blob: ee3cb3a5e2782292bba4807c0d9cc887de48a31e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020031#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070032#include <linux/ratelimit.h>
Li Zefan55782132009-06-09 13:43:05 +080033
34#define CREATE_TRACE_POINTS
35#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jens Axboe8324aa92008-01-29 14:51:59 +010037#include "blk.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080038#include "blk-cgroup.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010039
Mike Snitzerd07335e2010-11-16 12:52:38 +010040EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020041EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Li Zefan55782132009-06-09 13:43:05 +080042EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010043
Tejun Heoa73f7302011-12-14 00:33:37 +010044DEFINE_IDA(blk_queue_ida);
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/*
47 * For the allocated request tables
48 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010049static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51/*
52 * For queue allocation
53 */
Jens Axboe6728cb02008-01-31 13:03:55 +010054struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 * Controlling structure to kblockd
58 */
Jens Axboeff856ba2006-01-09 16:02:34 +010059static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Jens Axboe26b82562008-01-29 13:54:41 +010061static void drive_stat_acct(struct request *rq, int new_io)
62{
Jens Axboe28f13702008-05-07 10:15:46 +020063 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010064 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090065 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010066
Jens Axboec2553b52009-04-24 08:10:11 +020067 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010068 return;
69
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090071
Jerome Marchand09e099d2011-01-05 16:57:38 +010072 if (!new_io) {
73 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090074 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010075 } else {
76 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010077 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010078 /*
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
81 *
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
85 */
86 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010087 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010088 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090089 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020090 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010091 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010092 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020093
Tejun Heo074a7ac2008-08-25 19:56:14 +090094 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010095}
96
Jens Axboe8324aa92008-01-29 14:51:59 +010097void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
99 int nr;
100
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
103 nr = q->nr_requests;
104 q->nr_congestion_on = nr;
105
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 if (nr < 1)
108 nr = 1;
109 q->nr_congestion_off = nr;
110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/**
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * @bdev: device
115 *
116 * Locates the passed device's request queue and returns the address of its
117 * backing_dev_info
118 *
119 * Will return NULL if the request queue cannot be located.
120 */
121struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122{
123 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200124 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126 if (q)
127 ret = &q->backing_dev_info;
128 return ret;
129}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130EXPORT_SYMBOL(blk_get_backing_dev_info);
131
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200132void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200134 memset(rq, 0, sizeof(*rq));
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700137 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200138 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100139 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900140 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200141 INIT_HLIST_NODE(&rq->hash);
142 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200143 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800144 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100145 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100146 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900147 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700148 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100149 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200151EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
NeilBrown5bb23a62007-09-27 12:46:13 +0200153static void req_bio_endio(struct request *rq, struct bio *bio,
154 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100155{
Tejun Heo143a87f2011-01-25 12:43:52 +0100156 if (error)
157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
158 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
159 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100160
Tejun Heo143a87f2011-01-25 12:43:52 +0100161 if (unlikely(nbytes > bio->bi_size)) {
162 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
163 __func__, nbytes, bio->bi_size);
164 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200165 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100166
167 if (unlikely(rq->cmd_flags & REQ_QUIET))
168 set_bit(BIO_QUIET, &bio->bi_flags);
169
170 bio->bi_size -= nbytes;
171 bio->bi_sector += (nbytes >> 9);
172
173 if (bio_integrity(bio))
174 bio_integrity_advance(bio, nbytes);
175
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
178 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100179}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181void blk_dump_rq_flags(struct request *rq, char *msg)
182{
183 int bit;
184
Jens Axboe6728cb02008-01-31 13:03:55 +0100185 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200186 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
187 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Tejun Heo83096eb2009-05-07 22:24:39 +0900189 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq),
191 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900193 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200195 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100196 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200197 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 printk("%02x ", rq->cmd[bit]);
199 printk("\n");
200 }
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_dump_rq_flags);
203
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500204static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200205{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500206 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200207
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500208 q = container_of(work, struct request_queue, delay_work.work);
209 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200210 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500211 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 *
219 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time.
223 */
224void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Jens Axboe4521cc42011-04-18 11:36:39 +0200226 queue_delayed_work(kblockd_workqueue, &q->delay_work,
227 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500229EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231/**
232 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200233 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 *
235 * Description:
236 * blk_start_queue() will clear the stop flag on the queue, and call
237 * the request_fn for the queue if it was in a stopped state when
238 * entered. Also see blk_stop_queue(). Queue lock must be held.
239 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200240void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200242 WARN_ON(!irqs_disabled());
243
Nick Piggin75ad23b2008-04-29 14:48:33 +0200244 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200245 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247EXPORT_SYMBOL(blk_start_queue);
248
249/**
250 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200251 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 *
253 * Description:
254 * The Linux block layer assumes that a block driver will consume all
255 * entries on the request queue when the request_fn strategy is called.
256 * Often this will not happen, because of hardware limitations (queue
257 * depth settings). If a device driver gets a 'queue full' response,
258 * or if it simply chooses not to queue more I/O at one point, it can
259 * call this function to prevent the request_fn from being called until
260 * the driver has signalled it's ready to go again. This happens by calling
261 * blk_start_queue() to restart queue operations. Queue lock must be held.
262 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200263void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264{
Jens Axboead3d9d72011-03-25 16:58:59 +0100265 __cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200266 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268EXPORT_SYMBOL(blk_stop_queue);
269
270/**
271 * blk_sync_queue - cancel any pending callbacks on a queue
272 * @q: the queue
273 *
274 * Description:
275 * The block layer may perform asynchronous callback activity
276 * on a queue, such as calling the unplug function after a timeout.
277 * A block device may call blk_sync_queue to ensure that any
278 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200279 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 * that its ->make_request_fn will not re-add plugging prior to calling
281 * this function.
282 *
Vivek Goyalda527772011-03-02 19:05:33 -0500283 * This function does not cancel any asynchronous activity arising
284 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800285 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500286 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 */
288void blk_sync_queue(struct request_queue *q)
289{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100290 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500291 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293EXPORT_SYMBOL(blk_sync_queue);
294
295/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200296 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200298 *
299 * Description:
300 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200301 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200303void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Tejun Heoa538cd02009-04-23 11:05:17 +0900305 if (unlikely(blk_queue_stopped(q)))
306 return;
307
Jens Axboec21e6be2011-04-19 13:32:46 +0200308 q->request_fn(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200309}
310EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200311
Nick Piggin75ad23b2008-04-29 14:48:33 +0200312/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200313 * blk_run_queue_async - run a single device queue in workqueue context
314 * @q: The queue to run
315 *
316 * Description:
317 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
318 * of us.
319 */
320void blk_run_queue_async(struct request_queue *q)
321{
Shaohua Li3ec717b2011-05-18 11:22:43 +0200322 if (likely(!blk_queue_stopped(q))) {
323 __cancel_delayed_work(&q->delay_work);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Shaohua Li3ec717b2011-05-18 11:22:43 +0200325 }
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200326}
Jens Axboec21e6be2011-04-19 13:32:46 +0200327EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200328
329/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200330 * blk_run_queue - run a single device queue
331 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200332 *
333 * Description:
334 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900335 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200336 */
337void blk_run_queue(struct request_queue *q)
338{
339 unsigned long flags;
340
341 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200342 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 spin_unlock_irqrestore(q->queue_lock, flags);
344}
345EXPORT_SYMBOL(blk_run_queue);
346
Jens Axboe165125e2007-07-24 09:28:11 +0200347void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500348{
349 kobject_put(&q->kobj);
350}
Jens Axboed86e0e82011-05-27 07:44:43 +0200351EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500352
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353/**
354 * blk_drain_queue - drain requests from request_queue
355 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200356 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200357 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200358 * Drain requests from @q. If @drain_all is set, all requests are drained.
359 * If not, only ELVPRIV requests are drained. The caller is responsible
360 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200361 */
Tejun Heoc9a929d2011-10-19 14:42:16 +0200362void blk_drain_queue(struct request_queue *q, bool drain_all)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200363{
Asias He458f27a2012-06-15 08:45:25 +0200364 int i;
365
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200366 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100367 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200368
369 spin_lock_irq(q->queue_lock);
370
Tejun Heob855b042012-03-06 21:24:55 +0100371 /*
372 * The caller might be trying to drain @q before its
373 * elevator is initialized.
374 */
375 if (q->elevator)
376 elv_drain_elevator(q);
377
Tejun Heo5efd6112012-03-05 13:15:12 -0800378 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200379
Tejun Heo4eabc942011-12-15 20:03:04 +0100380 /*
381 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100382 * driver init after queue creation or is not yet fully
383 * active yet. Some drivers (e.g. fd and loop) get unhappy
384 * in such cases. Kick queue iff dispatch queue has
385 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100386 */
Tejun Heob855b042012-03-06 21:24:55 +0100387 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100388 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200389
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700390 drain |= q->nr_rqs_elvpriv;
Tejun Heo481a7d62011-12-14 00:33:37 +0100391
392 /*
393 * Unfortunately, requests are queued at and tracked from
394 * multiple places and there's no single counter which can
395 * be drained. Check all the queues and counters.
396 */
397 if (drain_all) {
398 drain |= !list_empty(&q->queue_head);
399 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700400 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100401 drain |= q->in_flight[i];
402 drain |= !list_empty(&q->flush_queue[i]);
403 }
404 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200405
406 spin_unlock_irq(q->queue_lock);
407
Tejun Heo481a7d62011-12-14 00:33:37 +0100408 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200409 break;
410 msleep(10);
411 }
Asias He458f27a2012-06-15 08:45:25 +0200412
413 /*
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
417 */
418 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700419 struct request_list *rl;
420
Asias He458f27a2012-06-15 08:45:25 +0200421 spin_lock_irq(q->queue_lock);
Tejun Heoa0516612012-06-26 15:05:44 -0700422
423 blk_queue_for_each_rl(rl, q)
424 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
425 wake_up_all(&rl->wait[i]);
426
Asias He458f27a2012-06-15 08:45:25 +0200427 spin_unlock_irq(q->queue_lock);
428 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200429}
430
Tejun Heoc9a929d2011-10-19 14:42:16 +0200431/**
Tejun Heod7325802012-03-05 13:14:58 -0800432 * blk_queue_bypass_start - enter queue bypass mode
433 * @q: queue of interest
434 *
435 * In bypass mode, only the dispatch FIFO queue of @q is used. This
436 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800437 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700438 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
439 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800440 */
441void blk_queue_bypass_start(struct request_queue *q)
442{
Tejun Heob82d4b12012-04-13 13:11:31 -0700443 bool drain;
444
Tejun Heod7325802012-03-05 13:14:58 -0800445 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700446 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800447 queue_flag_set(QUEUE_FLAG_BYPASS, q);
448 spin_unlock_irq(q->queue_lock);
449
Tejun Heob82d4b12012-04-13 13:11:31 -0700450 if (drain) {
451 blk_drain_queue(q, false);
452 /* ensure blk_queue_bypass() is %true inside RCU read lock */
453 synchronize_rcu();
454 }
Tejun Heod7325802012-03-05 13:14:58 -0800455}
456EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
457
458/**
459 * blk_queue_bypass_end - leave queue bypass mode
460 * @q: queue of interest
461 *
462 * Leave bypass mode and restore the normal queueing behavior.
463 */
464void blk_queue_bypass_end(struct request_queue *q)
465{
466 spin_lock_irq(q->queue_lock);
467 if (!--q->bypass_depth)
468 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
469 WARN_ON_ONCE(q->bypass_depth < 0);
470 spin_unlock_irq(q->queue_lock);
471}
472EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
473
474/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200475 * blk_cleanup_queue - shutdown a request queue
476 * @q: request queue to shutdown
477 *
478 * Mark @q DEAD, drain all pending requests, destroy and put it. All
479 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500480 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100481void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500482{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200483 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700484
Tejun Heoc9a929d2011-10-19 14:42:16 +0200485 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500486 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200487 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200488 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800489
Tejun Heo80fd9972012-04-13 14:50:53 -0700490 /*
491 * Dead queue is permanently in bypass mode till released. Note
492 * that, unlike blk_queue_bypass_start(), we aren't performing
493 * synchronize_rcu() after entering bypass mode to avoid the delay
494 * as some drivers create and destroy a lot of queues while
495 * probing. This is still safe because blk_release_queue() will be
496 * called only after the queue refcnt drops to zero and nothing,
497 * RCU or not, would be traversing the queue by then.
498 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800499 q->bypass_depth++;
500 queue_flag_set(QUEUE_FLAG_BYPASS, q);
501
Tejun Heoc9a929d2011-10-19 14:42:16 +0200502 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
503 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
504 queue_flag_set(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200505 spin_unlock_irq(lock);
506 mutex_unlock(&q->sysfs_lock);
507
Tejun Heob855b042012-03-06 21:24:55 +0100508 /* drain all requests queued before DEAD marking */
509 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200510
511 /* @q won't process any more request, flush async actions */
512 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
513 blk_sync_queue(q);
514
Asias He5e5cfac2012-05-24 23:28:52 +0800515 spin_lock_irq(lock);
516 if (q->queue_lock != &q->__queue_lock)
517 q->queue_lock = &q->__queue_lock;
518 spin_unlock_irq(lock);
519
Tejun Heoc9a929d2011-10-19 14:42:16 +0200520 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500521 blk_put_queue(q);
522}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523EXPORT_SYMBOL(blk_cleanup_queue);
524
Tejun Heo5b788ce2012-06-04 20:40:59 -0700525int blk_init_rl(struct request_list *rl, struct request_queue *q,
526 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400528 if (unlikely(rl->rq_pool))
529 return 0;
530
Tejun Heo5b788ce2012-06-04 20:40:59 -0700531 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200532 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
533 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200534 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
535 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Christoph Lameter19460892005-06-23 00:08:19 -0700537 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700538 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700539 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (!rl->rq_pool)
541 return -ENOMEM;
542
543 return 0;
544}
545
Tejun Heo5b788ce2012-06-04 20:40:59 -0700546void blk_exit_rl(struct request_list *rl)
547{
548 if (rl->rq_pool)
549 mempool_destroy(rl->rq_pool);
550}
551
Jens Axboe165125e2007-07-24 09:28:11 +0200552struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
Christoph Lameter19460892005-06-23 00:08:19 -0700554 return blk_alloc_queue_node(gfp_mask, -1);
555}
556EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Jens Axboe165125e2007-07-24 09:28:11 +0200558struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700559{
Jens Axboe165125e2007-07-24 09:28:11 +0200560 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700561 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700562
Jens Axboe8324aa92008-01-29 14:51:59 +0100563 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700564 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (!q)
566 return NULL;
567
Dan Carpenter00380a42012-03-23 09:58:54 +0100568 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100569 if (q->id < 0)
570 goto fail_q;
571
Jens Axboe0989a022009-06-12 14:42:56 +0200572 q->backing_dev_info.ra_pages =
573 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
574 q->backing_dev_info.state = 0;
575 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200576 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100577 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200578
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700579 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100580 if (err)
581 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700582
Matthew Garrett31373d02010-04-06 14:25:14 +0200583 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
584 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700585 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100586 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700587 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100588 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800589#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800590 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800591#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100592 INIT_LIST_HEAD(&q->flush_queue[0]);
593 INIT_LIST_HEAD(&q->flush_queue[1]);
594 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500595 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500596
Jens Axboe8324aa92008-01-29 14:51:59 +0100597 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Al Viro483f4af2006-03-18 18:34:37 -0500599 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700600 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500601
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500602 /*
603 * By default initialize queue_lock to internal lock and driver can
604 * override it later if need be.
605 */
606 q->queue_lock = &q->__queue_lock;
607
Tejun Heob82d4b12012-04-13 13:11:31 -0700608 /*
609 * A queue starts its life with bypass turned on to avoid
610 * unnecessary bypass on/off overhead and nasty surprises during
611 * init. The initial bypass will be finished at the end of
612 * blk_init_allocated_queue().
613 */
614 q->bypass_depth = 1;
615 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
616
Tejun Heo5efd6112012-03-05 13:15:12 -0800617 if (blkcg_init_queue(q))
Tejun Heof51b8022012-03-05 13:15:05 -0800618 goto fail_id;
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100621
622fail_id:
623 ida_simple_remove(&blk_queue_ida, q->id);
624fail_q:
625 kmem_cache_free(blk_requestq_cachep, q);
626 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}
Christoph Lameter19460892005-06-23 00:08:19 -0700628EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
630/**
631 * blk_init_queue - prepare a request queue for use with a block device
632 * @rfn: The function to be called to process requests that have been
633 * placed on the queue.
634 * @lock: Request queue spin lock
635 *
636 * Description:
637 * If a block device wishes to use the standard request handling procedures,
638 * which sorts requests and coalesces adjacent requests, then it must
639 * call blk_init_queue(). The function @rfn will be called when there
640 * are requests on the queue that need to be processed. If the device
641 * supports plugging, then @rfn may not be called immediately when requests
642 * are available on the queue, but may be called at some time later instead.
643 * Plugged queues are generally unplugged when a buffer belonging to one
644 * of the requests on the queue is needed, or due to memory pressure.
645 *
646 * @rfn is not required, or even expected, to remove all requests off the
647 * queue, but only as many as it can handle at a time. If it does leave
648 * requests on the queue, it is responsible for arranging that the requests
649 * get dealt with eventually.
650 *
651 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200652 * request queue; this lock will be taken also from interrupt context, so irq
653 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200655 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * it didn't succeed.
657 *
658 * Note:
659 * blk_init_queue() must be paired with a blk_cleanup_queue() call
660 * when the block device is deactivated (such as at module unload).
661 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700662
Jens Axboe165125e2007-07-24 09:28:11 +0200663struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Christoph Lameter19460892005-06-23 00:08:19 -0700665 return blk_init_queue_node(rfn, lock, -1);
666}
667EXPORT_SYMBOL(blk_init_queue);
668
Jens Axboe165125e2007-07-24 09:28:11 +0200669struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700670blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
671{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600672 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600674 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
675 if (!uninit_q)
676 return NULL;
677
Mike Snitzer51514122011-11-23 10:59:13 +0100678 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600679 if (!q)
680 blk_cleanup_queue(uninit_q);
681
682 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200683}
684EXPORT_SYMBOL(blk_init_queue_node);
685
686struct request_queue *
687blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
688 spinlock_t *lock)
689{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!q)
691 return NULL;
692
Tejun Heoa0516612012-06-26 15:05:44 -0700693 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500694 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900698 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100699 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500700
701 /* Override internal queue lock with supplied lock pointer */
702 if (lock)
703 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Jens Axboef3b144a2009-03-06 08:48:33 +0100705 /*
706 * This also sets hw/phys segments, boundary and size
707 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200708 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Alan Stern44ec9542007-02-20 11:01:57 -0500710 q->sg_reserved_size = INT_MAX;
711
Tejun Heob82d4b12012-04-13 13:11:31 -0700712 /* init elevator */
713 if (elevator_init(q, NULL))
714 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Tejun Heob82d4b12012-04-13 13:11:31 -0700716 blk_queue_congestion_threshold(q);
717
718 /* all done, end the initial bypass */
719 blk_queue_bypass_end(q);
720 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
Mike Snitzer51514122011-11-23 10:59:13 +0100722EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Tejun Heo09ac46c2011-12-14 00:33:38 +0100724bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Tejun Heo34f60552011-12-14 00:33:37 +0100726 if (likely(!blk_queue_dead(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100727 __blk_get_queue(q);
728 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
730
Tejun Heo09ac46c2011-12-14 00:33:38 +0100731 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
Jens Axboed86e0e82011-05-27 07:44:43 +0200733EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Tejun Heo5b788ce2012-06-04 20:40:59 -0700735static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100737 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700738 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100739 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100740 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100741 }
742
Tejun Heo5b788ce2012-06-04 20:40:59 -0700743 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/*
747 * ioc_batching returns true if the ioc is a valid batching request and
748 * should be given priority access to a request.
749 */
Jens Axboe165125e2007-07-24 09:28:11 +0200750static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
752 if (!ioc)
753 return 0;
754
755 /*
756 * Make sure the process is able to allocate at least 1 request
757 * even if the batch times out, otherwise we could theoretically
758 * lose wakeups.
759 */
760 return ioc->nr_batch_requests == q->nr_batching ||
761 (ioc->nr_batch_requests > 0
762 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
763}
764
765/*
766 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
767 * will cause the process to be a "batcher" on all queues in the system. This
768 * is the behaviour we want though - once it gets a wakeup it should be given
769 * a nice run.
770 */
Jens Axboe165125e2007-07-24 09:28:11 +0200771static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
773 if (!ioc || ioc_batching(q, ioc))
774 return;
775
776 ioc->nr_batch_requests = q->nr_batching;
777 ioc->last_waited = jiffies;
778}
779
Tejun Heo5b788ce2012-06-04 20:40:59 -0700780static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700782 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Tejun Heoa0516612012-06-26 15:05:44 -0700784 /*
785 * bdi isn't aware of blkcg yet. As all async IOs end up root
786 * blkcg anyway, just use root blkcg state.
787 */
788 if (rl == &q->root_rl &&
789 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200790 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Jens Axboe1faa16d2009-04-06 14:48:01 +0200792 if (rl->count[sync] + 1 <= q->nr_requests) {
793 if (waitqueue_active(&rl->wait[sync]))
794 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Tejun Heo5b788ce2012-06-04 20:40:59 -0700796 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 }
798}
799
800/*
801 * A request has just been released. Account for it, update the full and
802 * congestion status, wake up any waiters. Called under q->queue_lock.
803 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700804static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700806 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200807 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700809 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200810 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200811 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700812 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Tejun Heo5b788ce2012-06-04 20:40:59 -0700814 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Jens Axboe1faa16d2009-04-06 14:48:01 +0200816 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700817 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100821 * Determine if elevator data should be initialized when allocating the
822 * request associated with @bio.
823 */
824static bool blk_rq_should_init_elevator(struct bio *bio)
825{
826 if (!bio)
827 return true;
828
829 /*
830 * Flush requests do not use the elevator so skip initialization.
831 * This allows a request to share the flush and elevator data.
832 */
833 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
834 return false;
835
836 return true;
837}
838
Tejun Heoda8303c2011-10-19 14:33:05 +0200839/**
Tejun Heo852c7882012-03-05 13:15:27 -0800840 * rq_ioc - determine io_context for request allocation
841 * @bio: request being allocated is for this bio (can be %NULL)
842 *
843 * Determine io_context to use for request allocation for @bio. May return
844 * %NULL if %current->io_context doesn't exist.
845 */
846static struct io_context *rq_ioc(struct bio *bio)
847{
848#ifdef CONFIG_BLK_CGROUP
849 if (bio && bio->bi_ioc)
850 return bio->bi_ioc;
851#endif
852 return current->io_context;
853}
854
855/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700856 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700857 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200858 * @rw_flags: RW and SYNC flags
859 * @bio: bio to allocate request for (can be %NULL)
860 * @gfp_mask: allocation mask
861 *
862 * Get a free request from @q. This function may fail under memory
863 * pressure or if @q is dead.
864 *
865 * Must be callled with @q->queue_lock held and,
866 * Returns %NULL on failure, with @q->queue_lock held.
867 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700869static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700870 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700872 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800873 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700874 struct elevator_type *et = q->elevator->type;
875 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100876 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200877 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200878 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Tejun Heo34f60552011-12-14 00:33:37 +0100880 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200881 return NULL;
882
Jens Axboe7749a8d2006-12-13 13:02:26 +0100883 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100884 if (may_queue == ELV_MQUEUE_NO)
885 goto rq_starved;
886
Jens Axboe1faa16d2009-04-06 14:48:01 +0200887 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
888 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100889 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100890 * The queue will fill after this allocation, so set
891 * it as full, and mark this process as "batching".
892 * This process will be allowed to complete a batch of
893 * requests, others will be blocked.
894 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700895 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100896 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700897 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100898 } else {
899 if (may_queue != ELV_MQUEUE_MUST
900 && !ioc_batching(q, ioc)) {
901 /*
902 * The queue is full and the allocating
903 * process is not a "batcher", and not
904 * exempted by the IO scheduler
905 */
Tejun Heob6792812012-03-05 13:15:23 -0800906 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100907 }
908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
Tejun Heoa0516612012-06-26 15:05:44 -0700910 /*
911 * bdi isn't aware of blkcg yet. As all async IOs end up
912 * root blkcg anyway, just use root blkcg state.
913 */
914 if (rl == &q->root_rl)
915 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
917
Jens Axboe082cf692005-06-28 16:35:11 +0200918 /*
919 * Only allow batching queuers to allocate up to 50% over the defined
920 * limit of requests, otherwise we could have thousands of requests
921 * allocated with any setting of ->nr_requests
922 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200923 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800924 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100925
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700926 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200927 rl->count[is_sync]++;
928 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200929
Tejun Heof1f8cc92011-12-14 00:33:42 +0100930 /*
931 * Decide whether the new request will be managed by elevator. If
932 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
933 * prevent the current elevator from being destroyed until the new
934 * request is freed. This guarantees icq's won't be destroyed and
935 * makes creating new ones safe.
936 *
937 * Also, lookup icq while holding queue_lock. If it doesn't exist,
938 * it will be created after releasing queue_lock.
939 */
Tejun Heod7325802012-03-05 13:14:58 -0800940 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200941 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700942 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100943 if (et->icq_cache && ioc)
944 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100945 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200946
Jens Axboef253b862010-10-24 22:06:02 +0200947 if (blk_queue_io_stat(q))
948 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 spin_unlock_irq(q->queue_lock);
950
Tejun Heo29e2b092012-04-19 16:29:21 -0700951 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700952 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700953 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800954 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Tejun Heo29e2b092012-04-19 16:29:21 -0700956 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700957 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700958 rq->cmd_flags = rw_flags | REQ_ALLOCED;
959
Tejun Heoaaf7c682012-04-19 16:29:22 -0700960 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700961 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700962 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700963 if (ioc)
964 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700965 if (!icq)
966 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700967 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700968
969 rq->elv.icq = icq;
970 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
971 goto fail_elvpriv;
972
973 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -0700974 if (icq)
975 get_io_context(icq->ioc);
976 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700977out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100978 /*
979 * ioc may be NULL here, and ioc_batching will be false. That's
980 * OK, if the queue is under the request limit then requests need
981 * not count toward the nr_batch_requests limit. There will always
982 * be some limit enforced by BLK_BATCH_TIME.
983 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 if (ioc_batching(q, ioc))
985 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100986
Jens Axboe1faa16d2009-04-06 14:48:01 +0200987 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return rq;
Tejun Heob6792812012-03-05 13:15:23 -0800989
Tejun Heoaaf7c682012-04-19 16:29:22 -0700990fail_elvpriv:
991 /*
992 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
993 * and may fail indefinitely under memory pressure and thus
994 * shouldn't stall IO. Treat this request as !elvpriv. This will
995 * disturb iosched and blkcg but weird is bettern than dead.
996 */
997 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
998 dev_name(q->backing_dev_info.dev));
999
1000 rq->cmd_flags &= ~REQ_ELVPRIV;
1001 rq->elv.icq = NULL;
1002
1003 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001004 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001005 spin_unlock_irq(q->queue_lock);
1006 goto out;
1007
Tejun Heob6792812012-03-05 13:15:23 -08001008fail_alloc:
1009 /*
1010 * Allocation failed presumably due to memory. Undo anything we
1011 * might have messed up.
1012 *
1013 * Allocating task should really be put onto the front of the wait
1014 * queue, but this is pretty rare.
1015 */
1016 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001017 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001018
1019 /*
1020 * in the very unlikely event that allocation failed and no
1021 * requests for this direction was pending, mark us starved so that
1022 * freeing of a request in the other direction will notice
1023 * us. another possible fix would be to split the rq mempool into
1024 * READ and WRITE
1025 */
1026rq_starved:
1027 if (unlikely(rl->count[is_sync] == 0))
1028 rl->starved[is_sync] = 1;
1029 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
Tejun Heoda8303c2011-10-19 14:33:05 +02001032/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001033 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001034 * @q: request_queue to allocate request from
1035 * @rw_flags: RW and SYNC flags
1036 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001037 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001038 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001039 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1040 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001041 *
1042 * Must be callled with @q->queue_lock held and,
1043 * Returns %NULL on failure, with @q->queue_lock held.
1044 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001046static struct request *get_request(struct request_queue *q, int rw_flags,
1047 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001049 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001050 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001051 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001053
1054 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001055retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001056 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001057 if (rq)
1058 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Tejun Heoa0516612012-06-26 15:05:44 -07001060 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dead(q))) {
1061 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001062 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Tejun Heoa06e05e2012-06-04 20:40:55 -07001065 /* wait on @rl and retry */
1066 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1067 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001068
Tejun Heoa06e05e2012-06-04 20:40:55 -07001069 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Tejun Heoa06e05e2012-06-04 20:40:55 -07001071 spin_unlock_irq(q->queue_lock);
1072 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Tejun Heoa06e05e2012-06-04 20:40:55 -07001074 /*
1075 * After sleeping, we become a "batching" process and will be able
1076 * to allocate at least one request, and up to a big batch of them
1077 * for a small period time. See ioc_batching, ioc_set_batching
1078 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001079 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Tejun Heoa06e05e2012-06-04 20:40:55 -07001081 spin_lock_irq(q->queue_lock);
1082 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001083
Tejun Heoa06e05e2012-06-04 20:40:55 -07001084 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085}
1086
Jens Axboe165125e2007-07-24 09:28:11 +02001087struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088{
1089 struct request *rq;
1090
1091 BUG_ON(rw != READ && rw != WRITE);
1092
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001093 /* create ioc upfront */
1094 create_io_context(gfp_mask, q->node);
1095
Nick Piggind6344532005-06-28 20:45:14 -07001096 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001097 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001098 if (!rq)
1099 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001100 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 return rq;
1103}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104EXPORT_SYMBOL(blk_get_request);
1105
1106/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001107 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001108 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001109 * @bio: The bio describing the memory mappings that will be submitted for IO.
1110 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001111 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001112 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001113 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1114 * type commands. Where the struct request needs to be farther initialized by
1115 * the caller. It is passed a &struct bio, which describes the memory info of
1116 * the I/O transfer.
1117 *
1118 * The caller of blk_make_request must make sure that bi_io_vec
1119 * are set to describe the memory buffers. That bio_data_dir() will return
1120 * the needed direction of the request. (And all bio's in the passed bio-chain
1121 * are properly set accordingly)
1122 *
1123 * If called under none-sleepable conditions, mapped bio buffers must not
1124 * need bouncing, by calling the appropriate masked or flagged allocator,
1125 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1126 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001127 *
1128 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1129 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1130 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1131 * completion of a bio that hasn't been submitted yet, thus resulting in a
1132 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1133 * of bio_alloc(), as that avoids the mempool deadlock.
1134 * If possible a big IO should be split into smaller parts when allocation
1135 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001136 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001137struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1138 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001139{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001140 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1141
1142 if (unlikely(!rq))
1143 return ERR_PTR(-ENOMEM);
1144
1145 for_each_bio(bio) {
1146 struct bio *bounce_bio = bio;
1147 int ret;
1148
1149 blk_queue_bounce(q, &bounce_bio);
1150 ret = blk_rq_append_bio(q, rq, bounce_bio);
1151 if (unlikely(ret)) {
1152 blk_put_request(rq);
1153 return ERR_PTR(ret);
1154 }
1155 }
1156
1157 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001158}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001159EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001160
1161/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 * blk_requeue_request - put a request back on queue
1163 * @q: request queue where request should be inserted
1164 * @rq: request to be inserted
1165 *
1166 * Description:
1167 * Drivers often keep queueing requests until the hardware cannot accept
1168 * more, when that condition happens we need to put the request back
1169 * on the queue. Must be called with queue lock held.
1170 */
Jens Axboe165125e2007-07-24 09:28:11 +02001171void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001173 blk_delete_timer(rq);
1174 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001175 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (blk_rq_tagged(rq))
1178 blk_queue_end_tag(q, rq);
1179
James Bottomleyba396a62009-05-27 14:17:08 +02001180 BUG_ON(blk_queued_rq(rq));
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 elv_requeue_request(q, rq);
1183}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184EXPORT_SYMBOL(blk_requeue_request);
1185
Jens Axboe73c10102011-03-08 13:19:51 +01001186static void add_acct_request(struct request_queue *q, struct request *rq,
1187 int where)
1188{
1189 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001190 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001191}
1192
Tejun Heo074a7ac2008-08-25 19:56:14 +09001193static void part_round_stats_single(int cpu, struct hd_struct *part,
1194 unsigned long now)
1195{
1196 if (now == part->stamp)
1197 return;
1198
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001199 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001200 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001201 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001202 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1203 }
1204 part->stamp = now;
1205}
1206
1207/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001208 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1209 * @cpu: cpu number for stats access
1210 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 *
1212 * The average IO queue length and utilisation statistics are maintained
1213 * by observing the current state of the queue length and the amount of
1214 * time it has been in this state for.
1215 *
1216 * Normally, that accounting is done on IO completion, but that can result
1217 * in more than a second's worth of IO being accounted for within any one
1218 * second, leading to >100% utilisation. To deal with that, we call this
1219 * function to do a round-off before returning the results when reading
1220 * /proc/diskstats. This accounts immediately for all queue usage up to
1221 * the current jiffies and restarts the counters again.
1222 */
Tejun Heoc9959052008-08-25 19:47:21 +09001223void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001224{
1225 unsigned long now = jiffies;
1226
Tejun Heo074a7ac2008-08-25 19:56:14 +09001227 if (part->partno)
1228 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1229 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001230}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001231EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001232
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233/*
1234 * queue lock must be held
1235 */
Jens Axboe165125e2007-07-24 09:28:11 +02001236void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 if (unlikely(!q))
1239 return;
1240 if (unlikely(--req->ref_count))
1241 return;
1242
Tejun Heo8922e162005-10-20 16:23:44 +02001243 elv_completed_request(q, req);
1244
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001245 /* this is a bio leak */
1246 WARN_ON(req->bio != NULL);
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 /*
1249 * Request may not have originated from ll_rw_blk. if not,
1250 * it didn't come out of our reserved rq pools
1251 */
Jens Axboe49171e52006-08-10 08:59:11 +02001252 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001253 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001254 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001257 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Tejun Heoa0516612012-06-26 15:05:44 -07001259 blk_free_request(rl, req);
1260 freed_request(rl, flags);
1261 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
1263}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001264EXPORT_SYMBOL_GPL(__blk_put_request);
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266void blk_put_request(struct request *req)
1267{
Tejun Heo8922e162005-10-20 16:23:44 +02001268 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001269 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001271 spin_lock_irqsave(q->queue_lock, flags);
1272 __blk_put_request(q, req);
1273 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275EXPORT_SYMBOL(blk_put_request);
1276
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001277/**
1278 * blk_add_request_payload - add a payload to a request
1279 * @rq: request to update
1280 * @page: page backing the payload
1281 * @len: length of the payload.
1282 *
1283 * This allows to later add a payload to an already submitted request by
1284 * a block driver. The driver needs to take care of freeing the payload
1285 * itself.
1286 *
1287 * Note that this is a quite horrible hack and nothing but handling of
1288 * discard requests should ever use it.
1289 */
1290void blk_add_request_payload(struct request *rq, struct page *page,
1291 unsigned int len)
1292{
1293 struct bio *bio = rq->bio;
1294
1295 bio->bi_io_vec->bv_page = page;
1296 bio->bi_io_vec->bv_offset = 0;
1297 bio->bi_io_vec->bv_len = len;
1298
1299 bio->bi_size = len;
1300 bio->bi_vcnt = 1;
1301 bio->bi_phys_segments = 1;
1302
1303 rq->__data_len = rq->resid_len = len;
1304 rq->nr_phys_segments = 1;
1305 rq->buffer = bio_data(bio);
1306}
1307EXPORT_SYMBOL_GPL(blk_add_request_payload);
1308
Jens Axboe73c10102011-03-08 13:19:51 +01001309static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1310 struct bio *bio)
1311{
1312 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1313
Jens Axboe73c10102011-03-08 13:19:51 +01001314 if (!ll_back_merge_fn(q, req, bio))
1315 return false;
1316
1317 trace_block_bio_backmerge(q, bio);
1318
1319 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1320 blk_rq_set_mixed_merge(req);
1321
1322 req->biotail->bi_next = bio;
1323 req->biotail = bio;
1324 req->__data_len += bio->bi_size;
1325 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1326
1327 drive_stat_acct(req, 0);
1328 return true;
1329}
1330
1331static bool bio_attempt_front_merge(struct request_queue *q,
1332 struct request *req, struct bio *bio)
1333{
1334 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001335
Jens Axboe73c10102011-03-08 13:19:51 +01001336 if (!ll_front_merge_fn(q, req, bio))
1337 return false;
1338
1339 trace_block_bio_frontmerge(q, bio);
1340
1341 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1342 blk_rq_set_mixed_merge(req);
1343
Jens Axboe73c10102011-03-08 13:19:51 +01001344 bio->bi_next = req->bio;
1345 req->bio = bio;
1346
1347 /*
1348 * may not be valid. if the low level driver said
1349 * it didn't need a bounce buffer then it better
1350 * not touch req->buffer either...
1351 */
1352 req->buffer = bio_data(bio);
1353 req->__sector = bio->bi_sector;
1354 req->__data_len += bio->bi_size;
1355 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1356
1357 drive_stat_acct(req, 0);
1358 return true;
1359}
1360
Tejun Heobd87b582011-10-19 14:33:08 +02001361/**
1362 * attempt_plug_merge - try to merge with %current's plugged list
1363 * @q: request_queue new bio is being queued at
1364 * @bio: new bio being queued
1365 * @request_count: out parameter for number of traversed plugged requests
1366 *
1367 * Determine whether @bio being queued on @q can be merged with a request
1368 * on %current's plugged list. Returns %true if merge was successful,
1369 * otherwise %false.
1370 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001371 * Plugging coalesces IOs from the same issuer for the same purpose without
1372 * going through @q->queue_lock. As such it's more of an issuing mechanism
1373 * than scheduling, and the request, while may have elvpriv data, is not
1374 * added on the elevator at this point. In addition, we don't have
1375 * reliable access to the elevator outside queue lock. Only check basic
1376 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001377 */
Tejun Heobd87b582011-10-19 14:33:08 +02001378static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1379 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001380{
1381 struct blk_plug *plug;
1382 struct request *rq;
1383 bool ret = false;
1384
Tejun Heobd87b582011-10-19 14:33:08 +02001385 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001386 if (!plug)
1387 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001388 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001389
1390 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1391 int el_ret;
1392
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001393 if (rq->q == q)
1394 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001395
Tejun Heo07c2bd32012-02-08 09:19:42 +01001396 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001397 continue;
1398
Tejun Heo050c8ea2012-02-08 09:19:38 +01001399 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001400 if (el_ret == ELEVATOR_BACK_MERGE) {
1401 ret = bio_attempt_back_merge(q, rq, bio);
1402 if (ret)
1403 break;
1404 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1405 ret = bio_attempt_front_merge(q, rq, bio);
1406 if (ret)
1407 break;
1408 }
1409 }
1410out:
1411 return ret;
1412}
1413
Jens Axboe86db1e22008-01-29 14:53:40 +01001414void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001415{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001416 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001417
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001418 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1419 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001420 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001421
Tejun Heo52d9e672006-01-06 09:49:58 +01001422 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001423 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001424 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001425 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001426}
1427
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001428void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001430 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001431 struct blk_plug *plug;
1432 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1433 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001434 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 /*
1437 * low level driver can indicate that it wants pages above a
1438 * certain limit bounced to low memory (ie for highmem, or even
1439 * ISA dma in theory)
1440 */
1441 blk_queue_bounce(q, &bio);
1442
Tejun Heo4fed9472010-09-03 11:56:17 +02001443 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001444 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001445 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001446 goto get_rq;
1447 }
1448
Jens Axboe73c10102011-03-08 13:19:51 +01001449 /*
1450 * Check if we can merge with the plugged list before grabbing
1451 * any locks.
1452 */
Tejun Heobd87b582011-10-19 14:33:08 +02001453 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001454 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001455
1456 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
1458 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001459 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001460 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001461 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001462 if (!attempt_back_merge(q, req))
1463 elv_merged_request(q, req, el_ret);
1464 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001465 }
Jens Axboe73c10102011-03-08 13:19:51 +01001466 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001467 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001468 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001469 if (!attempt_front_merge(q, req))
1470 elv_merged_request(q, req, el_ret);
1471 goto out_unlock;
1472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 }
1474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001476 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001477 * This sync check and mask will be re-done in init_request_from_bio(),
1478 * but we need to set it earlier to expose the sync flag to the
1479 * rq allocator and io schedulers.
1480 */
1481 rw_flags = bio_data_dir(bio);
1482 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001483 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001484
1485 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001486 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001487 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001488 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001489 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001490 if (unlikely(!req)) {
1491 bio_endio(bio, -ENODEV); /* @q is dead */
1492 goto out_unlock;
1493 }
Nick Piggind6344532005-06-28 20:45:14 -07001494
Nick Piggin450991b2005-06-28 20:45:13 -07001495 /*
1496 * After dropping the lock and possibly sleeping here, our request
1497 * may now be mergeable after it had proven unmergeable (above).
1498 * We don't worry about that case for efficiency. It won't happen
1499 * often, and the elevators are able to handle it.
1500 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001501 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Tao Ma9562ad92011-10-24 16:11:30 +02001503 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001504 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001505
Jens Axboe73c10102011-03-08 13:19:51 +01001506 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001507 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001508 /*
1509 * If this is the first request added after a plug, fire
1510 * of a plug trace. If others have been added before, check
1511 * if we have multiple devices in this plug. If so, make a
1512 * note to sort the list before dispatch.
1513 */
1514 if (list_empty(&plug->list))
1515 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001516 else {
1517 if (!plug->should_sort) {
1518 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001519
Shaohua Li3540d5e2011-11-16 09:21:50 +01001520 __rq = list_entry_rq(plug->list.prev);
1521 if (__rq->q != q)
1522 plug->should_sort = 1;
1523 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001524 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001525 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001526 trace_block_plug(q);
1527 }
Jens Axboe73c10102011-03-08 13:19:51 +01001528 }
Shaohua Lia6327162011-08-24 16:04:32 +02001529 list_add_tail(&req->queuelist, &plug->list);
1530 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001531 } else {
1532 spin_lock_irq(q->queue_lock);
1533 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001534 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001535out_unlock:
1536 spin_unlock_irq(q->queue_lock);
1537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
Jens Axboec20e8de2011-09-12 12:03:37 +02001539EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541/*
1542 * If bio->bi_dev is a partition, remap the location
1543 */
1544static inline void blk_partition_remap(struct bio *bio)
1545{
1546 struct block_device *bdev = bio->bi_bdev;
1547
Jens Axboebf2de6f2007-09-27 13:01:25 +02001548 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 bio->bi_sector += p->start_sect;
1552 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001553
Mike Snitzerd07335e2010-11-16 12:52:38 +01001554 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1555 bdev->bd_dev,
1556 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 }
1558}
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560static void handle_bad_sector(struct bio *bio)
1561{
1562 char b[BDEVNAME_SIZE];
1563
1564 printk(KERN_INFO "attempt to access beyond end of device\n");
1565 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1566 bdevname(bio->bi_bdev, b),
1567 bio->bi_rw,
1568 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001569 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
1571 set_bit(BIO_EOF, &bio->bi_flags);
1572}
1573
Akinobu Mitac17bb492006-12-08 02:39:46 -08001574#ifdef CONFIG_FAIL_MAKE_REQUEST
1575
1576static DECLARE_FAULT_ATTR(fail_make_request);
1577
1578static int __init setup_fail_make_request(char *str)
1579{
1580 return setup_fault_attr(&fail_make_request, str);
1581}
1582__setup("fail_make_request=", setup_fail_make_request);
1583
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001584static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001585{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001586 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001587}
1588
1589static int __init fail_make_request_debugfs(void)
1590{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001591 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1592 NULL, &fail_make_request);
1593
1594 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001595}
1596
1597late_initcall(fail_make_request_debugfs);
1598
1599#else /* CONFIG_FAIL_MAKE_REQUEST */
1600
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001601static inline bool should_fail_request(struct hd_struct *part,
1602 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001603{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001604 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001605}
1606
1607#endif /* CONFIG_FAIL_MAKE_REQUEST */
1608
Jens Axboec07e2b42007-07-18 13:27:58 +02001609/*
1610 * Check whether this bio extends beyond the end of the device.
1611 */
1612static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1613{
1614 sector_t maxsector;
1615
1616 if (!nr_sectors)
1617 return 0;
1618
1619 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001620 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001621 if (maxsector) {
1622 sector_t sector = bio->bi_sector;
1623
1624 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1625 /*
1626 * This may well happen - the kernel calls bread()
1627 * without checking the size of the device, e.g., when
1628 * mounting a device.
1629 */
1630 handle_bad_sector(bio);
1631 return 1;
1632 }
1633 }
1634
1635 return 0;
1636}
1637
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001638static noinline_for_stack bool
1639generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640{
Jens Axboe165125e2007-07-24 09:28:11 +02001641 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001642 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001643 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001644 char b[BDEVNAME_SIZE];
1645 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
1647 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Jens Axboec07e2b42007-07-18 13:27:58 +02001649 if (bio_check_eod(bio, nr_sectors))
1650 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001652 q = bdev_get_queue(bio->bi_bdev);
1653 if (unlikely(!q)) {
1654 printk(KERN_ERR
1655 "generic_make_request: Trying to access "
1656 "nonexistent block-device %s (%Lu)\n",
1657 bdevname(bio->bi_bdev, b),
1658 (long long) bio->bi_sector);
1659 goto end_io;
1660 }
1661
1662 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1663 nr_sectors > queue_max_hw_sectors(q))) {
1664 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1665 bdevname(bio->bi_bdev, b),
1666 bio_sectors(bio),
1667 queue_max_hw_sectors(q));
1668 goto end_io;
1669 }
1670
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001671 part = bio->bi_bdev->bd_part;
1672 if (should_fail_request(part, bio->bi_size) ||
1673 should_fail_request(&part_to_disk(part)->part0,
1674 bio->bi_size))
1675 goto end_io;
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001678 * If this device has partitions, remap block n
1679 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001681 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001683 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1684 goto end_io;
1685
1686 if (bio_check_eod(bio, nr_sectors))
1687 goto end_io;
1688
1689 /*
1690 * Filter flush bio's early so that make_request based
1691 * drivers without flush support don't have to worry
1692 * about them.
1693 */
1694 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1695 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1696 if (!nr_sectors) {
1697 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001698 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001702 if ((bio->bi_rw & REQ_DISCARD) &&
1703 (!blk_queue_discard(q) ||
1704 ((bio->bi_rw & REQ_SECURE) &&
1705 !blk_queue_secdiscard(q)))) {
1706 err = -EOPNOTSUPP;
1707 goto end_io;
1708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001710 /*
1711 * Various block parts want %current->io_context and lazy ioc
1712 * allocation ends up trading a lot of pain for a small amount of
1713 * memory. Just allocate it upfront. This may fail and block
1714 * layer knows how to live with it.
1715 */
1716 create_io_context(GFP_ATOMIC, q->node);
1717
Tejun Heobc16a4f2011-10-19 14:33:01 +02001718 if (blk_throtl_bio(q, bio))
1719 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001720
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001721 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001722 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001723
1724end_io:
1725 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001726 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727}
1728
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001729/**
1730 * generic_make_request - hand a buffer to its device driver for I/O
1731 * @bio: The bio describing the location in memory and on the device.
1732 *
1733 * generic_make_request() is used to make I/O requests of block
1734 * devices. It is passed a &struct bio, which describes the I/O that needs
1735 * to be done.
1736 *
1737 * generic_make_request() does not return any status. The
1738 * success/failure status of the request, along with notification of
1739 * completion, is delivered asynchronously through the bio->bi_end_io
1740 * function described (one day) else where.
1741 *
1742 * The caller of generic_make_request must make sure that bi_io_vec
1743 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1744 * set to describe the device address, and the
1745 * bi_end_io and optionally bi_private are set to describe how
1746 * completion notification should be signaled.
1747 *
1748 * generic_make_request and the drivers it calls may use bi_next if this
1749 * bio happens to be merged with someone else, and may resubmit the bio to
1750 * a lower device by calling into generic_make_request recursively, which
1751 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001752 */
1753void generic_make_request(struct bio *bio)
1754{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001755 struct bio_list bio_list_on_stack;
1756
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001757 if (!generic_make_request_checks(bio))
1758 return;
1759
1760 /*
1761 * We only want one ->make_request_fn to be active at a time, else
1762 * stack usage with stacked devices could be a problem. So use
1763 * current->bio_list to keep a list of requests submited by a
1764 * make_request_fn function. current->bio_list is also used as a
1765 * flag to say if generic_make_request is currently active in this
1766 * task or not. If it is NULL, then no make_request is active. If
1767 * it is non-NULL, then a make_request is active, and new requests
1768 * should be added at the tail
1769 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001770 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001771 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001772 return;
1773 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001774
Neil Brownd89d8792007-05-01 09:53:42 +02001775 /* following loop may be a bit non-obvious, and so deserves some
1776 * explanation.
1777 * Before entering the loop, bio->bi_next is NULL (as all callers
1778 * ensure that) so we have a list with a single bio.
1779 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001780 * we assign bio_list to a pointer to the bio_list_on_stack,
1781 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001782 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001783 * through a recursive call to generic_make_request. If it
1784 * did, we find a non-NULL value in bio_list and re-enter the loop
1785 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001786 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001787 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001788 */
1789 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001790 bio_list_init(&bio_list_on_stack);
1791 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001792 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001793 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1794
1795 q->make_request_fn(q, bio);
1796
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001797 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001798 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001799 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001800}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801EXPORT_SYMBOL(generic_make_request);
1802
1803/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001804 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1806 * @bio: The &struct bio which describes the I/O
1807 *
1808 * submit_bio() is very similar in purpose to generic_make_request(), and
1809 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001810 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 *
1812 */
1813void submit_bio(int rw, struct bio *bio)
1814{
1815 int count = bio_sectors(bio);
1816
Jens Axboe22e2c502005-06-27 10:55:12 +02001817 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Jens Axboebf2de6f2007-09-27 13:01:25 +02001819 /*
1820 * If it's a regular read/write or a barrier with data attached,
1821 * go through the normal accounting stuff before submission.
1822 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001823 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001824 if (rw & WRITE) {
1825 count_vm_events(PGPGOUT, count);
1826 } else {
1827 task_io_account_read(bio->bi_size);
1828 count_vm_events(PGPGIN, count);
1829 }
1830
1831 if (unlikely(block_dump)) {
1832 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001833 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001834 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001835 (rw & WRITE) ? "WRITE" : "READ",
1836 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001837 bdevname(bio->bi_bdev, b),
1838 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 }
1841
1842 generic_make_request(bio);
1843}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844EXPORT_SYMBOL(submit_bio);
1845
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001846/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001847 * blk_rq_check_limits - Helper function to check a request for the queue limit
1848 * @q: the queue
1849 * @rq: the request being checked
1850 *
1851 * Description:
1852 * @rq may have been made based on weaker limitations of upper-level queues
1853 * in request stacking drivers, and it may violate the limitation of @q.
1854 * Since the block layer and the underlying device driver trust @rq
1855 * after it is inserted to @q, it should be checked against @q before
1856 * the insertion using this generic function.
1857 *
1858 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001859 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001860 * Request stacking drivers like request-based dm may change the queue
1861 * limits while requests are in the queue (e.g. dm's table swapping).
1862 * Such request stacking drivers should check those requests agaist
1863 * the new queue limits again when they dispatch those requests,
1864 * although such checkings are also done against the old queue limits
1865 * when submitting requests.
1866 */
1867int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1868{
ike Snitzer33839772010-08-08 12:11:33 -04001869 if (rq->cmd_flags & REQ_DISCARD)
1870 return 0;
1871
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001872 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1873 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001874 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1875 return -EIO;
1876 }
1877
1878 /*
1879 * queue's settings related to segment counting like q->bounce_pfn
1880 * may differ from that of other stacking queues.
1881 * Recalculate it to check the request correctly on this queue's
1882 * limitation.
1883 */
1884 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001885 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001886 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1887 return -EIO;
1888 }
1889
1890 return 0;
1891}
1892EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1893
1894/**
1895 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1896 * @q: the queue to submit the request
1897 * @rq: the request being queued
1898 */
1899int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1900{
1901 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001902 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001903
1904 if (blk_rq_check_limits(q, rq))
1905 return -EIO;
1906
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001907 if (rq->rq_disk &&
1908 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001909 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001910
1911 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo8ba61432011-12-14 00:33:37 +01001912 if (unlikely(blk_queue_dead(q))) {
1913 spin_unlock_irqrestore(q->queue_lock, flags);
1914 return -ENODEV;
1915 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001916
1917 /*
1918 * Submitting request must be dequeued before calling this function
1919 * because it will be linked to another request_queue
1920 */
1921 BUG_ON(blk_queued_rq(rq));
1922
Jeff Moyer4853aba2011-08-15 21:37:25 +02001923 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1924 where = ELEVATOR_INSERT_FLUSH;
1925
1926 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001927 if (where == ELEVATOR_INSERT_FLUSH)
1928 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001929 spin_unlock_irqrestore(q->queue_lock, flags);
1930
1931 return 0;
1932}
1933EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1934
Tejun Heo80a761f2009-07-03 17:48:17 +09001935/**
1936 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1937 * @rq: request to examine
1938 *
1939 * Description:
1940 * A request could be merge of IOs which require different failure
1941 * handling. This function determines the number of bytes which
1942 * can be failed from the beginning of the request without
1943 * crossing into area which need to be retried further.
1944 *
1945 * Return:
1946 * The number of bytes to fail.
1947 *
1948 * Context:
1949 * queue_lock must be held.
1950 */
1951unsigned int blk_rq_err_bytes(const struct request *rq)
1952{
1953 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1954 unsigned int bytes = 0;
1955 struct bio *bio;
1956
1957 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1958 return blk_rq_bytes(rq);
1959
1960 /*
1961 * Currently the only 'mixing' which can happen is between
1962 * different fastfail types. We can safely fail portions
1963 * which have all the failfast bits that the first one has -
1964 * the ones which are at least as eager to fail as the first
1965 * one.
1966 */
1967 for (bio = rq->bio; bio; bio = bio->bi_next) {
1968 if ((bio->bi_rw & ff) != ff)
1969 break;
1970 bytes += bio->bi_size;
1971 }
1972
1973 /* this could lead to infinite loop */
1974 BUG_ON(blk_rq_bytes(rq) && !bytes);
1975 return bytes;
1976}
1977EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1978
Jens Axboebc58ba92009-01-23 10:54:44 +01001979static void blk_account_io_completion(struct request *req, unsigned int bytes)
1980{
Jens Axboec2553b52009-04-24 08:10:11 +02001981 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001982 const int rw = rq_data_dir(req);
1983 struct hd_struct *part;
1984 int cpu;
1985
1986 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001987 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001988 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1989 part_stat_unlock();
1990 }
1991}
1992
1993static void blk_account_io_done(struct request *req)
1994{
Jens Axboebc58ba92009-01-23 10:54:44 +01001995 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001996 * Account IO completion. flush_rq isn't accounted as a
1997 * normal IO on queueing nor completion. Accounting the
1998 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001999 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002000 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002001 unsigned long duration = jiffies - req->start_time;
2002 const int rw = rq_data_dir(req);
2003 struct hd_struct *part;
2004 int cpu;
2005
2006 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002007 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002008
2009 part_stat_inc(cpu, part, ios[rw]);
2010 part_stat_add(cpu, part, ticks[rw], duration);
2011 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002012 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002013
Jens Axboe6c23a962011-01-07 08:43:37 +01002014 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002015 part_stat_unlock();
2016 }
2017}
2018
Tejun Heo53a08802008-12-03 12:41:26 +01002019/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002020 * blk_peek_request - peek at the top of a request queue
2021 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002022 *
2023 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002024 * Return the request at the top of @q. The returned request
2025 * should be started using blk_start_request() before LLD starts
2026 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002027 *
2028 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002029 * Pointer to the request at the top of @q if available. Null
2030 * otherwise.
2031 *
2032 * Context:
2033 * queue_lock must be held.
2034 */
2035struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002036{
2037 struct request *rq;
2038 int ret;
2039
2040 while ((rq = __elv_next_request(q)) != NULL) {
2041 if (!(rq->cmd_flags & REQ_STARTED)) {
2042 /*
2043 * This is the first time the device driver
2044 * sees this request (possibly after
2045 * requeueing). Notify IO scheduler.
2046 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002047 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002048 elv_activate_rq(q, rq);
2049
2050 /*
2051 * just mark as started even if we don't start
2052 * it, a request that has been delayed should
2053 * not be passed by new incoming requests
2054 */
2055 rq->cmd_flags |= REQ_STARTED;
2056 trace_block_rq_issue(q, rq);
2057 }
2058
2059 if (!q->boundary_rq || q->boundary_rq == rq) {
2060 q->end_sector = rq_end_sector(rq);
2061 q->boundary_rq = NULL;
2062 }
2063
2064 if (rq->cmd_flags & REQ_DONTPREP)
2065 break;
2066
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002067 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002068 /*
2069 * make sure space for the drain appears we
2070 * know we can do this because max_hw_segments
2071 * has been adjusted to be one fewer than the
2072 * device can handle
2073 */
2074 rq->nr_phys_segments++;
2075 }
2076
2077 if (!q->prep_rq_fn)
2078 break;
2079
2080 ret = q->prep_rq_fn(q, rq);
2081 if (ret == BLKPREP_OK) {
2082 break;
2083 } else if (ret == BLKPREP_DEFER) {
2084 /*
2085 * the request may have been (partially) prepped.
2086 * we need to keep this request in the front to
2087 * avoid resource deadlock. REQ_STARTED will
2088 * prevent other fs requests from passing this one.
2089 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002090 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002091 !(rq->cmd_flags & REQ_DONTPREP)) {
2092 /*
2093 * remove the space for the drain we added
2094 * so that we don't add it again
2095 */
2096 --rq->nr_phys_segments;
2097 }
2098
2099 rq = NULL;
2100 break;
2101 } else if (ret == BLKPREP_KILL) {
2102 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002103 /*
2104 * Mark this request as started so we don't trigger
2105 * any debug logic in the end I/O path.
2106 */
2107 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002108 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002109 } else {
2110 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2111 break;
2112 }
2113 }
2114
2115 return rq;
2116}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002117EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002118
Tejun Heo9934c8c2009-05-08 11:54:16 +09002119void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002120{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002121 struct request_queue *q = rq->q;
2122
Tejun Heo158dbda2009-04-23 11:05:18 +09002123 BUG_ON(list_empty(&rq->queuelist));
2124 BUG_ON(ELV_ON_HASH(rq));
2125
2126 list_del_init(&rq->queuelist);
2127
2128 /*
2129 * the time frame between a request being removed from the lists
2130 * and to it is freed is accounted as io that is in progress at
2131 * the driver side.
2132 */
Divyesh Shah91952912010-04-01 15:01:41 -07002133 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002134 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002135 set_io_start_time_ns(rq);
2136 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002137}
2138
Tejun Heo5efccd12009-04-23 11:05:18 +09002139/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002140 * blk_start_request - start request processing on the driver
2141 * @req: request to dequeue
2142 *
2143 * Description:
2144 * Dequeue @req and start timeout timer on it. This hands off the
2145 * request to the driver.
2146 *
2147 * Block internal functions which don't want to start timer should
2148 * call blk_dequeue_request().
2149 *
2150 * Context:
2151 * queue_lock must be held.
2152 */
2153void blk_start_request(struct request *req)
2154{
2155 blk_dequeue_request(req);
2156
2157 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002158 * We are now handing the request to the hardware, initialize
2159 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002160 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002161 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002162 if (unlikely(blk_bidi_rq(req)))
2163 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2164
Tejun Heo9934c8c2009-05-08 11:54:16 +09002165 blk_add_timer(req);
2166}
2167EXPORT_SYMBOL(blk_start_request);
2168
2169/**
2170 * blk_fetch_request - fetch a request from a request queue
2171 * @q: request queue to fetch a request from
2172 *
2173 * Description:
2174 * Return the request at the top of @q. The request is started on
2175 * return and LLD can start processing it immediately.
2176 *
2177 * Return:
2178 * Pointer to the request at the top of @q if available. Null
2179 * otherwise.
2180 *
2181 * Context:
2182 * queue_lock must be held.
2183 */
2184struct request *blk_fetch_request(struct request_queue *q)
2185{
2186 struct request *rq;
2187
2188 rq = blk_peek_request(q);
2189 if (rq)
2190 blk_start_request(rq);
2191 return rq;
2192}
2193EXPORT_SYMBOL(blk_fetch_request);
2194
2195/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002196 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002197 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002198 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002199 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002200 *
2201 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002202 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2203 * the request structure even if @req doesn't have leftover.
2204 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002205 *
2206 * This special helper function is only for request stacking drivers
2207 * (e.g. request-based dm) so that they can handle partial completion.
2208 * Actual device drivers should use blk_end_request instead.
2209 *
2210 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2211 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002212 *
2213 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002214 * %false - this request doesn't have any more data
2215 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002216 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002217bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002219 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 struct bio *bio;
2221
Tejun Heo2e60e022009-04-23 11:05:18 +09002222 if (!req->bio)
2223 return false;
2224
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002225 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002226
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002228 * For fs requests, rq is just carrier of independent bio's
2229 * and each partial completion should be handled separately.
2230 * Reset per-request error on each partial completion.
2231 *
2232 * TODO: tj: This is too subtle. It would be better to let
2233 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002235 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 req->errors = 0;
2237
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002238 if (error && req->cmd_type == REQ_TYPE_FS &&
2239 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002240 char *error_type;
2241
2242 switch (error) {
2243 case -ENOLINK:
2244 error_type = "recoverable transport";
2245 break;
2246 case -EREMOTEIO:
2247 error_type = "critical target";
2248 break;
2249 case -EBADE:
2250 error_type = "critical nexus";
2251 break;
2252 case -EIO:
2253 default:
2254 error_type = "I/O";
2255 break;
2256 }
Yi Zou37d7b342012-08-30 16:26:25 -07002257 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2258 error_type, req->rq_disk ?
2259 req->rq_disk->disk_name : "?",
2260 (unsigned long long)blk_rq_pos(req));
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 }
2263
Jens Axboebc58ba92009-01-23 10:54:44 +01002264 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 total_bytes = bio_nbytes = 0;
2267 while ((bio = req->bio) != NULL) {
2268 int nbytes;
2269
2270 if (nr_bytes >= bio->bi_size) {
2271 req->bio = bio->bi_next;
2272 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002273 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 next_idx = 0;
2275 bio_nbytes = 0;
2276 } else {
2277 int idx = bio->bi_idx + next_idx;
2278
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002279 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002281 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002282 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 break;
2284 }
2285
2286 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2287 BIO_BUG_ON(nbytes > bio->bi_size);
2288
2289 /*
2290 * not a complete bvec done
2291 */
2292 if (unlikely(nbytes > nr_bytes)) {
2293 bio_nbytes += nr_bytes;
2294 total_bytes += nr_bytes;
2295 break;
2296 }
2297
2298 /*
2299 * advance to the next vector
2300 */
2301 next_idx++;
2302 bio_nbytes += nbytes;
2303 }
2304
2305 total_bytes += nbytes;
2306 nr_bytes -= nbytes;
2307
Jens Axboe6728cb02008-01-31 13:03:55 +01002308 bio = req->bio;
2309 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 /*
2311 * end more in this run, or just return 'not-done'
2312 */
2313 if (unlikely(nr_bytes <= 0))
2314 break;
2315 }
2316 }
2317
2318 /*
2319 * completely done
2320 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002321 if (!req->bio) {
2322 /*
2323 * Reset counters so that the request stacking driver
2324 * can find how many bytes remain in the request
2325 * later.
2326 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002327 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002328 return false;
2329 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
2331 /*
2332 * if the request wasn't completed, update state
2333 */
2334 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002335 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 bio->bi_idx += next_idx;
2337 bio_iovec(bio)->bv_offset += nr_bytes;
2338 bio_iovec(bio)->bv_len -= nr_bytes;
2339 }
2340
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002341 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002342 req->buffer = bio_data(req->bio);
2343
2344 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002345 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002346 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002347
Tejun Heo80a761f2009-07-03 17:48:17 +09002348 /* mixed attributes always follow the first bio */
2349 if (req->cmd_flags & REQ_MIXED_MERGE) {
2350 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2351 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2352 }
2353
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002354 /*
2355 * If total number of sectors is less than the first segment
2356 * size, something has gone terribly wrong.
2357 */
2358 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002359 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002360 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002361 }
2362
2363 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002365
Tejun Heo2e60e022009-04-23 11:05:18 +09002366 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367}
Tejun Heo2e60e022009-04-23 11:05:18 +09002368EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Tejun Heo2e60e022009-04-23 11:05:18 +09002370static bool blk_update_bidi_request(struct request *rq, int error,
2371 unsigned int nr_bytes,
2372 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002373{
Tejun Heo2e60e022009-04-23 11:05:18 +09002374 if (blk_update_request(rq, error, nr_bytes))
2375 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002376
Tejun Heo2e60e022009-04-23 11:05:18 +09002377 /* Bidi request must be completed as a whole */
2378 if (unlikely(blk_bidi_rq(rq)) &&
2379 blk_update_request(rq->next_rq, error, bidi_bytes))
2380 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002381
Jens Axboee2e1a142010-06-09 10:42:09 +02002382 if (blk_queue_add_random(rq->q))
2383 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002384
2385 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386}
2387
James Bottomley28018c22010-07-01 19:49:17 +09002388/**
2389 * blk_unprep_request - unprepare a request
2390 * @req: the request
2391 *
2392 * This function makes a request ready for complete resubmission (or
2393 * completion). It happens only after all error handling is complete,
2394 * so represents the appropriate moment to deallocate any resources
2395 * that were allocated to the request in the prep_rq_fn. The queue
2396 * lock is held when calling this.
2397 */
2398void blk_unprep_request(struct request *req)
2399{
2400 struct request_queue *q = req->q;
2401
2402 req->cmd_flags &= ~REQ_DONTPREP;
2403 if (q->unprep_rq_fn)
2404 q->unprep_rq_fn(q, req);
2405}
2406EXPORT_SYMBOL_GPL(blk_unprep_request);
2407
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408/*
2409 * queue lock must be held
2410 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002411static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002413 if (blk_rq_tagged(req))
2414 blk_queue_end_tag(req->q, req);
2415
James Bottomleyba396a62009-05-27 14:17:08 +02002416 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002418 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002419 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
Mike Andersone78042e2008-10-30 02:16:20 -07002421 blk_delete_timer(req);
2422
James Bottomley28018c22010-07-01 19:49:17 +09002423 if (req->cmd_flags & REQ_DONTPREP)
2424 blk_unprep_request(req);
2425
2426
Jens Axboebc58ba92009-01-23 10:54:44 +01002427 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002428
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002430 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002431 else {
2432 if (blk_bidi_rq(req))
2433 __blk_put_request(req->next_rq->q, req->next_rq);
2434
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
2437}
2438
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002439/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002440 * blk_end_bidi_request - Complete a bidi request
2441 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002442 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002443 * @nr_bytes: number of bytes to complete @rq
2444 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002445 *
2446 * Description:
2447 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002448 * Drivers that supports bidi can safely call this member for any
2449 * type of request, bidi or uni. In the later case @bidi_bytes is
2450 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002451 *
2452 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002453 * %false - we are done with this request
2454 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002455 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002456static bool blk_end_bidi_request(struct request *rq, int error,
2457 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002458{
2459 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002460 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002461
Tejun Heo2e60e022009-04-23 11:05:18 +09002462 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2463 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002464
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002465 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002466 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002467 spin_unlock_irqrestore(q->queue_lock, flags);
2468
Tejun Heo2e60e022009-04-23 11:05:18 +09002469 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002470}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002471
2472/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002473 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2474 * @rq: the request to complete
2475 * @error: %0 for success, < %0 for error
2476 * @nr_bytes: number of bytes to complete @rq
2477 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002478 *
2479 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002480 * Identical to blk_end_bidi_request() except that queue lock is
2481 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002482 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002483 * Return:
2484 * %false - we are done with this request
2485 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002486 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002487bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002488 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002489{
Tejun Heo2e60e022009-04-23 11:05:18 +09002490 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2491 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002492
Tejun Heo2e60e022009-04-23 11:05:18 +09002493 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002494
Tejun Heo2e60e022009-04-23 11:05:18 +09002495 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002496}
2497
2498/**
2499 * blk_end_request - Helper function for drivers to complete the request.
2500 * @rq: the request being processed
2501 * @error: %0 for success, < %0 for error
2502 * @nr_bytes: number of bytes to complete
2503 *
2504 * Description:
2505 * Ends I/O on a number of bytes attached to @rq.
2506 * If @rq has leftover, sets it up for the next range of segments.
2507 *
2508 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002509 * %false - we are done with this request
2510 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002511 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002512bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002513{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002514 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002515}
Jens Axboe56ad1742009-07-28 22:11:24 +02002516EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002517
2518/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002519 * blk_end_request_all - Helper function for drives to finish the request.
2520 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002521 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002522 *
2523 * Description:
2524 * Completely finish @rq.
2525 */
2526void blk_end_request_all(struct request *rq, int error)
2527{
2528 bool pending;
2529 unsigned int bidi_bytes = 0;
2530
2531 if (unlikely(blk_bidi_rq(rq)))
2532 bidi_bytes = blk_rq_bytes(rq->next_rq);
2533
2534 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2535 BUG_ON(pending);
2536}
Jens Axboe56ad1742009-07-28 22:11:24 +02002537EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002538
2539/**
2540 * blk_end_request_cur - Helper function to finish the current request chunk.
2541 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002542 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002543 *
2544 * Description:
2545 * Complete the current consecutively mapped chunk from @rq.
2546 *
2547 * Return:
2548 * %false - we are done with this request
2549 * %true - still buffers pending for this request
2550 */
2551bool blk_end_request_cur(struct request *rq, int error)
2552{
2553 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2554}
Jens Axboe56ad1742009-07-28 22:11:24 +02002555EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002556
2557/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002558 * blk_end_request_err - Finish a request till the next failure boundary.
2559 * @rq: the request to finish till the next failure boundary for
2560 * @error: must be negative errno
2561 *
2562 * Description:
2563 * Complete @rq till the next failure boundary.
2564 *
2565 * Return:
2566 * %false - we are done with this request
2567 * %true - still buffers pending for this request
2568 */
2569bool blk_end_request_err(struct request *rq, int error)
2570{
2571 WARN_ON(error >= 0);
2572 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2573}
2574EXPORT_SYMBOL_GPL(blk_end_request_err);
2575
2576/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002577 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002578 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002579 * @error: %0 for success, < %0 for error
2580 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002581 *
2582 * Description:
2583 * Must be called with queue lock held unlike blk_end_request().
2584 *
2585 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002586 * %false - we are done with this request
2587 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002588 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002589bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002590{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002591 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002592}
Jens Axboe56ad1742009-07-28 22:11:24 +02002593EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002594
2595/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002596 * __blk_end_request_all - Helper function for drives to finish the request.
2597 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002598 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002599 *
2600 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002601 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002602 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002603void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002604{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002605 bool pending;
2606 unsigned int bidi_bytes = 0;
2607
2608 if (unlikely(blk_bidi_rq(rq)))
2609 bidi_bytes = blk_rq_bytes(rq->next_rq);
2610
2611 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2612 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002613}
Jens Axboe56ad1742009-07-28 22:11:24 +02002614EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002615
2616/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002617 * __blk_end_request_cur - Helper function to finish the current request chunk.
2618 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002619 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002620 *
2621 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002622 * Complete the current consecutively mapped chunk from @rq. Must
2623 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002624 *
2625 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002626 * %false - we are done with this request
2627 * %true - still buffers pending for this request
2628 */
2629bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002630{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002631 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002632}
Jens Axboe56ad1742009-07-28 22:11:24 +02002633EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002634
Tejun Heo80a761f2009-07-03 17:48:17 +09002635/**
2636 * __blk_end_request_err - Finish a request till the next failure boundary.
2637 * @rq: the request to finish till the next failure boundary for
2638 * @error: must be negative errno
2639 *
2640 * Description:
2641 * Complete @rq till the next failure boundary. Must be called
2642 * with queue lock held.
2643 *
2644 * Return:
2645 * %false - we are done with this request
2646 * %true - still buffers pending for this request
2647 */
2648bool __blk_end_request_err(struct request *rq, int error)
2649{
2650 WARN_ON(error >= 0);
2651 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2652}
2653EXPORT_SYMBOL_GPL(__blk_end_request_err);
2654
Jens Axboe86db1e22008-01-29 14:53:40 +01002655void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2656 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002658 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002659 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
David Woodhousefb2dce82008-08-05 18:01:53 +01002661 if (bio_has_data(bio)) {
2662 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002663 rq->buffer = bio_data(bio);
2664 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002665 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667
NeilBrown66846572007-08-16 13:31:28 +02002668 if (bio->bi_bdev)
2669 rq->rq_disk = bio->bi_bdev->bd_disk;
2670}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002672#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2673/**
2674 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2675 * @rq: the request to be flushed
2676 *
2677 * Description:
2678 * Flush all pages in @rq.
2679 */
2680void rq_flush_dcache_pages(struct request *rq)
2681{
2682 struct req_iterator iter;
2683 struct bio_vec *bvec;
2684
2685 rq_for_each_segment(bvec, rq, iter)
2686 flush_dcache_page(bvec->bv_page);
2687}
2688EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2689#endif
2690
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002691/**
2692 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2693 * @q : the queue of the device being checked
2694 *
2695 * Description:
2696 * Check if underlying low-level drivers of a device are busy.
2697 * If the drivers want to export their busy state, they must set own
2698 * exporting function using blk_queue_lld_busy() first.
2699 *
2700 * Basically, this function is used only by request stacking drivers
2701 * to stop dispatching requests to underlying devices when underlying
2702 * devices are busy. This behavior helps more I/O merging on the queue
2703 * of the request stacking driver and prevents I/O throughput regression
2704 * on burst I/O load.
2705 *
2706 * Return:
2707 * 0 - Not busy (The request stacking driver should dispatch request)
2708 * 1 - Busy (The request stacking driver should stop dispatching request)
2709 */
2710int blk_lld_busy(struct request_queue *q)
2711{
2712 if (q->lld_busy_fn)
2713 return q->lld_busy_fn(q);
2714
2715 return 0;
2716}
2717EXPORT_SYMBOL_GPL(blk_lld_busy);
2718
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002719/**
2720 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2721 * @rq: the clone request to be cleaned up
2722 *
2723 * Description:
2724 * Free all bios in @rq for a cloned request.
2725 */
2726void blk_rq_unprep_clone(struct request *rq)
2727{
2728 struct bio *bio;
2729
2730 while ((bio = rq->bio) != NULL) {
2731 rq->bio = bio->bi_next;
2732
2733 bio_put(bio);
2734 }
2735}
2736EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2737
2738/*
2739 * Copy attributes of the original request to the clone request.
2740 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2741 */
2742static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2743{
2744 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002745 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002746 dst->cmd_type = src->cmd_type;
2747 dst->__sector = blk_rq_pos(src);
2748 dst->__data_len = blk_rq_bytes(src);
2749 dst->nr_phys_segments = src->nr_phys_segments;
2750 dst->ioprio = src->ioprio;
2751 dst->extra_len = src->extra_len;
2752}
2753
2754/**
2755 * blk_rq_prep_clone - Helper function to setup clone request
2756 * @rq: the request to be setup
2757 * @rq_src: original request to be cloned
2758 * @bs: bio_set that bios for clone are allocated from
2759 * @gfp_mask: memory allocation mask for bio
2760 * @bio_ctr: setup function to be called for each clone bio.
2761 * Returns %0 for success, non %0 for failure.
2762 * @data: private data to be passed to @bio_ctr
2763 *
2764 * Description:
2765 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2766 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2767 * are not copied, and copying such parts is the caller's responsibility.
2768 * Also, pages which the original bios are pointing to are not copied
2769 * and the cloned bios just point same pages.
2770 * So cloned bios must be completed before original bios, which means
2771 * the caller must complete @rq before @rq_src.
2772 */
2773int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2774 struct bio_set *bs, gfp_t gfp_mask,
2775 int (*bio_ctr)(struct bio *, struct bio *, void *),
2776 void *data)
2777{
2778 struct bio *bio, *bio_src;
2779
2780 if (!bs)
2781 bs = fs_bio_set;
2782
2783 blk_rq_init(NULL, rq);
2784
2785 __rq_for_each_bio(bio_src, rq_src) {
2786 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2787 if (!bio)
2788 goto free_and_out;
2789
2790 __bio_clone(bio, bio_src);
2791
2792 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002793 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002794 goto free_and_out;
2795
2796 if (bio_ctr && bio_ctr(bio, bio_src, data))
2797 goto free_and_out;
2798
2799 if (rq->bio) {
2800 rq->biotail->bi_next = bio;
2801 rq->biotail = bio;
2802 } else
2803 rq->bio = rq->biotail = bio;
2804 }
2805
2806 __blk_rq_prep_clone(rq, rq_src);
2807
2808 return 0;
2809
2810free_and_out:
2811 if (bio)
2812 bio_free(bio, bs);
2813 blk_rq_unprep_clone(rq);
2814
2815 return -ENOMEM;
2816}
2817EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2818
Jens Axboe18887ad2008-07-28 13:08:45 +02002819int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820{
2821 return queue_work(kblockd_workqueue, work);
2822}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823EXPORT_SYMBOL(kblockd_schedule_work);
2824
Vivek Goyale43473b2010-09-15 17:06:35 -04002825int kblockd_schedule_delayed_work(struct request_queue *q,
2826 struct delayed_work *dwork, unsigned long delay)
2827{
2828 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2829}
2830EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2831
Jens Axboe73c10102011-03-08 13:19:51 +01002832#define PLUG_MAGIC 0x91827364
2833
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002834/**
2835 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2836 * @plug: The &struct blk_plug that needs to be initialized
2837 *
2838 * Description:
2839 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2840 * pending I/O should the task end up blocking between blk_start_plug() and
2841 * blk_finish_plug(). This is important from a performance perspective, but
2842 * also ensures that we don't deadlock. For instance, if the task is blocking
2843 * for a memory allocation, memory reclaim could end up wanting to free a
2844 * page belonging to that request that is currently residing in our private
2845 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2846 * this kind of deadlock.
2847 */
Jens Axboe73c10102011-03-08 13:19:51 +01002848void blk_start_plug(struct blk_plug *plug)
2849{
2850 struct task_struct *tsk = current;
2851
2852 plug->magic = PLUG_MAGIC;
2853 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002854 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002855 plug->should_sort = 0;
2856
2857 /*
2858 * If this is a nested plug, don't actually assign it. It will be
2859 * flushed on its own.
2860 */
2861 if (!tsk->plug) {
2862 /*
2863 * Store ordering should not be needed here, since a potential
2864 * preempt will imply a full memory barrier
2865 */
2866 tsk->plug = plug;
2867 }
2868}
2869EXPORT_SYMBOL(blk_start_plug);
2870
2871static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2872{
2873 struct request *rqa = container_of(a, struct request, queuelist);
2874 struct request *rqb = container_of(b, struct request, queuelist);
2875
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002876 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002877}
2878
Jens Axboe49cac012011-04-16 13:51:05 +02002879/*
2880 * If 'from_schedule' is true, then postpone the dispatch of requests
2881 * until a safe kblockd context. We due this to avoid accidental big
2882 * additional stack usage in driver dispatch, in places where the originally
2883 * plugger did not intend it.
2884 */
Jens Axboef6603782011-04-15 15:49:07 +02002885static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002886 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002887 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002888{
Jens Axboe49cac012011-04-16 13:51:05 +02002889 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002890
2891 /*
Tejun Heo8ba61432011-12-14 00:33:37 +01002892 * Don't mess with dead queue.
2893 */
2894 if (unlikely(blk_queue_dead(q))) {
2895 spin_unlock(q->queue_lock);
2896 return;
2897 }
2898
2899 /*
Jens Axboe99e22592011-04-18 09:59:55 +02002900 * If we are punting this to kblockd, then we can safely drop
2901 * the queue_lock before waking kblockd (which needs to take
2902 * this lock).
2903 */
2904 if (from_schedule) {
2905 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002906 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002907 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002908 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002909 spin_unlock(q->queue_lock);
2910 }
2911
Jens Axboe94b5eb22011-04-12 10:12:19 +02002912}
2913
NeilBrown74018dc2012-07-31 09:08:15 +02002914static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002915{
2916 LIST_HEAD(callbacks);
2917
Shaohua Li2a7d5552012-07-31 09:08:15 +02002918 while (!list_empty(&plug->cb_list)) {
2919 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002920
Shaohua Li2a7d5552012-07-31 09:08:15 +02002921 while (!list_empty(&callbacks)) {
2922 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002923 struct blk_plug_cb,
2924 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002925 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02002926 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002927 }
NeilBrown048c9372011-04-18 09:52:22 +02002928 }
2929}
2930
NeilBrown9cbb1752012-07-31 09:08:14 +02002931struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2932 int size)
2933{
2934 struct blk_plug *plug = current->plug;
2935 struct blk_plug_cb *cb;
2936
2937 if (!plug)
2938 return NULL;
2939
2940 list_for_each_entry(cb, &plug->cb_list, list)
2941 if (cb->callback == unplug && cb->data == data)
2942 return cb;
2943
2944 /* Not currently on the callback list */
2945 BUG_ON(size < sizeof(*cb));
2946 cb = kzalloc(size, GFP_ATOMIC);
2947 if (cb) {
2948 cb->data = data;
2949 cb->callback = unplug;
2950 list_add(&cb->list, &plug->cb_list);
2951 }
2952 return cb;
2953}
2954EXPORT_SYMBOL(blk_check_plugged);
2955
Jens Axboe49cac012011-04-16 13:51:05 +02002956void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002957{
2958 struct request_queue *q;
2959 unsigned long flags;
2960 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002961 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002962 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002963
2964 BUG_ON(plug->magic != PLUG_MAGIC);
2965
NeilBrown74018dc2012-07-31 09:08:15 +02002966 flush_plug_callbacks(plug, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002967 if (list_empty(&plug->list))
2968 return;
2969
NeilBrown109b8122011-04-11 14:13:10 +02002970 list_splice_init(&plug->list, &list);
2971
2972 if (plug->should_sort) {
2973 list_sort(NULL, &list, plug_rq_cmp);
2974 plug->should_sort = 0;
2975 }
Jens Axboe73c10102011-03-08 13:19:51 +01002976
2977 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002978 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002979
2980 /*
2981 * Save and disable interrupts here, to avoid doing it for every
2982 * queue lock we have to take.
2983 */
Jens Axboe73c10102011-03-08 13:19:51 +01002984 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002985 while (!list_empty(&list)) {
2986 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002987 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002988 BUG_ON(!rq->q);
2989 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002990 /*
2991 * This drops the queue lock
2992 */
2993 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002994 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002995 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002996 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002997 spin_lock(q->queue_lock);
2998 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002999
3000 /*
3001 * Short-circuit if @q is dead
3002 */
3003 if (unlikely(blk_queue_dead(q))) {
3004 __blk_end_request_all(rq, -ENODEV);
3005 continue;
3006 }
3007
Jens Axboe73c10102011-03-08 13:19:51 +01003008 /*
3009 * rq is already accounted, so use raw insert
3010 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003011 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3012 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3013 else
3014 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003015
3016 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003017 }
3018
Jens Axboe99e22592011-04-18 09:59:55 +02003019 /*
3020 * This drops the queue lock
3021 */
3022 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003023 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003024
Jens Axboe73c10102011-03-08 13:19:51 +01003025 local_irq_restore(flags);
3026}
Jens Axboe73c10102011-03-08 13:19:51 +01003027
3028void blk_finish_plug(struct blk_plug *plug)
3029{
Jens Axboef6603782011-04-15 15:49:07 +02003030 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003031
3032 if (plug == current->plug)
3033 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003034}
3035EXPORT_SYMBOL(blk_finish_plug);
3036
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037int __init blk_dev_init(void)
3038{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003039 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3040 sizeof(((struct request *)0)->cmd_flags));
3041
Tejun Heo89b90be2011-01-03 15:01:47 +01003042 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3043 kblockd_workqueue = alloc_workqueue("kblockd",
3044 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 if (!kblockd_workqueue)
3046 panic("Failed to create kblockd\n");
3047
3048 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003049 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050
Jens Axboe8324aa92008-01-29 14:51:59 +01003051 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003052 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 return 0;
3055}