blob: 3fe00a14822a203cacc0529ad0610cf5ce7dae06 [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}
Al Viro483f4af2006-03-18 18:34:37 -0500348
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500349/*
350 * Note: If a driver supplied the queue lock, it should not zap that lock
351 * unexpectedly as some queue cleanup components like elevator_exit() and
352 * blk_throtl_exit() need queue lock.
353 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100354void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500355{
Jens Axboee3335de2008-09-18 09:22:54 -0700356 /*
357 * We know we have process context here, so we can be a little
358 * cautious and ensure that pending block actions on this device
359 * are done before moving on. Going into this function, we should
360 * not have processes doing IO to this device.
361 */
362 blk_sync_queue(q);
363
Matthew Garrett31373d02010-04-06 14:25:14 +0200364 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
Al Viro483f4af2006-03-18 18:34:37 -0500365 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200366 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500367 mutex_unlock(&q->sysfs_lock);
368
369 if (q->elevator)
370 elevator_exit(q->elevator);
371
Vivek Goyalda527772011-03-02 19:05:33 -0500372 blk_throtl_exit(q);
373
Al Viro483f4af2006-03-18 18:34:37 -0500374 blk_put_queue(q);
375}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376EXPORT_SYMBOL(blk_cleanup_queue);
377
Jens Axboe165125e2007-07-24 09:28:11 +0200378static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
380 struct request_list *rl = &q->rq;
381
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400382 if (unlikely(rl->rq_pool))
383 return 0;
384
Jens Axboe1faa16d2009-04-06 14:48:01 +0200385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200387 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Christoph Lameter19460892005-06-23 00:08:19 -0700391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 if (!rl->rq_pool)
395 return -ENOMEM;
396
397 return 0;
398}
399
Jens Axboe165125e2007-07-24 09:28:11 +0200400struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Christoph Lameter19460892005-06-23 00:08:19 -0700402 return blk_alloc_queue_node(gfp_mask, -1);
403}
404EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Jens Axboe165125e2007-07-24 09:28:11 +0200406struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700407{
Jens Axboe165125e2007-07-24 09:28:11 +0200408 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700409 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700410
Jens Axboe8324aa92008-01-29 14:51:59 +0100411 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700412 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (!q)
414 return NULL;
415
Jens Axboe0989a022009-06-12 14:42:56 +0200416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200420 q->backing_dev_info.name = "block";
Jens Axboe0989a022009-06-12 14:42:56 +0200421
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700422 err = bdi_init(&q->backing_dev_info);
423 if (err) {
Jens Axboe8324aa92008-01-29 14:51:59 +0100424 kmem_cache_free(blk_requestq_cachep, q);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700425 return NULL;
426 }
427
Vivek Goyale43473b2010-09-15 17:06:35 -0400428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
430 return NULL;
431 }
432
Matthew Garrett31373d02010-04-06 14:25:14 +0200433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoae1b1532011-01-25 12:43:54 +0100437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500441
Jens Axboe8324aa92008-01-29 14:51:59 +0100442 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Al Viro483f4af2006-03-18 18:34:37 -0500444 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700445 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500446
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500447 /*
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
450 */
451 q->queue_lock = &q->__queue_lock;
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 return q;
454}
Christoph Lameter19460892005-06-23 00:08:19 -0700455EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457/**
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
462 *
463 * Description:
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
472 *
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
477 *
478 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200482 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 * it didn't succeed.
484 *
485 * Note:
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
488 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700489
Jens Axboe165125e2007-07-24 09:28:11 +0200490struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Christoph Lameter19460892005-06-23 00:08:19 -0700492 return blk_init_queue_node(rfn, lock, -1);
493}
494EXPORT_SYMBOL(blk_init_queue);
495
Jens Axboe165125e2007-07-24 09:28:11 +0200496struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700497blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
498{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600499 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
502 if (!uninit_q)
503 return NULL;
504
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
506 if (!q)
507 blk_cleanup_queue(uninit_q);
508
509 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200510}
511EXPORT_SYMBOL(blk_init_queue_node);
512
513struct request_queue *
514blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
515 spinlock_t *lock)
516{
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
518}
519EXPORT_SYMBOL(blk_init_allocated_queue);
520
521struct request_queue *
522blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
524{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (!q)
526 return NULL;
527
Christoph Lameter19460892005-06-23 00:08:19 -0700528 q->node = node_id;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600529 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500530 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
532 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900534 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100535 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500536
537 /* Override internal queue lock with supplied lock pointer */
538 if (lock)
539 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Jens Axboef3b144a2009-03-06 08:48:33 +0100541 /*
542 * This also sets hw/phys segments, boundary and size
543 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 blk_queue_make_request(q, __make_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Alan Stern44ec9542007-02-20 11:01:57 -0500546 q->sg_reserved_size = INT_MAX;
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 /*
549 * all done
550 */
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
553 return q;
554 }
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return NULL;
557}
Mike Snitzer01effb02010-05-11 08:57:42 +0200558EXPORT_SYMBOL(blk_init_allocated_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Jens Axboe165125e2007-07-24 09:28:11 +0200560int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
Nick Pigginfde6ad22005-06-23 00:08:53 -0700562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -0500563 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return 0;
565 }
566
567 return 1;
568}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Jens Axboe165125e2007-07-24 09:28:11 +0200570static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Jens Axboe73c10102011-03-08 13:19:51 +0100572 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
573
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
1113 /*
1114 * Debug stuff, kill later
1115 */
1116 if (!rq_mergeable(req)) {
1117 blk_dump_rq_flags(req, "back");
1118 return false;
1119 }
1120
1121 if (!ll_back_merge_fn(q, req, bio))
1122 return false;
1123
1124 trace_block_bio_backmerge(q, bio);
1125
1126 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1127 blk_rq_set_mixed_merge(req);
1128
1129 req->biotail->bi_next = bio;
1130 req->biotail = bio;
1131 req->__data_len += bio->bi_size;
1132 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1133
1134 drive_stat_acct(req, 0);
1135 return true;
1136}
1137
1138static bool bio_attempt_front_merge(struct request_queue *q,
1139 struct request *req, struct bio *bio)
1140{
1141 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1142 sector_t sector;
1143
1144 /*
1145 * Debug stuff, kill later
1146 */
1147 if (!rq_mergeable(req)) {
1148 blk_dump_rq_flags(req, "front");
1149 return false;
1150 }
1151
1152 if (!ll_front_merge_fn(q, req, bio))
1153 return false;
1154
1155 trace_block_bio_frontmerge(q, bio);
1156
1157 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1158 blk_rq_set_mixed_merge(req);
1159
1160 sector = bio->bi_sector;
1161
1162 bio->bi_next = req->bio;
1163 req->bio = bio;
1164
1165 /*
1166 * may not be valid. if the low level driver said
1167 * it didn't need a bounce buffer then it better
1168 * not touch req->buffer either...
1169 */
1170 req->buffer = bio_data(bio);
1171 req->__sector = bio->bi_sector;
1172 req->__data_len += bio->bi_size;
1173 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1174
1175 drive_stat_acct(req, 0);
1176 return true;
1177}
1178
1179/*
1180 * Attempts to merge with the plugged list in the current process. Returns
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001181 * true if merge was successful, otherwise false.
Jens Axboe73c10102011-03-08 13:19:51 +01001182 */
1183static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1184 struct bio *bio)
1185{
1186 struct blk_plug *plug;
1187 struct request *rq;
1188 bool ret = false;
1189
1190 plug = tsk->plug;
1191 if (!plug)
1192 goto out;
1193
1194 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1195 int el_ret;
1196
1197 if (rq->q != q)
1198 continue;
1199
1200 el_ret = elv_try_merge(rq, bio);
1201 if (el_ret == ELEVATOR_BACK_MERGE) {
1202 ret = bio_attempt_back_merge(q, rq, bio);
1203 if (ret)
1204 break;
1205 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1206 ret = bio_attempt_front_merge(q, rq, bio);
1207 if (ret)
1208 break;
1209 }
1210 }
1211out:
1212 return ret;
1213}
1214
Jens Axboe86db1e22008-01-29 14:53:40 +01001215void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001216{
Jens Axboec7c22e42008-09-13 20:26:01 +02001217 req->cpu = bio->bi_comp_cpu;
Jens Axboe4aff5e22006-08-10 08:44:47 +02001218 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001219
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001220 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1221 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001222 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001223
Tejun Heo52d9e672006-01-06 09:49:58 +01001224 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001225 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001226 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001227 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001228}
1229
Jens Axboe165125e2007-07-24 09:28:11 +02001230static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001232 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001233 struct blk_plug *plug;
1234 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1235 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 /*
1238 * low level driver can indicate that it wants pages above a
1239 * certain limit bounced to low memory (ie for highmem, or even
1240 * ISA dma in theory)
1241 */
1242 blk_queue_bounce(q, &bio);
1243
Tejun Heo4fed9472010-09-03 11:56:17 +02001244 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001245 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001246 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001247 goto get_rq;
1248 }
1249
Jens Axboe73c10102011-03-08 13:19:51 +01001250 /*
1251 * Check if we can merge with the plugged list before grabbing
1252 * any locks.
1253 */
1254 if (attempt_plug_merge(current, q, bio))
1255 goto out;
1256
1257 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001260 if (el_ret == ELEVATOR_BACK_MERGE) {
1261 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1262 if (bio_attempt_back_merge(q, req, bio)) {
1263 if (!attempt_back_merge(q, req))
1264 elv_merged_request(q, req, el_ret);
1265 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001266 }
Jens Axboe73c10102011-03-08 13:19:51 +01001267 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1268 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1269 if (bio_attempt_front_merge(q, req, bio)) {
1270 if (!attempt_front_merge(q, req))
1271 elv_merged_request(q, req, el_ret);
1272 goto out_unlock;
1273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 }
1275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001277 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001278 * This sync check and mask will be re-done in init_request_from_bio(),
1279 * but we need to set it earlier to expose the sync flag to the
1280 * rq allocator and io schedulers.
1281 */
1282 rw_flags = bio_data_dir(bio);
1283 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001284 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001285
1286 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001287 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001288 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001289 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001290 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07001291
Nick Piggin450991b2005-06-28 20:45:13 -07001292 /*
1293 * After dropping the lock and possibly sleeping here, our request
1294 * may now be mergeable after it had proven unmergeable (above).
1295 * We don't worry about that case for efficiency. It won't happen
1296 * often, and the elevators are able to handle it.
1297 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001298 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Jens Axboec7c22e42008-09-13 20:26:01 +02001300 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
Jens Axboe73c10102011-03-08 13:19:51 +01001301 bio_flagged(bio, BIO_CPU_AFFINE)) {
1302 req->cpu = blk_cpu_to_group(get_cpu());
1303 put_cpu();
1304 }
Tejun Heodd831002010-09-03 11:56:16 +02001305
Jens Axboe73c10102011-03-08 13:19:51 +01001306 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001307 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001308 /*
1309 * If this is the first request added after a plug, fire
1310 * of a plug trace. If others have been added before, check
1311 * if we have multiple devices in this plug. If so, make a
1312 * note to sort the list before dispatch.
1313 */
1314 if (list_empty(&plug->list))
1315 trace_block_plug(q);
1316 else if (!plug->should_sort) {
Jens Axboe73c10102011-03-08 13:19:51 +01001317 struct request *__rq;
1318
1319 __rq = list_entry_rq(plug->list.prev);
1320 if (__rq->q != q)
1321 plug->should_sort = 1;
1322 }
1323 /*
1324 * Debug flag, kill later
1325 */
1326 req->cmd_flags |= REQ_ON_PLUG;
1327 list_add_tail(&req->queuelist, &plug->list);
1328 drive_stat_acct(req, 1);
1329 } else {
1330 spin_lock_irq(q->queue_lock);
1331 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001332 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001333out_unlock:
1334 spin_unlock_irq(q->queue_lock);
1335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338}
1339
1340/*
1341 * If bio->bi_dev is a partition, remap the location
1342 */
1343static inline void blk_partition_remap(struct bio *bio)
1344{
1345 struct block_device *bdev = bio->bi_bdev;
1346
Jens Axboebf2de6f2007-09-27 13:01:25 +02001347 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 bio->bi_sector += p->start_sect;
1351 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001352
Mike Snitzerd07335e2010-11-16 12:52:38 +01001353 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1354 bdev->bd_dev,
1355 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357}
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359static void handle_bad_sector(struct bio *bio)
1360{
1361 char b[BDEVNAME_SIZE];
1362
1363 printk(KERN_INFO "attempt to access beyond end of device\n");
1364 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1365 bdevname(bio->bi_bdev, b),
1366 bio->bi_rw,
1367 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001368 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 set_bit(BIO_EOF, &bio->bi_flags);
1371}
1372
Akinobu Mitac17bb492006-12-08 02:39:46 -08001373#ifdef CONFIG_FAIL_MAKE_REQUEST
1374
1375static DECLARE_FAULT_ATTR(fail_make_request);
1376
1377static int __init setup_fail_make_request(char *str)
1378{
1379 return setup_fault_attr(&fail_make_request, str);
1380}
1381__setup("fail_make_request=", setup_fail_make_request);
1382
1383static int should_fail_request(struct bio *bio)
1384{
Tejun Heoeddb2e22008-08-25 19:56:13 +09001385 struct hd_struct *part = bio->bi_bdev->bd_part;
1386
1387 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001388 return should_fail(&fail_make_request, bio->bi_size);
1389
1390 return 0;
1391}
1392
1393static int __init fail_make_request_debugfs(void)
1394{
1395 return init_fault_attr_dentries(&fail_make_request,
1396 "fail_make_request");
1397}
1398
1399late_initcall(fail_make_request_debugfs);
1400
1401#else /* CONFIG_FAIL_MAKE_REQUEST */
1402
1403static inline int should_fail_request(struct bio *bio)
1404{
1405 return 0;
1406}
1407
1408#endif /* CONFIG_FAIL_MAKE_REQUEST */
1409
Jens Axboec07e2b42007-07-18 13:27:58 +02001410/*
1411 * Check whether this bio extends beyond the end of the device.
1412 */
1413static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1414{
1415 sector_t maxsector;
1416
1417 if (!nr_sectors)
1418 return 0;
1419
1420 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001421 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001422 if (maxsector) {
1423 sector_t sector = bio->bi_sector;
1424
1425 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1426 /*
1427 * This may well happen - the kernel calls bread()
1428 * without checking the size of the device, e.g., when
1429 * mounting a device.
1430 */
1431 handle_bad_sector(bio);
1432 return 1;
1433 }
1434 }
1435
1436 return 0;
1437}
1438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001440 * generic_make_request - hand a buffer to its device driver for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 * @bio: The bio describing the location in memory and on the device.
1442 *
1443 * generic_make_request() is used to make I/O requests of block
1444 * devices. It is passed a &struct bio, which describes the I/O that needs
1445 * to be done.
1446 *
1447 * generic_make_request() does not return any status. The
1448 * success/failure status of the request, along with notification of
1449 * completion, is delivered asynchronously through the bio->bi_end_io
1450 * function described (one day) else where.
1451 *
1452 * The caller of generic_make_request must make sure that bi_io_vec
1453 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1454 * set to describe the device address, and the
1455 * bi_end_io and optionally bi_private are set to describe how
1456 * completion notification should be signaled.
1457 *
1458 * generic_make_request and the drivers it calls may use bi_next if this
1459 * bio happens to be merged with someone else, and may change bi_dev and
1460 * bi_sector for remaps as it sees fit. So the values of these fields
1461 * should NOT be depended on after the call to generic_make_request.
1462 */
Neil Brownd89d8792007-05-01 09:53:42 +02001463static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Jens Axboe165125e2007-07-24 09:28:11 +02001465 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08001466 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01001468 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01001469 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470
1471 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Jens Axboec07e2b42007-07-18 13:27:58 +02001473 if (bio_check_eod(bio, nr_sectors))
1474 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 /*
1477 * Resolve the mapping until finished. (drivers are
1478 * still free to implement/resolve their own stacking
1479 * by explicitly returning 0)
1480 *
1481 * NOTE: we don't repeat the blk_size check for each new device.
1482 * Stacking drivers are expected to know what they are doing.
1483 */
NeilBrown5ddfe962006-10-30 22:07:21 -08001484 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01001485 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 do {
1487 char b[BDEVNAME_SIZE];
1488
1489 q = bdev_get_queue(bio->bi_bdev);
Tejun Heoa7384672008-11-28 13:32:03 +09001490 if (unlikely(!q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 printk(KERN_ERR
1492 "generic_make_request: Trying to access "
1493 "nonexistent block-device %s (%Lu)\n",
1494 bdevname(bio->bi_bdev, b),
1495 (long long) bio->bi_sector);
Tejun Heoa7384672008-11-28 13:32:03 +09001496 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
1498
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001499 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
Christoph Hellwig67efc922009-09-30 13:54:20 +02001500 nr_sectors > queue_max_hw_sectors(q))) {
Jens Axboe6728cb02008-01-31 13:03:55 +01001501 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001502 bdevname(bio->bi_bdev, b),
1503 bio_sectors(bio),
1504 queue_max_hw_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 goto end_io;
1506 }
1507
Nick Pigginfde6ad22005-06-23 00:08:53 -07001508 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 goto end_io;
1510
Akinobu Mitac17bb492006-12-08 02:39:46 -08001511 if (should_fail_request(bio))
1512 goto end_io;
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 /*
1515 * If this device has partitions, remap block n
1516 * of partition p to block n+start(p) of the disk.
1517 */
1518 blk_partition_remap(bio);
1519
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001520 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1521 goto end_io;
1522
NeilBrown5ddfe962006-10-30 22:07:21 -08001523 if (old_sector != -1)
Mike Snitzerd07335e2010-11-16 12:52:38 +01001524 trace_block_bio_remap(q, bio, old_dev, old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01001525
NeilBrown5ddfe962006-10-30 22:07:21 -08001526 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01001527 old_dev = bio->bi_bdev->bd_dev;
1528
Jens Axboec07e2b42007-07-18 13:27:58 +02001529 if (bio_check_eod(bio, nr_sectors))
1530 goto end_io;
Tejun Heoa7384672008-11-28 13:32:03 +09001531
Tejun Heo1e879012010-09-03 11:56:17 +02001532 /*
1533 * Filter flush bio's early so that make_request based
1534 * drivers without flush support don't have to worry
1535 * about them.
1536 */
1537 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1538 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1539 if (!nr_sectors) {
1540 err = 0;
1541 goto end_io;
1542 }
1543 }
1544
Adrian Hunter8d57a982010-08-11 14:17:49 -07001545 if ((bio->bi_rw & REQ_DISCARD) &&
1546 (!blk_queue_discard(q) ||
1547 ((bio->bi_rw & REQ_SECURE) &&
1548 !blk_queue_secdiscard(q)))) {
Jens Axboe51fd77b2007-11-02 08:49:08 +01001549 err = -EOPNOTSUPP;
1550 goto end_io;
1551 }
NeilBrown5ddfe962006-10-30 22:07:21 -08001552
Vivek Goyale43473b2010-09-15 17:06:35 -04001553 blk_throtl_bio(q, &bio);
1554
1555 /*
1556 * If bio = NULL, bio has been throttled and will be submitted
1557 * later.
1558 */
1559 if (!bio)
1560 break;
1561
Minchan Kim01edede2009-09-08 21:56:38 +02001562 trace_block_bio_queue(q, bio);
1563
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 ret = q->make_request_fn(q, bio);
1565 } while (ret);
Tejun Heoa7384672008-11-28 13:32:03 +09001566
1567 return;
1568
1569end_io:
1570 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
1572
Neil Brownd89d8792007-05-01 09:53:42 +02001573/*
1574 * We only want one ->make_request_fn to be active at a time,
1575 * else stack usage with stacked devices could be a problem.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001576 * So use current->bio_list to keep a list of requests
Neil Brownd89d8792007-05-01 09:53:42 +02001577 * submited by a make_request_fn function.
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001578 * current->bio_list is also used as a flag to say if
Neil Brownd89d8792007-05-01 09:53:42 +02001579 * generic_make_request is currently active in this task or not.
1580 * If it is NULL, then no make_request is active. If it is non-NULL,
1581 * then a make_request is active, and new requests should be added
1582 * at the tail
1583 */
1584void generic_make_request(struct bio *bio)
1585{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001586 struct bio_list bio_list_on_stack;
1587
1588 if (current->bio_list) {
Neil Brownd89d8792007-05-01 09:53:42 +02001589 /* make_request is active */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001590 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001591 return;
1592 }
1593 /* following loop may be a bit non-obvious, and so deserves some
1594 * explanation.
1595 * Before entering the loop, bio->bi_next is NULL (as all callers
1596 * ensure that) so we have a list with a single bio.
1597 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001598 * we assign bio_list to a pointer to the bio_list_on_stack,
1599 * thus initialising the bio_list of new bios to be
Neil Brownd89d8792007-05-01 09:53:42 +02001600 * added. __generic_make_request may indeed add some more bios
1601 * through a recursive call to generic_make_request. If it
1602 * did, we find a non-NULL value in bio_list and re-enter the loop
1603 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001604 * of the top of the list (no pretending) and so remove it from
1605 * bio_list, and call into __generic_make_request again.
Neil Brownd89d8792007-05-01 09:53:42 +02001606 *
1607 * The loop was structured like this to make only one call to
1608 * __generic_make_request (which is important as it is large and
1609 * inlined) and to keep the structure simple.
1610 */
1611 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001612 bio_list_init(&bio_list_on_stack);
1613 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001614 do {
Neil Brownd89d8792007-05-01 09:53:42 +02001615 __generic_make_request(bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001616 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001617 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001618 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620EXPORT_SYMBOL(generic_make_request);
1621
1622/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001623 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1625 * @bio: The &struct bio which describes the I/O
1626 *
1627 * submit_bio() is very similar in purpose to generic_make_request(), and
1628 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001629 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 *
1631 */
1632void submit_bio(int rw, struct bio *bio)
1633{
1634 int count = bio_sectors(bio);
1635
Jens Axboe22e2c502005-06-27 10:55:12 +02001636 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Jens Axboebf2de6f2007-09-27 13:01:25 +02001638 /*
1639 * If it's a regular read/write or a barrier with data attached,
1640 * go through the normal accounting stuff before submission.
1641 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001642 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001643 if (rw & WRITE) {
1644 count_vm_events(PGPGOUT, count);
1645 } else {
1646 task_io_account_read(bio->bi_size);
1647 count_vm_events(PGPGIN, count);
1648 }
1649
1650 if (unlikely(block_dump)) {
1651 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001652 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001653 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001654 (rw & WRITE) ? "WRITE" : "READ",
1655 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001656 bdevname(bio->bi_bdev, b),
1657 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
1660
1661 generic_make_request(bio);
1662}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663EXPORT_SYMBOL(submit_bio);
1664
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001665/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001666 * blk_rq_check_limits - Helper function to check a request for the queue limit
1667 * @q: the queue
1668 * @rq: the request being checked
1669 *
1670 * Description:
1671 * @rq may have been made based on weaker limitations of upper-level queues
1672 * in request stacking drivers, and it may violate the limitation of @q.
1673 * Since the block layer and the underlying device driver trust @rq
1674 * after it is inserted to @q, it should be checked against @q before
1675 * the insertion using this generic function.
1676 *
1677 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001678 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001679 * Request stacking drivers like request-based dm may change the queue
1680 * limits while requests are in the queue (e.g. dm's table swapping).
1681 * Such request stacking drivers should check those requests agaist
1682 * the new queue limits again when they dispatch those requests,
1683 * although such checkings are also done against the old queue limits
1684 * when submitting requests.
1685 */
1686int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1687{
ike Snitzer33839772010-08-08 12:11:33 -04001688 if (rq->cmd_flags & REQ_DISCARD)
1689 return 0;
1690
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001691 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1692 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001693 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1694 return -EIO;
1695 }
1696
1697 /*
1698 * queue's settings related to segment counting like q->bounce_pfn
1699 * may differ from that of other stacking queues.
1700 * Recalculate it to check the request correctly on this queue's
1701 * limitation.
1702 */
1703 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001704 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001705 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1706 return -EIO;
1707 }
1708
1709 return 0;
1710}
1711EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1712
1713/**
1714 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1715 * @q: the queue to submit the request
1716 * @rq: the request being queued
1717 */
1718int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1719{
1720 unsigned long flags;
1721
1722 if (blk_rq_check_limits(q, rq))
1723 return -EIO;
1724
1725#ifdef CONFIG_FAIL_MAKE_REQUEST
1726 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1727 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1728 return -EIO;
1729#endif
1730
1731 spin_lock_irqsave(q->queue_lock, flags);
1732
1733 /*
1734 * Submitting request must be dequeued before calling this function
1735 * because it will be linked to another request_queue
1736 */
1737 BUG_ON(blk_queued_rq(rq));
1738
Jens Axboe73c10102011-03-08 13:19:51 +01001739 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001740 spin_unlock_irqrestore(q->queue_lock, flags);
1741
1742 return 0;
1743}
1744EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1745
Tejun Heo80a761f2009-07-03 17:48:17 +09001746/**
1747 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1748 * @rq: request to examine
1749 *
1750 * Description:
1751 * A request could be merge of IOs which require different failure
1752 * handling. This function determines the number of bytes which
1753 * can be failed from the beginning of the request without
1754 * crossing into area which need to be retried further.
1755 *
1756 * Return:
1757 * The number of bytes to fail.
1758 *
1759 * Context:
1760 * queue_lock must be held.
1761 */
1762unsigned int blk_rq_err_bytes(const struct request *rq)
1763{
1764 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1765 unsigned int bytes = 0;
1766 struct bio *bio;
1767
1768 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1769 return blk_rq_bytes(rq);
1770
1771 /*
1772 * Currently the only 'mixing' which can happen is between
1773 * different fastfail types. We can safely fail portions
1774 * which have all the failfast bits that the first one has -
1775 * the ones which are at least as eager to fail as the first
1776 * one.
1777 */
1778 for (bio = rq->bio; bio; bio = bio->bi_next) {
1779 if ((bio->bi_rw & ff) != ff)
1780 break;
1781 bytes += bio->bi_size;
1782 }
1783
1784 /* this could lead to infinite loop */
1785 BUG_ON(blk_rq_bytes(rq) && !bytes);
1786 return bytes;
1787}
1788EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1789
Jens Axboebc58ba92009-01-23 10:54:44 +01001790static void blk_account_io_completion(struct request *req, unsigned int bytes)
1791{
Jens Axboec2553b52009-04-24 08:10:11 +02001792 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001793 const int rw = rq_data_dir(req);
1794 struct hd_struct *part;
1795 int cpu;
1796
1797 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001798 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001799 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1800 part_stat_unlock();
1801 }
1802}
1803
1804static void blk_account_io_done(struct request *req)
1805{
Jens Axboebc58ba92009-01-23 10:54:44 +01001806 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001807 * Account IO completion. flush_rq isn't accounted as a
1808 * normal IO on queueing nor completion. Accounting the
1809 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001810 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001811 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001812 unsigned long duration = jiffies - req->start_time;
1813 const int rw = rq_data_dir(req);
1814 struct hd_struct *part;
1815 int cpu;
1816
1817 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001818 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001819
1820 part_stat_inc(cpu, part, ios[rw]);
1821 part_stat_add(cpu, part, ticks[rw], duration);
1822 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001823 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001824
Jens Axboe6c23a962011-01-07 08:43:37 +01001825 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001826 part_stat_unlock();
1827 }
1828}
1829
Tejun Heo53a08802008-12-03 12:41:26 +01001830/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001831 * blk_peek_request - peek at the top of a request queue
1832 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001833 *
1834 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001835 * Return the request at the top of @q. The returned request
1836 * should be started using blk_start_request() before LLD starts
1837 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001838 *
1839 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001840 * Pointer to the request at the top of @q if available. Null
1841 * otherwise.
1842 *
1843 * Context:
1844 * queue_lock must be held.
1845 */
1846struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09001847{
1848 struct request *rq;
1849 int ret;
1850
1851 while ((rq = __elv_next_request(q)) != NULL) {
1852 if (!(rq->cmd_flags & REQ_STARTED)) {
1853 /*
1854 * This is the first time the device driver
1855 * sees this request (possibly after
1856 * requeueing). Notify IO scheduler.
1857 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001858 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09001859 elv_activate_rq(q, rq);
1860
1861 /*
1862 * just mark as started even if we don't start
1863 * it, a request that has been delayed should
1864 * not be passed by new incoming requests
1865 */
1866 rq->cmd_flags |= REQ_STARTED;
1867 trace_block_rq_issue(q, rq);
1868 }
1869
1870 if (!q->boundary_rq || q->boundary_rq == rq) {
1871 q->end_sector = rq_end_sector(rq);
1872 q->boundary_rq = NULL;
1873 }
1874
1875 if (rq->cmd_flags & REQ_DONTPREP)
1876 break;
1877
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001878 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09001879 /*
1880 * make sure space for the drain appears we
1881 * know we can do this because max_hw_segments
1882 * has been adjusted to be one fewer than the
1883 * device can handle
1884 */
1885 rq->nr_phys_segments++;
1886 }
1887
1888 if (!q->prep_rq_fn)
1889 break;
1890
1891 ret = q->prep_rq_fn(q, rq);
1892 if (ret == BLKPREP_OK) {
1893 break;
1894 } else if (ret == BLKPREP_DEFER) {
1895 /*
1896 * the request may have been (partially) prepped.
1897 * we need to keep this request in the front to
1898 * avoid resource deadlock. REQ_STARTED will
1899 * prevent other fs requests from passing this one.
1900 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001901 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09001902 !(rq->cmd_flags & REQ_DONTPREP)) {
1903 /*
1904 * remove the space for the drain we added
1905 * so that we don't add it again
1906 */
1907 --rq->nr_phys_segments;
1908 }
1909
1910 rq = NULL;
1911 break;
1912 } else if (ret == BLKPREP_KILL) {
1913 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02001914 /*
1915 * Mark this request as started so we don't trigger
1916 * any debug logic in the end I/O path.
1917 */
1918 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09001919 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09001920 } else {
1921 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1922 break;
1923 }
1924 }
1925
1926 return rq;
1927}
Tejun Heo9934c8c2009-05-08 11:54:16 +09001928EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09001929
Tejun Heo9934c8c2009-05-08 11:54:16 +09001930void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09001931{
Tejun Heo9934c8c2009-05-08 11:54:16 +09001932 struct request_queue *q = rq->q;
1933
Tejun Heo158dbda2009-04-23 11:05:18 +09001934 BUG_ON(list_empty(&rq->queuelist));
1935 BUG_ON(ELV_ON_HASH(rq));
1936
1937 list_del_init(&rq->queuelist);
1938
1939 /*
1940 * the time frame between a request being removed from the lists
1941 * and to it is freed is accounted as io that is in progress at
1942 * the driver side.
1943 */
Divyesh Shah91952912010-04-01 15:01:41 -07001944 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02001945 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07001946 set_io_start_time_ns(rq);
1947 }
Tejun Heo158dbda2009-04-23 11:05:18 +09001948}
1949
Tejun Heo5efccd12009-04-23 11:05:18 +09001950/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001951 * blk_start_request - start request processing on the driver
1952 * @req: request to dequeue
1953 *
1954 * Description:
1955 * Dequeue @req and start timeout timer on it. This hands off the
1956 * request to the driver.
1957 *
1958 * Block internal functions which don't want to start timer should
1959 * call blk_dequeue_request().
1960 *
1961 * Context:
1962 * queue_lock must be held.
1963 */
1964void blk_start_request(struct request *req)
1965{
1966 blk_dequeue_request(req);
1967
1968 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09001969 * We are now handing the request to the hardware, initialize
1970 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09001971 */
Tejun Heo5f49f632009-05-19 18:33:05 +09001972 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02001973 if (unlikely(blk_bidi_rq(req)))
1974 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1975
Tejun Heo9934c8c2009-05-08 11:54:16 +09001976 blk_add_timer(req);
1977}
1978EXPORT_SYMBOL(blk_start_request);
1979
1980/**
1981 * blk_fetch_request - fetch a request from a request queue
1982 * @q: request queue to fetch a request from
1983 *
1984 * Description:
1985 * Return the request at the top of @q. The request is started on
1986 * return and LLD can start processing it immediately.
1987 *
1988 * Return:
1989 * Pointer to the request at the top of @q if available. Null
1990 * otherwise.
1991 *
1992 * Context:
1993 * queue_lock must be held.
1994 */
1995struct request *blk_fetch_request(struct request_queue *q)
1996{
1997 struct request *rq;
1998
1999 rq = blk_peek_request(q);
2000 if (rq)
2001 blk_start_request(rq);
2002 return rq;
2003}
2004EXPORT_SYMBOL(blk_fetch_request);
2005
2006/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002007 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002008 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002009 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002010 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002011 *
2012 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002013 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2014 * the request structure even if @req doesn't have leftover.
2015 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002016 *
2017 * This special helper function is only for request stacking drivers
2018 * (e.g. request-based dm) so that they can handle partial completion.
2019 * Actual device drivers should use blk_end_request instead.
2020 *
2021 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2022 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002023 *
2024 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002025 * %false - this request doesn't have any more data
2026 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002027 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002028bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002030 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 struct bio *bio;
2032
Tejun Heo2e60e022009-04-23 11:05:18 +09002033 if (!req->bio)
2034 return false;
2035
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002036 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002039 * For fs requests, rq is just carrier of independent bio's
2040 * and each partial completion should be handled separately.
2041 * Reset per-request error on each partial completion.
2042 *
2043 * TODO: tj: This is too subtle. It would be better to let
2044 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002046 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 req->errors = 0;
2048
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002049 if (error && req->cmd_type == REQ_TYPE_FS &&
2050 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002051 char *error_type;
2052
2053 switch (error) {
2054 case -ENOLINK:
2055 error_type = "recoverable transport";
2056 break;
2057 case -EREMOTEIO:
2058 error_type = "critical target";
2059 break;
2060 case -EBADE:
2061 error_type = "critical nexus";
2062 break;
2063 case -EIO:
2064 default:
2065 error_type = "I/O";
2066 break;
2067 }
2068 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2069 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2070 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 }
2072
Jens Axboebc58ba92009-01-23 10:54:44 +01002073 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 total_bytes = bio_nbytes = 0;
2076 while ((bio = req->bio) != NULL) {
2077 int nbytes;
2078
2079 if (nr_bytes >= bio->bi_size) {
2080 req->bio = bio->bi_next;
2081 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002082 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 next_idx = 0;
2084 bio_nbytes = 0;
2085 } else {
2086 int idx = bio->bi_idx + next_idx;
2087
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002088 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002090 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002091 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 break;
2093 }
2094
2095 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2096 BIO_BUG_ON(nbytes > bio->bi_size);
2097
2098 /*
2099 * not a complete bvec done
2100 */
2101 if (unlikely(nbytes > nr_bytes)) {
2102 bio_nbytes += nr_bytes;
2103 total_bytes += nr_bytes;
2104 break;
2105 }
2106
2107 /*
2108 * advance to the next vector
2109 */
2110 next_idx++;
2111 bio_nbytes += nbytes;
2112 }
2113
2114 total_bytes += nbytes;
2115 nr_bytes -= nbytes;
2116
Jens Axboe6728cb02008-01-31 13:03:55 +01002117 bio = req->bio;
2118 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 /*
2120 * end more in this run, or just return 'not-done'
2121 */
2122 if (unlikely(nr_bytes <= 0))
2123 break;
2124 }
2125 }
2126
2127 /*
2128 * completely done
2129 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002130 if (!req->bio) {
2131 /*
2132 * Reset counters so that the request stacking driver
2133 * can find how many bytes remain in the request
2134 * later.
2135 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002136 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002137 return false;
2138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
2140 /*
2141 * if the request wasn't completed, update state
2142 */
2143 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002144 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 bio->bi_idx += next_idx;
2146 bio_iovec(bio)->bv_offset += nr_bytes;
2147 bio_iovec(bio)->bv_len -= nr_bytes;
2148 }
2149
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002150 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002151 req->buffer = bio_data(req->bio);
2152
2153 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002154 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002155 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002156
Tejun Heo80a761f2009-07-03 17:48:17 +09002157 /* mixed attributes always follow the first bio */
2158 if (req->cmd_flags & REQ_MIXED_MERGE) {
2159 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2160 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2161 }
2162
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002163 /*
2164 * If total number of sectors is less than the first segment
2165 * size, something has gone terribly wrong.
2166 */
2167 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002168 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002169 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002170 }
2171
2172 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002174
Tejun Heo2e60e022009-04-23 11:05:18 +09002175 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176}
Tejun Heo2e60e022009-04-23 11:05:18 +09002177EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
Tejun Heo2e60e022009-04-23 11:05:18 +09002179static bool blk_update_bidi_request(struct request *rq, int error,
2180 unsigned int nr_bytes,
2181 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002182{
Tejun Heo2e60e022009-04-23 11:05:18 +09002183 if (blk_update_request(rq, error, nr_bytes))
2184 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002185
Tejun Heo2e60e022009-04-23 11:05:18 +09002186 /* Bidi request must be completed as a whole */
2187 if (unlikely(blk_bidi_rq(rq)) &&
2188 blk_update_request(rq->next_rq, error, bidi_bytes))
2189 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002190
Jens Axboee2e1a142010-06-09 10:42:09 +02002191 if (blk_queue_add_random(rq->q))
2192 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002193
2194 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195}
2196
James Bottomley28018c22010-07-01 19:49:17 +09002197/**
2198 * blk_unprep_request - unprepare a request
2199 * @req: the request
2200 *
2201 * This function makes a request ready for complete resubmission (or
2202 * completion). It happens only after all error handling is complete,
2203 * so represents the appropriate moment to deallocate any resources
2204 * that were allocated to the request in the prep_rq_fn. The queue
2205 * lock is held when calling this.
2206 */
2207void blk_unprep_request(struct request *req)
2208{
2209 struct request_queue *q = req->q;
2210
2211 req->cmd_flags &= ~REQ_DONTPREP;
2212 if (q->unprep_rq_fn)
2213 q->unprep_rq_fn(q, req);
2214}
2215EXPORT_SYMBOL_GPL(blk_unprep_request);
2216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217/*
2218 * queue lock must be held
2219 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002220static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002222 if (blk_rq_tagged(req))
2223 blk_queue_end_tag(req->q, req);
2224
James Bottomleyba396a62009-05-27 14:17:08 +02002225 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002227 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002228 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Mike Andersone78042e2008-10-30 02:16:20 -07002230 blk_delete_timer(req);
2231
James Bottomley28018c22010-07-01 19:49:17 +09002232 if (req->cmd_flags & REQ_DONTPREP)
2233 blk_unprep_request(req);
2234
2235
Jens Axboebc58ba92009-01-23 10:54:44 +01002236 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002239 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002240 else {
2241 if (blk_bidi_rq(req))
2242 __blk_put_request(req->next_rq->q, req->next_rq);
2243
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
2246}
2247
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002248/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002249 * blk_end_bidi_request - Complete a bidi request
2250 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002251 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002252 * @nr_bytes: number of bytes to complete @rq
2253 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002254 *
2255 * Description:
2256 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002257 * Drivers that supports bidi can safely call this member for any
2258 * type of request, bidi or uni. In the later case @bidi_bytes is
2259 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002260 *
2261 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002262 * %false - we are done with this request
2263 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002264 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002265static bool blk_end_bidi_request(struct request *rq, int error,
2266 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002267{
2268 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002269 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002270
Tejun Heo2e60e022009-04-23 11:05:18 +09002271 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2272 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002273
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002274 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002275 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002276 spin_unlock_irqrestore(q->queue_lock, flags);
2277
Tejun Heo2e60e022009-04-23 11:05:18 +09002278 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002279}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002280
2281/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002282 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2283 * @rq: the request to complete
2284 * @error: %0 for success, < %0 for error
2285 * @nr_bytes: number of bytes to complete @rq
2286 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002287 *
2288 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002289 * Identical to blk_end_bidi_request() except that queue lock is
2290 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002291 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002292 * Return:
2293 * %false - we are done with this request
2294 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002295 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002296static bool __blk_end_bidi_request(struct request *rq, int error,
2297 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002298{
Tejun Heo2e60e022009-04-23 11:05:18 +09002299 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2300 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002301
Tejun Heo2e60e022009-04-23 11:05:18 +09002302 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002303
Tejun Heo2e60e022009-04-23 11:05:18 +09002304 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002305}
2306
2307/**
2308 * blk_end_request - Helper function for drivers to complete the request.
2309 * @rq: the request being processed
2310 * @error: %0 for success, < %0 for error
2311 * @nr_bytes: number of bytes to complete
2312 *
2313 * Description:
2314 * Ends I/O on a number of bytes attached to @rq.
2315 * If @rq has leftover, sets it up for the next range of segments.
2316 *
2317 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002318 * %false - we are done with this request
2319 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002320 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002321bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002322{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002323 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002324}
Jens Axboe56ad1742009-07-28 22:11:24 +02002325EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002326
2327/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002328 * blk_end_request_all - Helper function for drives to finish the request.
2329 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002330 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002331 *
2332 * Description:
2333 * Completely finish @rq.
2334 */
2335void blk_end_request_all(struct request *rq, int error)
2336{
2337 bool pending;
2338 unsigned int bidi_bytes = 0;
2339
2340 if (unlikely(blk_bidi_rq(rq)))
2341 bidi_bytes = blk_rq_bytes(rq->next_rq);
2342
2343 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2344 BUG_ON(pending);
2345}
Jens Axboe56ad1742009-07-28 22:11:24 +02002346EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002347
2348/**
2349 * blk_end_request_cur - Helper function to finish the current request chunk.
2350 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002351 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002352 *
2353 * Description:
2354 * Complete the current consecutively mapped chunk from @rq.
2355 *
2356 * Return:
2357 * %false - we are done with this request
2358 * %true - still buffers pending for this request
2359 */
2360bool blk_end_request_cur(struct request *rq, int error)
2361{
2362 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2363}
Jens Axboe56ad1742009-07-28 22:11:24 +02002364EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002365
2366/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002367 * blk_end_request_err - Finish a request till the next failure boundary.
2368 * @rq: the request to finish till the next failure boundary for
2369 * @error: must be negative errno
2370 *
2371 * Description:
2372 * Complete @rq till the next failure boundary.
2373 *
2374 * Return:
2375 * %false - we are done with this request
2376 * %true - still buffers pending for this request
2377 */
2378bool blk_end_request_err(struct request *rq, int error)
2379{
2380 WARN_ON(error >= 0);
2381 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2382}
2383EXPORT_SYMBOL_GPL(blk_end_request_err);
2384
2385/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002386 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002387 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002388 * @error: %0 for success, < %0 for error
2389 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002390 *
2391 * Description:
2392 * Must be called with queue lock held unlike blk_end_request().
2393 *
2394 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002395 * %false - we are done with this request
2396 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002397 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002398bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002399{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002400 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002401}
Jens Axboe56ad1742009-07-28 22:11:24 +02002402EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002403
2404/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002405 * __blk_end_request_all - Helper function for drives to finish the request.
2406 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002407 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002408 *
2409 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002410 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002411 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002412void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002413{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002414 bool pending;
2415 unsigned int bidi_bytes = 0;
2416
2417 if (unlikely(blk_bidi_rq(rq)))
2418 bidi_bytes = blk_rq_bytes(rq->next_rq);
2419
2420 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2421 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002422}
Jens Axboe56ad1742009-07-28 22:11:24 +02002423EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002424
2425/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002426 * __blk_end_request_cur - Helper function to finish the current request chunk.
2427 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002428 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002429 *
2430 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002431 * Complete the current consecutively mapped chunk from @rq. Must
2432 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002433 *
2434 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002435 * %false - we are done with this request
2436 * %true - still buffers pending for this request
2437 */
2438bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002439{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002440 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002441}
Jens Axboe56ad1742009-07-28 22:11:24 +02002442EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002443
Tejun Heo80a761f2009-07-03 17:48:17 +09002444/**
2445 * __blk_end_request_err - Finish a request till the next failure boundary.
2446 * @rq: the request to finish till the next failure boundary for
2447 * @error: must be negative errno
2448 *
2449 * Description:
2450 * Complete @rq till the next failure boundary. Must be called
2451 * with queue lock held.
2452 *
2453 * Return:
2454 * %false - we are done with this request
2455 * %true - still buffers pending for this request
2456 */
2457bool __blk_end_request_err(struct request *rq, int error)
2458{
2459 WARN_ON(error >= 0);
2460 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2461}
2462EXPORT_SYMBOL_GPL(__blk_end_request_err);
2463
Jens Axboe86db1e22008-01-29 14:53:40 +01002464void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2465 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002467 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002468 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
David Woodhousefb2dce82008-08-05 18:01:53 +01002470 if (bio_has_data(bio)) {
2471 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002472 rq->buffer = bio_data(bio);
2473 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002474 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
NeilBrown66846572007-08-16 13:31:28 +02002477 if (bio->bi_bdev)
2478 rq->rq_disk = bio->bi_bdev->bd_disk;
2479}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002481#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2482/**
2483 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2484 * @rq: the request to be flushed
2485 *
2486 * Description:
2487 * Flush all pages in @rq.
2488 */
2489void rq_flush_dcache_pages(struct request *rq)
2490{
2491 struct req_iterator iter;
2492 struct bio_vec *bvec;
2493
2494 rq_for_each_segment(bvec, rq, iter)
2495 flush_dcache_page(bvec->bv_page);
2496}
2497EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2498#endif
2499
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002500/**
2501 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2502 * @q : the queue of the device being checked
2503 *
2504 * Description:
2505 * Check if underlying low-level drivers of a device are busy.
2506 * If the drivers want to export their busy state, they must set own
2507 * exporting function using blk_queue_lld_busy() first.
2508 *
2509 * Basically, this function is used only by request stacking drivers
2510 * to stop dispatching requests to underlying devices when underlying
2511 * devices are busy. This behavior helps more I/O merging on the queue
2512 * of the request stacking driver and prevents I/O throughput regression
2513 * on burst I/O load.
2514 *
2515 * Return:
2516 * 0 - Not busy (The request stacking driver should dispatch request)
2517 * 1 - Busy (The request stacking driver should stop dispatching request)
2518 */
2519int blk_lld_busy(struct request_queue *q)
2520{
2521 if (q->lld_busy_fn)
2522 return q->lld_busy_fn(q);
2523
2524 return 0;
2525}
2526EXPORT_SYMBOL_GPL(blk_lld_busy);
2527
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002528/**
2529 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2530 * @rq: the clone request to be cleaned up
2531 *
2532 * Description:
2533 * Free all bios in @rq for a cloned request.
2534 */
2535void blk_rq_unprep_clone(struct request *rq)
2536{
2537 struct bio *bio;
2538
2539 while ((bio = rq->bio) != NULL) {
2540 rq->bio = bio->bi_next;
2541
2542 bio_put(bio);
2543 }
2544}
2545EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2546
2547/*
2548 * Copy attributes of the original request to the clone request.
2549 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2550 */
2551static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2552{
2553 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002554 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002555 dst->cmd_type = src->cmd_type;
2556 dst->__sector = blk_rq_pos(src);
2557 dst->__data_len = blk_rq_bytes(src);
2558 dst->nr_phys_segments = src->nr_phys_segments;
2559 dst->ioprio = src->ioprio;
2560 dst->extra_len = src->extra_len;
2561}
2562
2563/**
2564 * blk_rq_prep_clone - Helper function to setup clone request
2565 * @rq: the request to be setup
2566 * @rq_src: original request to be cloned
2567 * @bs: bio_set that bios for clone are allocated from
2568 * @gfp_mask: memory allocation mask for bio
2569 * @bio_ctr: setup function to be called for each clone bio.
2570 * Returns %0 for success, non %0 for failure.
2571 * @data: private data to be passed to @bio_ctr
2572 *
2573 * Description:
2574 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2575 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2576 * are not copied, and copying such parts is the caller's responsibility.
2577 * Also, pages which the original bios are pointing to are not copied
2578 * and the cloned bios just point same pages.
2579 * So cloned bios must be completed before original bios, which means
2580 * the caller must complete @rq before @rq_src.
2581 */
2582int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2583 struct bio_set *bs, gfp_t gfp_mask,
2584 int (*bio_ctr)(struct bio *, struct bio *, void *),
2585 void *data)
2586{
2587 struct bio *bio, *bio_src;
2588
2589 if (!bs)
2590 bs = fs_bio_set;
2591
2592 blk_rq_init(NULL, rq);
2593
2594 __rq_for_each_bio(bio_src, rq_src) {
2595 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2596 if (!bio)
2597 goto free_and_out;
2598
2599 __bio_clone(bio, bio_src);
2600
2601 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002602 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002603 goto free_and_out;
2604
2605 if (bio_ctr && bio_ctr(bio, bio_src, data))
2606 goto free_and_out;
2607
2608 if (rq->bio) {
2609 rq->biotail->bi_next = bio;
2610 rq->biotail = bio;
2611 } else
2612 rq->bio = rq->biotail = bio;
2613 }
2614
2615 __blk_rq_prep_clone(rq, rq_src);
2616
2617 return 0;
2618
2619free_and_out:
2620 if (bio)
2621 bio_free(bio, bs);
2622 blk_rq_unprep_clone(rq);
2623
2624 return -ENOMEM;
2625}
2626EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2627
Jens Axboe18887ad2008-07-28 13:08:45 +02002628int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
2630 return queue_work(kblockd_workqueue, work);
2631}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632EXPORT_SYMBOL(kblockd_schedule_work);
2633
Vivek Goyale43473b2010-09-15 17:06:35 -04002634int kblockd_schedule_delayed_work(struct request_queue *q,
2635 struct delayed_work *dwork, unsigned long delay)
2636{
2637 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2638}
2639EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2640
Jens Axboe73c10102011-03-08 13:19:51 +01002641#define PLUG_MAGIC 0x91827364
2642
2643void blk_start_plug(struct blk_plug *plug)
2644{
2645 struct task_struct *tsk = current;
2646
2647 plug->magic = PLUG_MAGIC;
2648 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002649 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002650 plug->should_sort = 0;
2651
2652 /*
2653 * If this is a nested plug, don't actually assign it. It will be
2654 * flushed on its own.
2655 */
2656 if (!tsk->plug) {
2657 /*
2658 * Store ordering should not be needed here, since a potential
2659 * preempt will imply a full memory barrier
2660 */
2661 tsk->plug = plug;
2662 }
2663}
2664EXPORT_SYMBOL(blk_start_plug);
2665
2666static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2667{
2668 struct request *rqa = container_of(a, struct request, queuelist);
2669 struct request *rqb = container_of(b, struct request, queuelist);
2670
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002671 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002672}
2673
Jens Axboe49cac012011-04-16 13:51:05 +02002674/*
2675 * If 'from_schedule' is true, then postpone the dispatch of requests
2676 * until a safe kblockd context. We due this to avoid accidental big
2677 * additional stack usage in driver dispatch, in places where the originally
2678 * plugger did not intend it.
2679 */
Jens Axboef6603782011-04-15 15:49:07 +02002680static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002681 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002682 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002683{
Jens Axboe49cac012011-04-16 13:51:05 +02002684 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002685
2686 /*
2687 * If we are punting this to kblockd, then we can safely drop
2688 * the queue_lock before waking kblockd (which needs to take
2689 * this lock).
2690 */
2691 if (from_schedule) {
2692 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002693 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002694 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002695 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002696 spin_unlock(q->queue_lock);
2697 }
2698
Jens Axboe94b5eb22011-04-12 10:12:19 +02002699}
2700
NeilBrown048c9372011-04-18 09:52:22 +02002701static void flush_plug_callbacks(struct blk_plug *plug)
2702{
2703 LIST_HEAD(callbacks);
2704
2705 if (list_empty(&plug->cb_list))
2706 return;
2707
2708 list_splice_init(&plug->cb_list, &callbacks);
2709
2710 while (!list_empty(&callbacks)) {
2711 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2712 struct blk_plug_cb,
2713 list);
2714 list_del(&cb->list);
2715 cb->callback(cb);
2716 }
2717}
2718
Jens Axboe49cac012011-04-16 13:51:05 +02002719void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002720{
2721 struct request_queue *q;
2722 unsigned long flags;
2723 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002724 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002725 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002726
2727 BUG_ON(plug->magic != PLUG_MAGIC);
2728
NeilBrown048c9372011-04-18 09:52:22 +02002729 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002730 if (list_empty(&plug->list))
2731 return;
2732
NeilBrown109b8122011-04-11 14:13:10 +02002733 list_splice_init(&plug->list, &list);
2734
2735 if (plug->should_sort) {
2736 list_sort(NULL, &list, plug_rq_cmp);
2737 plug->should_sort = 0;
2738 }
Jens Axboe73c10102011-03-08 13:19:51 +01002739
2740 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002741 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002742
2743 /*
2744 * Save and disable interrupts here, to avoid doing it for every
2745 * queue lock we have to take.
2746 */
Jens Axboe73c10102011-03-08 13:19:51 +01002747 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002748 while (!list_empty(&list)) {
2749 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002750 list_del_init(&rq->queuelist);
2751 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2752 BUG_ON(!rq->q);
2753 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002754 /*
2755 * This drops the queue lock
2756 */
2757 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002758 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002759 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002760 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002761 spin_lock(q->queue_lock);
2762 }
2763 rq->cmd_flags &= ~REQ_ON_PLUG;
2764
2765 /*
2766 * rq is already accounted, so use raw insert
2767 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002768 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2769 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2770 else
2771 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002772
2773 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002774 }
2775
Jens Axboe99e22592011-04-18 09:59:55 +02002776 /*
2777 * This drops the queue lock
2778 */
2779 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002780 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002781
Jens Axboe73c10102011-03-08 13:19:51 +01002782 local_irq_restore(flags);
2783}
Jens Axboe73c10102011-03-08 13:19:51 +01002784
2785void blk_finish_plug(struct blk_plug *plug)
2786{
Jens Axboef6603782011-04-15 15:49:07 +02002787 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002788
2789 if (plug == current->plug)
2790 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002791}
2792EXPORT_SYMBOL(blk_finish_plug);
2793
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794int __init blk_dev_init(void)
2795{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002796 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2797 sizeof(((struct request *)0)->cmd_flags));
2798
Tejun Heo89b90be2011-01-03 15:01:47 +01002799 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2800 kblockd_workqueue = alloc_workqueue("kblockd",
2801 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 if (!kblockd_workqueue)
2803 panic("Failed to create kblockd\n");
2804
2805 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002806 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807
Jens Axboe8324aa92008-01-29 14:51:59 +01002808 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02002809 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 return 0;
2812}