674 spin_unlock(q->queue_lock);
2675 }
2676
Jens Axboe94b5eb22011-04-12 10:12:19 +02002677}
2678
NeilBrown048c9372011-04-18 09:52:22 +02002679static void flush_plug_callbacks(struct blk_plug *plug)
2680{
2681 LIST_HEAD(callbacks);
2682
2683 if (list_empty(&plug->cb_list))
2684 return;
2685
2686 list_splice_init(&plug->cb_list, &callbacks);
2687
2688 while (!list_empty(&callbacks)) {
2689 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2690 struct blk_plug_cb,
2691 list);
2692 list_del(&cb->list);
2693 cb->callback(cb);
2694 }
2695}
2696
Jens Axboe49cac012011-04-16 13:51:05 +02002697void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002698{
2699 struct request_queue *q;
2700 unsigned long flags;
2701 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002702 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002703 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002704
2705 BUG_ON(plug->magic != PLUG_MAGIC);
2706
NeilBrown048c9372011-04-18 09:52:22 +02002707 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002708 if (list_empty(&plug->list))
2709 return;
2710
NeilBrown109b8122011-04-11 14:13:10 +02002711 list_splice_init(&plug->list, &list);
2712
2713 if (plug->should_sort) {
2714 list_sort(NULL, &list, plug_rq_cmp);
2715 plug->should_sort = 0;
2716 }
Jens Axboe73c10102011-03-08 13:19:51 +01002717
2718 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002719 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002720
2721 /*
2722 * Save and disable interrupts here, to avoid doing it for every
2723 * queue lock we have to take.
2724 */
Jens Axboe73c10102011-03-08 13:19:51 +01002725 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002726 while (!list_empty(&list)) {
2727 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002728 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002729 BUG_ON(!rq->q);
2730 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002731 /*
2732 * This drops the queue lock
2733 */
2734 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002735 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002736 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002737 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002738 spin_lock(q->queue_lock);
2739 }
Jens Axboe73c10102011-03-08 13:19:51 +01002740 /*
2741 * rq is already accounted, so use raw insert
2742 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002743 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2744 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2745 else
2746 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002747
2748 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002749 }
2750
Jens Axboe99e22592011-04-18 09:59:55 +02002751 /*
2752 * This drops the queue lock
2753 */
2754 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002755 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002756
Jens Axboe73c10102011-03-08 13:19:51 +01002757 local_irq_restore(flags);
2758}
Jens Axboe73c10102011-03-08 13:19:51 +01002759
2760void blk_finish_plug(struct blk_plug *plug)
2761{
Jens Axboef6603782011-04-15 15:49:07 +02002762 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002763
2764 if (plug == current->plug)
2765 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002766}
2767EXPORT_SYMBOL(blk_finish_plug);
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769int __init blk_dev_init(void)
2770{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002771 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2772 sizeof(((struct request *)0)->cmd_flags));
2773
Tejun Heo89b90be2011-01-03 15:01:47 +01002774 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2775 kblockd_workqueue = alloc_workqueue("kblockd",
2776 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 if (!kblockd_workqueue)
2778 panic("Failed to create kblockd\n");
2779
2780 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002781 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
Jens Axboe8324aa92008-01-29 14:51:59 +01002783 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002784 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 return 0;
2787}