blob: 3c923a7aeb56f1658142b091868c3c29ebffd3c5 [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{
364 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100365 bool drain = false;
366 int i;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200367
368 spin_lock_irq(q->queue_lock);
369
Tejun Heob855b042012-03-06 21:24:55 +0100370 /*
371 * The caller might be trying to drain @q before its
372 * elevator is initialized.
373 */
374 if (q->elevator)
375 elv_drain_elevator(q);
376
Tejun Heo5efd6112012-03-05 13:15:12 -0800377 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200378
Tejun Heo4eabc942011-12-15 20:03:04 +0100379 /*
380 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100381 * driver init after queue creation or is not yet fully
382 * active yet. Some drivers (e.g. fd and loop) get unhappy
383 * in such cases. Kick queue iff dispatch queue has
384 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100385 */
Tejun Heob855b042012-03-06 21:24:55 +0100386 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100387 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200388
Tejun Heo481a7d62011-12-14 00:33:37 +0100389 drain |= q->rq.elvpriv;
390
391 /*
392 * Unfortunately, requests are queued at and tracked from
393 * multiple places and there's no single counter which can
394 * be drained. Check all the queues and counters.
395 */
396 if (drain_all) {
397 drain |= !list_empty(&q->queue_head);
398 for (i = 0; i < 2; i++) {
399 drain |= q->rq.count[i];
400 drain |= q->in_flight[i];
401 drain |= !list_empty(&q->flush_queue[i]);
402 }
403 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404
405 spin_unlock_irq(q->queue_lock);
406
Tejun Heo481a7d62011-12-14 00:33:37 +0100407 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200408 break;
409 msleep(10);
410 }
411}
412
Tejun Heoc9a929d2011-10-19 14:42:16 +0200413/**
Tejun Heod7325802012-03-05 13:14:58 -0800414 * blk_queue_bypass_start - enter queue bypass mode
415 * @q: queue of interest
416 *
417 * In bypass mode, only the dispatch FIFO queue of @q is used. This
418 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800419 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700420 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
421 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800422 */
423void blk_queue_bypass_start(struct request_queue *q)
424{
Tejun Heob82d4b12012-04-13 13:11:31 -0700425 bool drain;
426
Tejun Heod7325802012-03-05 13:14:58 -0800427 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700428 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800429 queue_flag_set(QUEUE_FLAG_BYPASS, q);
430 spin_unlock_irq(q->queue_lock);
431
Tejun Heob82d4b12012-04-13 13:11:31 -0700432 if (drain) {
433 blk_drain_queue(q, false);
434 /* ensure blk_queue_bypass() is %true inside RCU read lock */
435 synchronize_rcu();
436 }
Tejun Heod7325802012-03-05 13:14:58 -0800437}
438EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
439
440/**
441 * blk_queue_bypass_end - leave queue bypass mode
442 * @q: queue of interest
443 *
444 * Leave bypass mode and restore the normal queueing behavior.
445 */
446void blk_queue_bypass_end(struct request_queue *q)
447{
448 spin_lock_irq(q->queue_lock);
449 if (!--q->bypass_depth)
450 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
451 WARN_ON_ONCE(q->bypass_depth < 0);
452 spin_unlock_irq(q->queue_lock);
453}
454EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
455
456/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200457 * blk_cleanup_queue - shutdown a request queue
458 * @q: request queue to shutdown
459 *
460 * Mark @q DEAD, drain all pending requests, destroy and put it. All
461 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500462 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100463void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500464{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200465 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700466
Tejun Heoc9a929d2011-10-19 14:42:16 +0200467 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500468 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200469 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200470
471 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800472
Tejun Heo80fd9972012-04-13 14:50:53 -0700473 /*
474 * Dead queue is permanently in bypass mode till released. Note
475 * that, unlike blk_queue_bypass_start(), we aren't performing
476 * synchronize_rcu() after entering bypass mode to avoid the delay
477 * as some drivers create and destroy a lot of queues while
478 * probing. This is still safe because blk_release_queue() will be
479 * called only after the queue refcnt drops to zero and nothing,
480 * RCU or not, would be traversing the queue by then.
481 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800482 q->bypass_depth++;
483 queue_flag_set(QUEUE_FLAG_BYPASS, q);
484
Tejun Heoc9a929d2011-10-19 14:42:16 +0200485 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
486 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
487 queue_flag_set(QUEUE_FLAG_DEAD, q);
Al Viro483f4af2006-03-18 18:34:37 -0500488
Hannes Reinecke777eb1b2011-09-28 08:07:01 -0600489 if (q->queue_lock != &q->__queue_lock)
490 q->queue_lock = &q->__queue_lock;
Vivek Goyalda527772011-03-02 19:05:33 -0500491
Tejun Heoc9a929d2011-10-19 14:42:16 +0200492 spin_unlock_irq(lock);
493 mutex_unlock(&q->sysfs_lock);
494
Tejun Heob855b042012-03-06 21:24:55 +0100495 /* drain all requests queued before DEAD marking */
496 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200497
498 /* @q won't process any more request, flush async actions */
499 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
500 blk_sync_queue(q);
501
502 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500503 blk_put_queue(q);
504}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505EXPORT_SYMBOL(blk_cleanup_queue);
506
Jens Axboe165125e2007-07-24 09:28:11 +0200507static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 struct request_list *rl = &q->rq;
510
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400511 if (unlikely(rl->rq_pool))
512 return 0;
513
Jens Axboe1faa16d2009-04-06 14:48:01 +0200514 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
515 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200516 rl->elvpriv = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200517 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
518 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Christoph Lameter19460892005-06-23 00:08:19 -0700520 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
521 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 if (!rl->rq_pool)
524 return -ENOMEM;
525
526 return 0;
527}
528
Jens Axboe165125e2007-07-24 09:28:11 +0200529struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530{
Christoph Lameter19460892005-06-23 00:08:19 -0700531 return blk_alloc_queue_node(gfp_mask, -1);
532}
533EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Jens Axboe165125e2007-07-24 09:28:11 +0200535struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700536{
Jens Axboe165125e2007-07-24 09:28:11 +0200537 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700538 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700539
Jens Axboe8324aa92008-01-29 14:51:59 +0100540 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700541 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (!q)
543 return NULL;
544
Dan Carpenter00380a42012-03-23 09:58:54 +0100545 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100546 if (q->id < 0)
547 goto fail_q;
548
Jens Axboe0989a022009-06-12 14:42:56 +0200549 q->backing_dev_info.ra_pages =
550 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
551 q->backing_dev_info.state = 0;
552 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200553 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100554 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200555
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700556 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100557 if (err)
558 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700559
Matthew Garrett31373d02010-04-06 14:25:14 +0200560 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
561 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700562 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100563 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700564 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100565 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800566#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800567 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800568#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100569 INIT_LIST_HEAD(&q->flush_queue[0]);
570 INIT_LIST_HEAD(&q->flush_queue[1]);
571 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500572 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500573
Jens Axboe8324aa92008-01-29 14:51:59 +0100574 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
Al Viro483f4af2006-03-18 18:34:37 -0500576 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700577 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500578
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500579 /*
580 * By default initialize queue_lock to internal lock and driver can
581 * override it later if need be.
582 */
583 q->queue_lock = &q->__queue_lock;
584
Tejun Heob82d4b12012-04-13 13:11:31 -0700585 /*
586 * A queue starts its life with bypass turned on to avoid
587 * unnecessary bypass on/off overhead and nasty surprises during
588 * init. The initial bypass will be finished at the end of
589 * blk_init_allocated_queue().
590 */
591 q->bypass_depth = 1;
592 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
593
Tejun Heo5efd6112012-03-05 13:15:12 -0800594 if (blkcg_init_queue(q))
Tejun Heof51b8022012-03-05 13:15:05 -0800595 goto fail_id;
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100598
599fail_id:
600 ida_simple_remove(&blk_queue_ida, q->id);
601fail_q:
602 kmem_cache_free(blk_requestq_cachep, q);
603 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
Christoph Lameter19460892005-06-23 00:08:19 -0700605EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607/**
608 * blk_init_queue - prepare a request queue for use with a block device
609 * @rfn: The function to be called to process requests that have been
610 * placed on the queue.
611 * @lock: Request queue spin lock
612 *
613 * Description:
614 * If a block device wishes to use the standard request handling procedures,
615 * which sorts requests and coalesces adjacent requests, then it must
616 * call blk_init_queue(). The function @rfn will be called when there
617 * are requests on the queue that need to be processed. If the device
618 * supports plugging, then @rfn may not be called immediately when requests
619 * are available on the queue, but may be called at some time later instead.
620 * Plugged queues are generally unplugged when a buffer belonging to one
621 * of the requests on the queue is needed, or due to memory pressure.
622 *
623 * @rfn is not required, or even expected, to remove all requests off the
624 * queue, but only as many as it can handle at a time. If it does leave
625 * requests on the queue, it is responsible for arranging that the requests
626 * get dealt with eventually.
627 *
628 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200629 * request queue; this lock will be taken also from interrupt context, so irq
630 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200632 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * it didn't succeed.
634 *
635 * Note:
636 * blk_init_queue() must be paired with a blk_cleanup_queue() call
637 * when the block device is deactivated (such as at module unload).
638 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700639
Jens Axboe165125e2007-07-24 09:28:11 +0200640struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Christoph Lameter19460892005-06-23 00:08:19 -0700642 return blk_init_queue_node(rfn, lock, -1);
643}
644EXPORT_SYMBOL(blk_init_queue);
645
Jens Axboe165125e2007-07-24 09:28:11 +0200646struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700647blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
648{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600649 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600651 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
652 if (!uninit_q)
653 return NULL;
654
Mike Snitzer51514122011-11-23 10:59:13 +0100655 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600656 if (!q)
657 blk_cleanup_queue(uninit_q);
658
659 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200660}
661EXPORT_SYMBOL(blk_init_queue_node);
662
663struct request_queue *
664blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
665 spinlock_t *lock)
666{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 if (!q)
668 return NULL;
669
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600670 if (blk_init_free_list(q))
Al Viro8669aaf2006-03-18 13:50:00 -0500671 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900675 q->unprep_rq_fn = NULL;
Jens Axboebc58ba92009-01-23 10:54:44 +0100676 q->queue_flags = QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500677
678 /* Override internal queue lock with supplied lock pointer */
679 if (lock)
680 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Jens Axboef3b144a2009-03-06 08:48:33 +0100682 /*
683 * This also sets hw/phys segments, boundary and size
684 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200685 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Alan Stern44ec9542007-02-20 11:01:57 -0500687 q->sg_reserved_size = INT_MAX;
688
Tejun Heob82d4b12012-04-13 13:11:31 -0700689 /* init elevator */
690 if (elevator_init(q, NULL))
691 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Tejun Heob82d4b12012-04-13 13:11:31 -0700693 blk_queue_congestion_threshold(q);
694
695 /* all done, end the initial bypass */
696 blk_queue_bypass_end(q);
697 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
Mike Snitzer51514122011-11-23 10:59:13 +0100699EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Tejun Heo09ac46c2011-12-14 00:33:38 +0100701bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Tejun Heo34f60552011-12-14 00:33:37 +0100703 if (likely(!blk_queue_dead(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100704 __blk_get_queue(q);
705 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707
Tejun Heo09ac46c2011-12-14 00:33:38 +0100708 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709}
Jens Axboed86e0e82011-05-27 07:44:43 +0200710EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Jens Axboe165125e2007-07-24 09:28:11 +0200712static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100714 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heocb98fc82005-10-28 08:29:39 +0200715 elv_put_request(q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100716 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100717 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100718 }
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 mempool_free(rq, q->rq.rq_pool);
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/*
724 * ioc_batching returns true if the ioc is a valid batching request and
725 * should be given priority access to a request.
726 */
Jens Axboe165125e2007-07-24 09:28:11 +0200727static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 if (!ioc)
730 return 0;
731
732 /*
733 * Make sure the process is able to allocate at least 1 request
734 * even if the batch times out, otherwise we could theoretically
735 * lose wakeups.
736 */
737 return ioc->nr_batch_requests == q->nr_batching ||
738 (ioc->nr_batch_requests > 0
739 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
740}
741
742/*
743 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
744 * will cause the process to be a "batcher" on all queues in the system. This
745 * is the behaviour we want though - once it gets a wakeup it should be given
746 * a nice run.
747 */
Jens Axboe165125e2007-07-24 09:28:11 +0200748static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 if (!ioc || ioc_batching(q, ioc))
751 return;
752
753 ioc->nr_batch_requests = q->nr_batching;
754 ioc->last_waited = jiffies;
755}
756
Jens Axboe1faa16d2009-04-06 14:48:01 +0200757static void __freed_request(struct request_queue *q, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
759 struct request_list *rl = &q->rq;
760
Jens Axboe1faa16d2009-04-06 14:48:01 +0200761 if (rl->count[sync] < queue_congestion_off_threshold(q))
762 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Jens Axboe1faa16d2009-04-06 14:48:01 +0200764 if (rl->count[sync] + 1 <= q->nr_requests) {
765 if (waitqueue_active(&rl->wait[sync]))
766 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Jens Axboe1faa16d2009-04-06 14:48:01 +0200768 blk_clear_queue_full(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 }
770}
771
772/*
773 * A request has just been released. Account for it, update the full and
774 * congestion status, wake up any waiters. Called under q->queue_lock.
775 */
Tejun Heo75eb6c32011-10-19 14:31:22 +0200776static void freed_request(struct request_queue *q, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
778 struct request_list *rl = &q->rq;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200779 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Jens Axboe1faa16d2009-04-06 14:48:01 +0200781 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200782 if (flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +0200783 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Jens Axboe1faa16d2009-04-06 14:48:01 +0200785 __freed_request(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Jens Axboe1faa16d2009-04-06 14:48:01 +0200787 if (unlikely(rl->starved[sync ^ 1]))
788 __freed_request(q, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100792 * Determine if elevator data should be initialized when allocating the
793 * request associated with @bio.
794 */
795static bool blk_rq_should_init_elevator(struct bio *bio)
796{
797 if (!bio)
798 return true;
799
800 /*
801 * Flush requests do not use the elevator so skip initialization.
802 * This allows a request to share the flush and elevator data.
803 */
804 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
805 return false;
806
807 return true;
808}
809
Tejun Heoda8303c2011-10-19 14:33:05 +0200810/**
Tejun Heo852c7882012-03-05 13:15:27 -0800811 * rq_ioc - determine io_context for request allocation
812 * @bio: request being allocated is for this bio (can be %NULL)
813 *
814 * Determine io_context to use for request allocation for @bio. May return
815 * %NULL if %current->io_context doesn't exist.
816 */
817static struct io_context *rq_ioc(struct bio *bio)
818{
819#ifdef CONFIG_BLK_CGROUP
820 if (bio && bio->bi_ioc)
821 return bio->bi_ioc;
822#endif
823 return current->io_context;
824}
825
826/**
Tejun Heoda8303c2011-10-19 14:33:05 +0200827 * get_request - get a free request
828 * @q: request_queue to allocate request from
829 * @rw_flags: RW and SYNC flags
830 * @bio: bio to allocate request for (can be %NULL)
831 * @gfp_mask: allocation mask
832 *
833 * Get a free request from @q. This function may fail under memory
834 * pressure or if @q is dead.
835 *
836 * Must be callled with @q->queue_lock held and,
837 * Returns %NULL on failure, with @q->queue_lock held.
838 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 */
Jens Axboe165125e2007-07-24 09:28:11 +0200840static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +0100841 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
Tejun Heob6792812012-03-05 13:15:23 -0800843 struct request *rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 struct request_list *rl = &q->rq;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100845 struct elevator_type *et;
Tejun Heof2dbd762011-12-14 00:33:40 +0100846 struct io_context *ioc;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100847 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200848 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heof2dbd762011-12-14 00:33:40 +0100849 bool retried = false;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200850 int may_queue;
Tejun Heof2dbd762011-12-14 00:33:40 +0100851retry:
Tejun Heof1f8cc92011-12-14 00:33:42 +0100852 et = q->elevator->type;
Tejun Heo852c7882012-03-05 13:15:27 -0800853 ioc = rq_ioc(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Tejun Heo34f60552011-12-14 00:33:37 +0100855 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200856 return NULL;
857
Jens Axboe7749a8d2006-12-13 13:02:26 +0100858 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100859 if (may_queue == ELV_MQUEUE_NO)
860 goto rq_starved;
861
Jens Axboe1faa16d2009-04-06 14:48:01 +0200862 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
863 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100864 /*
865 * We want ioc to record batching state. If it's
866 * not already there, creating a new one requires
867 * dropping queue_lock, which in turn requires
868 * retesting conditions to avoid queue hang.
869 */
870 if (!ioc && !retried) {
871 spin_unlock_irq(q->queue_lock);
Tejun Heo24acfc32012-03-05 13:15:24 -0800872 create_io_context(gfp_mask, q->node);
Tejun Heof2dbd762011-12-14 00:33:40 +0100873 spin_lock_irq(q->queue_lock);
874 retried = true;
875 goto retry;
876 }
877
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100878 /*
879 * The queue will fill after this allocation, so set
880 * it as full, and mark this process as "batching".
881 * This process will be allowed to complete a batch of
882 * requests, others will be blocked.
883 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200884 if (!blk_queue_full(q, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100885 ioc_set_batching(q, ioc);
Jens Axboe1faa16d2009-04-06 14:48:01 +0200886 blk_set_queue_full(q, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100887 } else {
888 if (may_queue != ELV_MQUEUE_MUST
889 && !ioc_batching(q, ioc)) {
890 /*
891 * The queue is full and the allocating
892 * process is not a "batcher", and not
893 * exempted by the IO scheduler
894 */
Tejun Heob6792812012-03-05 13:15:23 -0800895 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100896 }
897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
Jens Axboe1faa16d2009-04-06 14:48:01 +0200899 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 }
901
Jens Axboe082cf692005-06-28 16:35:11 +0200902 /*
903 * Only allow batching queuers to allocate up to 50% over the defined
904 * limit of requests, otherwise we could have thousands of requests
905 * allocated with any setting of ->nr_requests
906 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200907 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800908 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100909
Jens Axboe1faa16d2009-04-06 14:48:01 +0200910 rl->count[is_sync]++;
911 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200912
Tejun Heof1f8cc92011-12-14 00:33:42 +0100913 /*
914 * Decide whether the new request will be managed by elevator. If
915 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
916 * prevent the current elevator from being destroyed until the new
917 * request is freed. This guarantees icq's won't be destroyed and
918 * makes creating new ones safe.
919 *
920 * Also, lookup icq while holding queue_lock. If it doesn't exist,
921 * it will be created after releasing queue_lock.
922 */
Tejun Heod7325802012-03-05 13:14:58 -0800923 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200924 rw_flags |= REQ_ELVPRIV;
925 rl->elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100926 if (et->icq_cache && ioc)
927 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100928 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200929
Jens Axboef253b862010-10-24 22:06:02 +0200930 if (blk_queue_io_stat(q))
931 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 spin_unlock_irq(q->queue_lock);
933
Tejun Heo29e2b092012-04-19 16:29:21 -0700934 /* allocate and init request */
935 rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
936 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800937 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Tejun Heo29e2b092012-04-19 16:29:21 -0700939 blk_rq_init(q, rq);
940 rq->cmd_flags = rw_flags | REQ_ALLOCED;
941
Tejun Heoaaf7c682012-04-19 16:29:22 -0700942 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700943 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700944 if (unlikely(et->icq_cache && !icq)) {
945 create_io_context(gfp_mask, q->node);
946 ioc = rq_ioc(bio);
947 if (!ioc)
948 goto fail_elvpriv;
949
950 icq = ioc_create_icq(ioc, q, gfp_mask);
951 if (!icq)
952 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700953 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700954
955 rq->elv.icq = icq;
956 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
957 goto fail_elvpriv;
958
959 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -0700960 if (icq)
961 get_io_context(icq->ioc);
962 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700963out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100964 /*
965 * ioc may be NULL here, and ioc_batching will be false. That's
966 * OK, if the queue is under the request limit then requests need
967 * not count toward the nr_batch_requests limit. There will always
968 * be some limit enforced by BLK_BATCH_TIME.
969 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (ioc_batching(q, ioc))
971 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100972
Jens Axboe1faa16d2009-04-06 14:48:01 +0200973 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 return rq;
Tejun Heob6792812012-03-05 13:15:23 -0800975
Tejun Heoaaf7c682012-04-19 16:29:22 -0700976fail_elvpriv:
977 /*
978 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
979 * and may fail indefinitely under memory pressure and thus
980 * shouldn't stall IO. Treat this request as !elvpriv. This will
981 * disturb iosched and blkcg but weird is bettern than dead.
982 */
983 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
984 dev_name(q->backing_dev_info.dev));
985
986 rq->cmd_flags &= ~REQ_ELVPRIV;
987 rq->elv.icq = NULL;
988
989 spin_lock_irq(q->queue_lock);
990 rl->elvpriv--;
991 spin_unlock_irq(q->queue_lock);
992 goto out;
993
Tejun Heob6792812012-03-05 13:15:23 -0800994fail_alloc:
995 /*
996 * Allocation failed presumably due to memory. Undo anything we
997 * might have messed up.
998 *
999 * Allocating task should really be put onto the front of the wait
1000 * queue, but this is pretty rare.
1001 */
1002 spin_lock_irq(q->queue_lock);
1003 freed_request(q, rw_flags);
1004
1005 /*
1006 * in the very unlikely event that allocation failed and no
1007 * requests for this direction was pending, mark us starved so that
1008 * freeing of a request in the other direction will notice
1009 * us. another possible fix would be to split the rq mempool into
1010 * READ and WRITE
1011 */
1012rq_starved:
1013 if (unlikely(rl->count[is_sync] == 0))
1014 rl->starved[is_sync] = 1;
1015 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
Tejun Heoda8303c2011-10-19 14:33:05 +02001018/**
1019 * get_request_wait - get a free request with retry
1020 * @q: request_queue to allocate request from
1021 * @rw_flags: RW and SYNC flags
1022 * @bio: bio to allocate request for (can be %NULL)
Nick Piggind6344532005-06-28 20:45:14 -07001023 *
Tejun Heoda8303c2011-10-19 14:33:05 +02001024 * Get a free request from @q. This function keeps retrying under memory
1025 * pressure and fails iff @q is dead.
1026 *
1027 * Must be callled with @q->queue_lock held and,
1028 * Returns %NULL on failure, with @q->queue_lock held.
1029 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 */
Jens Axboe165125e2007-07-24 09:28:11 +02001031static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02001032 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001034 const bool is_sync = rw_is_sync(rw_flags) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 struct request *rq;
1036
Jens Axboe7749a8d2006-12-13 13:02:26 +01001037 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07001038 while (!rq) {
1039 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 struct request_list *rl = &q->rq;
1041
Tejun Heo34f60552011-12-14 00:33:37 +01001042 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +02001043 return NULL;
1044
Jens Axboe1faa16d2009-04-06 14:48:01 +02001045 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 TASK_UNINTERRUPTIBLE);
1047
Jens Axboe1faa16d2009-04-06 14:48:01 +02001048 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +02001050 spin_unlock_irq(q->queue_lock);
1051 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +02001053 /*
1054 * After sleeping, we become a "batching" process and
1055 * will be able to allocate at least one request, and
1056 * up to a big batch of them for a small period time.
1057 * See ioc_batching, ioc_set_batching
1058 */
Tejun Heo24acfc32012-03-05 13:15:24 -08001059 create_io_context(GFP_NOIO, q->node);
Tejun Heof2dbd762011-12-14 00:33:40 +01001060 ioc_set_batching(q, current->io_context);
Jens Axboe2056a782006-03-23 20:00:26 +01001061
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +02001062 spin_lock_irq(q->queue_lock);
Jens Axboe1faa16d2009-04-06 14:48:01 +02001063 finish_wait(&rl->wait[is_sync], &wait);
Zhang, Yanmin05caf8d2008-05-22 15:13:29 +02001064
1065 rq = get_request(q, rw_flags, bio, GFP_NOIO);
1066 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 return rq;
1069}
1070
Jens Axboe165125e2007-07-24 09:28:11 +02001071struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072{
1073 struct request *rq;
1074
1075 BUG_ON(rw != READ && rw != WRITE);
1076
Nick Piggind6344532005-06-28 20:45:14 -07001077 spin_lock_irq(q->queue_lock);
Tejun Heoda8303c2011-10-19 14:33:05 +02001078 if (gfp_mask & __GFP_WAIT)
Jens Axboe22e2c502005-06-27 10:55:12 +02001079 rq = get_request_wait(q, rw, NULL);
Tejun Heoda8303c2011-10-19 14:33:05 +02001080 else
Jens Axboe22e2c502005-06-27 10:55:12 +02001081 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001082 if (!rq)
1083 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001084 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 return rq;
1087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088EXPORT_SYMBOL(blk_get_request);
1089
1090/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001091 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001092 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001093 * @bio: The bio describing the memory mappings that will be submitted for IO.
1094 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001095 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001096 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001097 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1098 * type commands. Where the struct request needs to be farther initialized by
1099 * the caller. It is passed a &struct bio, which describes the memory info of
1100 * the I/O transfer.
1101 *
1102 * The caller of blk_make_request must make sure that bi_io_vec
1103 * are set to describe the memory buffers. That bio_data_dir() will return
1104 * the needed direction of the request. (And all bio's in the passed bio-chain
1105 * are properly set accordingly)
1106 *
1107 * If called under none-sleepable conditions, mapped bio buffers must not
1108 * need bouncing, by calling the appropriate masked or flagged allocator,
1109 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1110 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001111 *
1112 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1113 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1114 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1115 * completion of a bio that hasn't been submitted yet, thus resulting in a
1116 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1117 * of bio_alloc(), as that avoids the mempool deadlock.
1118 * If possible a big IO should be split into smaller parts when allocation
1119 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001120 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001121struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1122 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001123{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001124 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1125
1126 if (unlikely(!rq))
1127 return ERR_PTR(-ENOMEM);
1128
1129 for_each_bio(bio) {
1130 struct bio *bounce_bio = bio;
1131 int ret;
1132
1133 blk_queue_bounce(q, &bounce_bio);
1134 ret = blk_rq_append_bio(q, rq, bounce_bio);
1135 if (unlikely(ret)) {
1136 blk_put_request(rq);
1137 return ERR_PTR(ret);
1138 }
1139 }
1140
1141 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001142}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001143EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001144
1145/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 * blk_requeue_request - put a request back on queue
1147 * @q: request queue where request should be inserted
1148 * @rq: request to be inserted
1149 *
1150 * Description:
1151 * Drivers often keep queueing requests until the hardware cannot accept
1152 * more, when that condition happens we need to put the request back
1153 * on the queue. Must be called with queue lock held.
1154 */
Jens Axboe165125e2007-07-24 09:28:11 +02001155void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001157 blk_delete_timer(rq);
1158 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001159 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (blk_rq_tagged(rq))
1162 blk_queue_end_tag(q, rq);
1163
James Bottomleyba396a62009-05-27 14:17:08 +02001164 BUG_ON(blk_queued_rq(rq));
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 elv_requeue_request(q, rq);
1167}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168EXPORT_SYMBOL(blk_requeue_request);
1169
Jens Axboe73c10102011-03-08 13:19:51 +01001170static void add_acct_request(struct request_queue *q, struct request *rq,
1171 int where)
1172{
1173 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001174 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001175}
1176
Tejun Heo074a7ac2008-08-25 19:56:14 +09001177static void part_round_stats_single(int cpu, struct hd_struct *part,
1178 unsigned long now)
1179{
1180 if (now == part->stamp)
1181 return;
1182
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001183 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001184 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001185 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001186 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1187 }
1188 part->stamp = now;
1189}
1190
1191/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001192 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1193 * @cpu: cpu number for stats access
1194 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 *
1196 * The average IO queue length and utilisation statistics are maintained
1197 * by observing the current state of the queue length and the amount of
1198 * time it has been in this state for.
1199 *
1200 * Normally, that accounting is done on IO completion, but that can result
1201 * in more than a second's worth of IO being accounted for within any one
1202 * second, leading to >100% utilisation. To deal with that, we call this
1203 * function to do a round-off before returning the results when reading
1204 * /proc/diskstats. This accounts immediately for all queue usage up to
1205 * the current jiffies and restarts the counters again.
1206 */
Tejun Heoc9959052008-08-25 19:47:21 +09001207void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001208{
1209 unsigned long now = jiffies;
1210
Tejun Heo074a7ac2008-08-25 19:56:14 +09001211 if (part->partno)
1212 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1213 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001214}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001215EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/*
1218 * queue lock must be held
1219 */
Jens Axboe165125e2007-07-24 09:28:11 +02001220void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 if (unlikely(!q))
1223 return;
1224 if (unlikely(--req->ref_count))
1225 return;
1226
Tejun Heo8922e162005-10-20 16:23:44 +02001227 elv_completed_request(q, req);
1228
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001229 /* this is a bio leak */
1230 WARN_ON(req->bio != NULL);
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 /*
1233 * Request may not have originated from ll_rw_blk. if not,
1234 * it didn't come out of our reserved rq pools
1235 */
Jens Axboe49171e52006-08-10 08:59:11 +02001236 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001237 unsigned int flags = req->cmd_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001240 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 blk_free_request(q, req);
Tejun Heo75eb6c32011-10-19 14:31:22 +02001243 freed_request(q, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001246EXPORT_SYMBOL_GPL(__blk_put_request);
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248void blk_put_request(struct request *req)
1249{
Tejun Heo8922e162005-10-20 16:23:44 +02001250 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001251 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001253 spin_lock_irqsave(q->queue_lock, flags);
1254 __blk_put_request(q, req);
1255 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257EXPORT_SYMBOL(blk_put_request);
1258
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001259/**
1260 * blk_add_request_payload - add a payload to a request
1261 * @rq: request to update
1262 * @page: page backing the payload
1263 * @len: length of the payload.
1264 *
1265 * This allows to later add a payload to an already submitted request by
1266 * a block driver. The driver needs to take care of freeing the payload
1267 * itself.
1268 *
1269 * Note that this is a quite horrible hack and nothing but handling of
1270 * discard requests should ever use it.
1271 */
1272void blk_add_request_payload(struct request *rq, struct page *page,
1273 unsigned int len)
1274{
1275 struct bio *bio = rq->bio;
1276
1277 bio->bi_io_vec->bv_page = page;
1278 bio->bi_io_vec->bv_offset = 0;
1279 bio->bi_io_vec->bv_len = len;
1280
1281 bio->bi_size = len;
1282 bio->bi_vcnt = 1;
1283 bio->bi_phys_segments = 1;
1284
1285 rq->__data_len = rq->resid_len = len;
1286 rq->nr_phys_segments = 1;
1287 rq->buffer = bio_data(bio);
1288}
1289EXPORT_SYMBOL_GPL(blk_add_request_payload);
1290
Jens Axboe73c10102011-03-08 13:19:51 +01001291static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1292 struct bio *bio)
1293{
1294 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1295
Jens Axboe73c10102011-03-08 13:19:51 +01001296 if (!ll_back_merge_fn(q, req, bio))
1297 return false;
1298
1299 trace_block_bio_backmerge(q, bio);
1300
1301 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1302 blk_rq_set_mixed_merge(req);
1303
1304 req->biotail->bi_next = bio;
1305 req->biotail = bio;
1306 req->__data_len += bio->bi_size;
1307 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1308
1309 drive_stat_acct(req, 0);
1310 return true;
1311}
1312
1313static bool bio_attempt_front_merge(struct request_queue *q,
1314 struct request *req, struct bio *bio)
1315{
1316 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001317
Jens Axboe73c10102011-03-08 13:19:51 +01001318 if (!ll_front_merge_fn(q, req, bio))
1319 return false;
1320
1321 trace_block_bio_frontmerge(q, bio);
1322
1323 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1324 blk_rq_set_mixed_merge(req);
1325
Jens Axboe73c10102011-03-08 13:19:51 +01001326 bio->bi_next = req->bio;
1327 req->bio = bio;
1328
1329 /*
1330 * may not be valid. if the low level driver said
1331 * it didn't need a bounce buffer then it better
1332 * not touch req->buffer either...
1333 */
1334 req->buffer = bio_data(bio);
1335 req->__sector = bio->bi_sector;
1336 req->__data_len += bio->bi_size;
1337 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1338
1339 drive_stat_acct(req, 0);
1340 return true;
1341}
1342
Tejun Heobd87b582011-10-19 14:33:08 +02001343/**
1344 * attempt_plug_merge - try to merge with %current's plugged list
1345 * @q: request_queue new bio is being queued at
1346 * @bio: new bio being queued
1347 * @request_count: out parameter for number of traversed plugged requests
1348 *
1349 * Determine whether @bio being queued on @q can be merged with a request
1350 * on %current's plugged list. Returns %true if merge was successful,
1351 * otherwise %false.
1352 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001353 * Plugging coalesces IOs from the same issuer for the same purpose without
1354 * going through @q->queue_lock. As such it's more of an issuing mechanism
1355 * than scheduling, and the request, while may have elvpriv data, is not
1356 * added on the elevator at this point. In addition, we don't have
1357 * reliable access to the elevator outside queue lock. Only check basic
1358 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001359 */
Tejun Heobd87b582011-10-19 14:33:08 +02001360static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1361 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001362{
1363 struct blk_plug *plug;
1364 struct request *rq;
1365 bool ret = false;
1366
Tejun Heobd87b582011-10-19 14:33:08 +02001367 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001368 if (!plug)
1369 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001370 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001371
1372 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1373 int el_ret;
1374
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001375 if (rq->q == q)
1376 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001377
Tejun Heo07c2bd32012-02-08 09:19:42 +01001378 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001379 continue;
1380
Tejun Heo050c8ea2012-02-08 09:19:38 +01001381 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001382 if (el_ret == ELEVATOR_BACK_MERGE) {
1383 ret = bio_attempt_back_merge(q, rq, bio);
1384 if (ret)
1385 break;
1386 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1387 ret = bio_attempt_front_merge(q, rq, bio);
1388 if (ret)
1389 break;
1390 }
1391 }
1392out:
1393 return ret;
1394}
1395
Jens Axboe86db1e22008-01-29 14:53:40 +01001396void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001397{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001398 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001399
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001400 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1401 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001402 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001403
Tejun Heo52d9e672006-01-06 09:49:58 +01001404 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001405 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001406 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001407 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001408}
1409
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001410void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001412 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001413 struct blk_plug *plug;
1414 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1415 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001416 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 /*
1419 * low level driver can indicate that it wants pages above a
1420 * certain limit bounced to low memory (ie for highmem, or even
1421 * ISA dma in theory)
1422 */
1423 blk_queue_bounce(q, &bio);
1424
Tejun Heo4fed9472010-09-03 11:56:17 +02001425 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001426 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001427 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001428 goto get_rq;
1429 }
1430
Jens Axboe73c10102011-03-08 13:19:51 +01001431 /*
1432 * Check if we can merge with the plugged list before grabbing
1433 * any locks.
1434 */
Tejun Heobd87b582011-10-19 14:33:08 +02001435 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001436 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001437
1438 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001441 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001442 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001443 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001444 if (!attempt_back_merge(q, req))
1445 elv_merged_request(q, req, el_ret);
1446 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001447 }
Jens Axboe73c10102011-03-08 13:19:51 +01001448 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001449 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001450 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001451 if (!attempt_front_merge(q, req))
1452 elv_merged_request(q, req, el_ret);
1453 goto out_unlock;
1454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001458 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001459 * This sync check and mask will be re-done in init_request_from_bio(),
1460 * but we need to set it earlier to expose the sync flag to the
1461 * rq allocator and io schedulers.
1462 */
1463 rw_flags = bio_data_dir(bio);
1464 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001465 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001466
1467 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001468 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001469 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001470 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001471 req = get_request_wait(q, rw_flags, bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001472 if (unlikely(!req)) {
1473 bio_endio(bio, -ENODEV); /* @q is dead */
1474 goto out_unlock;
1475 }
Nick Piggind6344532005-06-28 20:45:14 -07001476
Nick Piggin450991b2005-06-28 20:45:13 -07001477 /*
1478 * After dropping the lock and possibly sleeping here, our request
1479 * may now be mergeable after it had proven unmergeable (above).
1480 * We don't worry about that case for efficiency. It won't happen
1481 * often, and the elevators are able to handle it.
1482 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001483 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Tao Ma9562ad92011-10-24 16:11:30 +02001485 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001486 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001487
Jens Axboe73c10102011-03-08 13:19:51 +01001488 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001489 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001490 /*
1491 * If this is the first request added after a plug, fire
1492 * of a plug trace. If others have been added before, check
1493 * if we have multiple devices in this plug. If so, make a
1494 * note to sort the list before dispatch.
1495 */
1496 if (list_empty(&plug->list))
1497 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001498 else {
1499 if (!plug->should_sort) {
1500 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001501
Shaohua Li3540d5e2011-11-16 09:21:50 +01001502 __rq = list_entry_rq(plug->list.prev);
1503 if (__rq->q != q)
1504 plug->should_sort = 1;
1505 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001506 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001507 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001508 trace_block_plug(q);
1509 }
Jens Axboe73c10102011-03-08 13:19:51 +01001510 }
Shaohua Lia6327162011-08-24 16:04:32 +02001511 list_add_tail(&req->queuelist, &plug->list);
1512 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001513 } else {
1514 spin_lock_irq(q->queue_lock);
1515 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001516 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001517out_unlock:
1518 spin_unlock_irq(q->queue_lock);
1519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520}
Jens Axboec20e8de2011-09-12 12:03:37 +02001521EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523/*
1524 * If bio->bi_dev is a partition, remap the location
1525 */
1526static inline void blk_partition_remap(struct bio *bio)
1527{
1528 struct block_device *bdev = bio->bi_bdev;
1529
Jens Axboebf2de6f2007-09-27 13:01:25 +02001530 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 bio->bi_sector += p->start_sect;
1534 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001535
Mike Snitzerd07335e2010-11-16 12:52:38 +01001536 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1537 bdev->bd_dev,
1538 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 }
1540}
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542static void handle_bad_sector(struct bio *bio)
1543{
1544 char b[BDEVNAME_SIZE];
1545
1546 printk(KERN_INFO "attempt to access beyond end of device\n");
1547 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1548 bdevname(bio->bi_bdev, b),
1549 bio->bi_rw,
1550 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001551 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553 set_bit(BIO_EOF, &bio->bi_flags);
1554}
1555
Akinobu Mitac17bb492006-12-08 02:39:46 -08001556#ifdef CONFIG_FAIL_MAKE_REQUEST
1557
1558static DECLARE_FAULT_ATTR(fail_make_request);
1559
1560static int __init setup_fail_make_request(char *str)
1561{
1562 return setup_fault_attr(&fail_make_request, str);
1563}
1564__setup("fail_make_request=", setup_fail_make_request);
1565
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001566static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001567{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001568 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001569}
1570
1571static int __init fail_make_request_debugfs(void)
1572{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001573 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1574 NULL, &fail_make_request);
1575
1576 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001577}
1578
1579late_initcall(fail_make_request_debugfs);
1580
1581#else /* CONFIG_FAIL_MAKE_REQUEST */
1582
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001583static inline bool should_fail_request(struct hd_struct *part,
1584 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001585{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001586 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001587}
1588
1589#endif /* CONFIG_FAIL_MAKE_REQUEST */
1590
Jens Axboec07e2b42007-07-18 13:27:58 +02001591/*
1592 * Check whether this bio extends beyond the end of the device.
1593 */
1594static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1595{
1596 sector_t maxsector;
1597
1598 if (!nr_sectors)
1599 return 0;
1600
1601 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001602 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001603 if (maxsector) {
1604 sector_t sector = bio->bi_sector;
1605
1606 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1607 /*
1608 * This may well happen - the kernel calls bread()
1609 * without checking the size of the device, e.g., when
1610 * mounting a device.
1611 */
1612 handle_bad_sector(bio);
1613 return 1;
1614 }
1615 }
1616
1617 return 0;
1618}
1619
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001620static noinline_for_stack bool
1621generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622{
Jens Axboe165125e2007-07-24 09:28:11 +02001623 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001624 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001625 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001626 char b[BDEVNAME_SIZE];
1627 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Jens Axboec07e2b42007-07-18 13:27:58 +02001631 if (bio_check_eod(bio, nr_sectors))
1632 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001634 q = bdev_get_queue(bio->bi_bdev);
1635 if (unlikely(!q)) {
1636 printk(KERN_ERR
1637 "generic_make_request: Trying to access "
1638 "nonexistent block-device %s (%Lu)\n",
1639 bdevname(bio->bi_bdev, b),
1640 (long long) bio->bi_sector);
1641 goto end_io;
1642 }
1643
1644 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1645 nr_sectors > queue_max_hw_sectors(q))) {
1646 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1647 bdevname(bio->bi_bdev, b),
1648 bio_sectors(bio),
1649 queue_max_hw_sectors(q));
1650 goto end_io;
1651 }
1652
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001653 part = bio->bi_bdev->bd_part;
1654 if (should_fail_request(part, bio->bi_size) ||
1655 should_fail_request(&part_to_disk(part)->part0,
1656 bio->bi_size))
1657 goto end_io;
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001660 * If this device has partitions, remap block n
1661 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001663 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001665 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1666 goto end_io;
1667
1668 if (bio_check_eod(bio, nr_sectors))
1669 goto end_io;
1670
1671 /*
1672 * Filter flush bio's early so that make_request based
1673 * drivers without flush support don't have to worry
1674 * about them.
1675 */
1676 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1677 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1678 if (!nr_sectors) {
1679 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001680 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001682 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001684 if ((bio->bi_rw & REQ_DISCARD) &&
1685 (!blk_queue_discard(q) ||
1686 ((bio->bi_rw & REQ_SECURE) &&
1687 !blk_queue_secdiscard(q)))) {
1688 err = -EOPNOTSUPP;
1689 goto end_io;
1690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
Tejun Heobc16a4f2011-10-19 14:33:01 +02001692 if (blk_throtl_bio(q, bio))
1693 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001694
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001695 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001696 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001697
1698end_io:
1699 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001700 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701}
1702
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001703/**
1704 * generic_make_request - hand a buffer to its device driver for I/O
1705 * @bio: The bio describing the location in memory and on the device.
1706 *
1707 * generic_make_request() is used to make I/O requests of block
1708 * devices. It is passed a &struct bio, which describes the I/O that needs
1709 * to be done.
1710 *
1711 * generic_make_request() does not return any status. The
1712 * success/failure status of the request, along with notification of
1713 * completion, is delivered asynchronously through the bio->bi_end_io
1714 * function described (one day) else where.
1715 *
1716 * The caller of generic_make_request must make sure that bi_io_vec
1717 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1718 * set to describe the device address, and the
1719 * bi_end_io and optionally bi_private are set to describe how
1720 * completion notification should be signaled.
1721 *
1722 * generic_make_request and the drivers it calls may use bi_next if this
1723 * bio happens to be merged with someone else, and may resubmit the bio to
1724 * a lower device by calling into generic_make_request recursively, which
1725 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001726 */
1727void generic_make_request(struct bio *bio)
1728{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001729 struct bio_list bio_list_on_stack;
1730
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001731 if (!generic_make_request_checks(bio))
1732 return;
1733
1734 /*
1735 * We only want one ->make_request_fn to be active at a time, else
1736 * stack usage with stacked devices could be a problem. So use
1737 * current->bio_list to keep a list of requests submited by a
1738 * make_request_fn function. current->bio_list is also used as a
1739 * flag to say if generic_make_request is currently active in this
1740 * task or not. If it is NULL, then no make_request is active. If
1741 * it is non-NULL, then a make_request is active, and new requests
1742 * should be added at the tail
1743 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001744 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001745 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001746 return;
1747 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001748
Neil Brownd89d8792007-05-01 09:53:42 +02001749 /* following loop may be a bit non-obvious, and so deserves some
1750 * explanation.
1751 * Before entering the loop, bio->bi_next is NULL (as all callers
1752 * ensure that) so we have a list with a single bio.
1753 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001754 * we assign bio_list to a pointer to the bio_list_on_stack,
1755 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001756 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001757 * through a recursive call to generic_make_request. If it
1758 * did, we find a non-NULL value in bio_list and re-enter the loop
1759 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001760 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001761 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001762 */
1763 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001764 bio_list_init(&bio_list_on_stack);
1765 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001766 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001767 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1768
1769 q->make_request_fn(q, bio);
1770
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001771 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001772 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001773 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001774}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775EXPORT_SYMBOL(generic_make_request);
1776
1777/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001778 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1780 * @bio: The &struct bio which describes the I/O
1781 *
1782 * submit_bio() is very similar in purpose to generic_make_request(), and
1783 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001784 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 *
1786 */
1787void submit_bio(int rw, struct bio *bio)
1788{
1789 int count = bio_sectors(bio);
1790
Jens Axboe22e2c502005-06-27 10:55:12 +02001791 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Jens Axboebf2de6f2007-09-27 13:01:25 +02001793 /*
1794 * If it's a regular read/write or a barrier with data attached,
1795 * go through the normal accounting stuff before submission.
1796 */
Jens Axboe3ffb52e2010-06-29 13:33:38 +02001797 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001798 if (rw & WRITE) {
1799 count_vm_events(PGPGOUT, count);
1800 } else {
1801 task_io_account_read(bio->bi_size);
1802 count_vm_events(PGPGIN, count);
1803 }
1804
1805 if (unlikely(block_dump)) {
1806 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001807 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001808 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001809 (rw & WRITE) ? "WRITE" : "READ",
1810 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001811 bdevname(bio->bi_bdev, b),
1812 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 }
1815
1816 generic_make_request(bio);
1817}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818EXPORT_SYMBOL(submit_bio);
1819
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001820/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001821 * blk_rq_check_limits - Helper function to check a request for the queue limit
1822 * @q: the queue
1823 * @rq: the request being checked
1824 *
1825 * Description:
1826 * @rq may have been made based on weaker limitations of upper-level queues
1827 * in request stacking drivers, and it may violate the limitation of @q.
1828 * Since the block layer and the underlying device driver trust @rq
1829 * after it is inserted to @q, it should be checked against @q before
1830 * the insertion using this generic function.
1831 *
1832 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001833 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001834 * Request stacking drivers like request-based dm may change the queue
1835 * limits while requests are in the queue (e.g. dm's table swapping).
1836 * Such request stacking drivers should check those requests agaist
1837 * the new queue limits again when they dispatch those requests,
1838 * although such checkings are also done against the old queue limits
1839 * when submitting requests.
1840 */
1841int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1842{
ike Snitzer33839772010-08-08 12:11:33 -04001843 if (rq->cmd_flags & REQ_DISCARD)
1844 return 0;
1845
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001846 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1847 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001848 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1849 return -EIO;
1850 }
1851
1852 /*
1853 * queue's settings related to segment counting like q->bounce_pfn
1854 * may differ from that of other stacking queues.
1855 * Recalculate it to check the request correctly on this queue's
1856 * limitation.
1857 */
1858 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001859 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001860 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1861 return -EIO;
1862 }
1863
1864 return 0;
1865}
1866EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1867
1868/**
1869 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1870 * @q: the queue to submit the request
1871 * @rq: the request being queued
1872 */
1873int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1874{
1875 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001876 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001877
1878 if (blk_rq_check_limits(q, rq))
1879 return -EIO;
1880
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001881 if (rq->rq_disk &&
1882 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001883 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001884
1885 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo8ba61432011-12-14 00:33:37 +01001886 if (unlikely(blk_queue_dead(q))) {
1887 spin_unlock_irqrestore(q->queue_lock, flags);
1888 return -ENODEV;
1889 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001890
1891 /*
1892 * Submitting request must be dequeued before calling this function
1893 * because it will be linked to another request_queue
1894 */
1895 BUG_ON(blk_queued_rq(rq));
1896
Jeff Moyer4853aba2011-08-15 21:37:25 +02001897 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1898 where = ELEVATOR_INSERT_FLUSH;
1899
1900 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001901 if (where == ELEVATOR_INSERT_FLUSH)
1902 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001903 spin_unlock_irqrestore(q->queue_lock, flags);
1904
1905 return 0;
1906}
1907EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1908
Tejun Heo80a761f2009-07-03 17:48:17 +09001909/**
1910 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1911 * @rq: request to examine
1912 *
1913 * Description:
1914 * A request could be merge of IOs which require different failure
1915 * handling. This function determines the number of bytes which
1916 * can be failed from the beginning of the request without
1917 * crossing into area which need to be retried further.
1918 *
1919 * Return:
1920 * The number of bytes to fail.
1921 *
1922 * Context:
1923 * queue_lock must be held.
1924 */
1925unsigned int blk_rq_err_bytes(const struct request *rq)
1926{
1927 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1928 unsigned int bytes = 0;
1929 struct bio *bio;
1930
1931 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1932 return blk_rq_bytes(rq);
1933
1934 /*
1935 * Currently the only 'mixing' which can happen is between
1936 * different fastfail types. We can safely fail portions
1937 * which have all the failfast bits that the first one has -
1938 * the ones which are at least as eager to fail as the first
1939 * one.
1940 */
1941 for (bio = rq->bio; bio; bio = bio->bi_next) {
1942 if ((bio->bi_rw & ff) != ff)
1943 break;
1944 bytes += bio->bi_size;
1945 }
1946
1947 /* this could lead to infinite loop */
1948 BUG_ON(blk_rq_bytes(rq) && !bytes);
1949 return bytes;
1950}
1951EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1952
Jens Axboebc58ba92009-01-23 10:54:44 +01001953static void blk_account_io_completion(struct request *req, unsigned int bytes)
1954{
Jens Axboec2553b52009-04-24 08:10:11 +02001955 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001956 const int rw = rq_data_dir(req);
1957 struct hd_struct *part;
1958 int cpu;
1959
1960 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001961 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001962 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1963 part_stat_unlock();
1964 }
1965}
1966
1967static void blk_account_io_done(struct request *req)
1968{
Jens Axboebc58ba92009-01-23 10:54:44 +01001969 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001970 * Account IO completion. flush_rq isn't accounted as a
1971 * normal IO on queueing nor completion. Accounting the
1972 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001973 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01001974 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001975 unsigned long duration = jiffies - req->start_time;
1976 const int rw = rq_data_dir(req);
1977 struct hd_struct *part;
1978 int cpu;
1979
1980 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001981 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001982
1983 part_stat_inc(cpu, part, ios[rw]);
1984 part_stat_add(cpu, part, ticks[rw], duration);
1985 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001986 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01001987
Jens Axboe6c23a962011-01-07 08:43:37 +01001988 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001989 part_stat_unlock();
1990 }
1991}
1992
Tejun Heo53a08802008-12-03 12:41:26 +01001993/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09001994 * blk_peek_request - peek at the top of a request queue
1995 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001996 *
1997 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09001998 * Return the request at the top of @q. The returned request
1999 * should be started using blk_start_request() before LLD starts
2000 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002001 *
2002 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002003 * Pointer to the request at the top of @q if available. Null
2004 * otherwise.
2005 *
2006 * Context:
2007 * queue_lock must be held.
2008 */
2009struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002010{
2011 struct request *rq;
2012 int ret;
2013
2014 while ((rq = __elv_next_request(q)) != NULL) {
2015 if (!(rq->cmd_flags & REQ_STARTED)) {
2016 /*
2017 * This is the first time the device driver
2018 * sees this request (possibly after
2019 * requeueing). Notify IO scheduler.
2020 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002021 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002022 elv_activate_rq(q, rq);
2023
2024 /*
2025 * just mark as started even if we don't start
2026 * it, a request that has been delayed should
2027 * not be passed by new incoming requests
2028 */
2029 rq->cmd_flags |= REQ_STARTED;
2030 trace_block_rq_issue(q, rq);
2031 }
2032
2033 if (!q->boundary_rq || q->boundary_rq == rq) {
2034 q->end_sector = rq_end_sector(rq);
2035 q->boundary_rq = NULL;
2036 }
2037
2038 if (rq->cmd_flags & REQ_DONTPREP)
2039 break;
2040
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002041 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002042 /*
2043 * make sure space for the drain appears we
2044 * know we can do this because max_hw_segments
2045 * has been adjusted to be one fewer than the
2046 * device can handle
2047 */
2048 rq->nr_phys_segments++;
2049 }
2050
2051 if (!q->prep_rq_fn)
2052 break;
2053
2054 ret = q->prep_rq_fn(q, rq);
2055 if (ret == BLKPREP_OK) {
2056 break;
2057 } else if (ret == BLKPREP_DEFER) {
2058 /*
2059 * the request may have been (partially) prepped.
2060 * we need to keep this request in the front to
2061 * avoid resource deadlock. REQ_STARTED will
2062 * prevent other fs requests from passing this one.
2063 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002064 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002065 !(rq->cmd_flags & REQ_DONTPREP)) {
2066 /*
2067 * remove the space for the drain we added
2068 * so that we don't add it again
2069 */
2070 --rq->nr_phys_segments;
2071 }
2072
2073 rq = NULL;
2074 break;
2075 } else if (ret == BLKPREP_KILL) {
2076 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002077 /*
2078 * Mark this request as started so we don't trigger
2079 * any debug logic in the end I/O path.
2080 */
2081 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002082 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002083 } else {
2084 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2085 break;
2086 }
2087 }
2088
2089 return rq;
2090}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002091EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002092
Tejun Heo9934c8c2009-05-08 11:54:16 +09002093void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002094{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002095 struct request_queue *q = rq->q;
2096
Tejun Heo158dbda2009-04-23 11:05:18 +09002097 BUG_ON(list_empty(&rq->queuelist));
2098 BUG_ON(ELV_ON_HASH(rq));
2099
2100 list_del_init(&rq->queuelist);
2101
2102 /*
2103 * the time frame between a request being removed from the lists
2104 * and to it is freed is accounted as io that is in progress at
2105 * the driver side.
2106 */
Divyesh Shah91952912010-04-01 15:01:41 -07002107 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002108 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002109 set_io_start_time_ns(rq);
2110 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002111}
2112
Tejun Heo5efccd12009-04-23 11:05:18 +09002113/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002114 * blk_start_request - start request processing on the driver
2115 * @req: request to dequeue
2116 *
2117 * Description:
2118 * Dequeue @req and start timeout timer on it. This hands off the
2119 * request to the driver.
2120 *
2121 * Block internal functions which don't want to start timer should
2122 * call blk_dequeue_request().
2123 *
2124 * Context:
2125 * queue_lock must be held.
2126 */
2127void blk_start_request(struct request *req)
2128{
2129 blk_dequeue_request(req);
2130
2131 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002132 * We are now handing the request to the hardware, initialize
2133 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002134 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002135 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002136 if (unlikely(blk_bidi_rq(req)))
2137 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2138
Tejun Heo9934c8c2009-05-08 11:54:16 +09002139 blk_add_timer(req);
2140}
2141EXPORT_SYMBOL(blk_start_request);
2142
2143/**
2144 * blk_fetch_request - fetch a request from a request queue
2145 * @q: request queue to fetch a request from
2146 *
2147 * Description:
2148 * Return the request at the top of @q. The request is started on
2149 * return and LLD can start processing it immediately.
2150 *
2151 * Return:
2152 * Pointer to the request at the top of @q if available. Null
2153 * otherwise.
2154 *
2155 * Context:
2156 * queue_lock must be held.
2157 */
2158struct request *blk_fetch_request(struct request_queue *q)
2159{
2160 struct request *rq;
2161
2162 rq = blk_peek_request(q);
2163 if (rq)
2164 blk_start_request(rq);
2165 return rq;
2166}
2167EXPORT_SYMBOL(blk_fetch_request);
2168
2169/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002170 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002171 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002172 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002173 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002174 *
2175 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002176 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2177 * the request structure even if @req doesn't have leftover.
2178 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002179 *
2180 * This special helper function is only for request stacking drivers
2181 * (e.g. request-based dm) so that they can handle partial completion.
2182 * Actual device drivers should use blk_end_request instead.
2183 *
2184 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2185 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002186 *
2187 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002188 * %false - this request doesn't have any more data
2189 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002190 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002191bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002193 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 struct bio *bio;
2195
Tejun Heo2e60e022009-04-23 11:05:18 +09002196 if (!req->bio)
2197 return false;
2198
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002199 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002200
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002202 * For fs requests, rq is just carrier of independent bio's
2203 * and each partial completion should be handled separately.
2204 * Reset per-request error on each partial completion.
2205 *
2206 * TODO: tj: This is too subtle. It would be better to let
2207 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002209 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 req->errors = 0;
2211
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002212 if (error && req->cmd_type == REQ_TYPE_FS &&
2213 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002214 char *error_type;
2215
2216 switch (error) {
2217 case -ENOLINK:
2218 error_type = "recoverable transport";
2219 break;
2220 case -EREMOTEIO:
2221 error_type = "critical target";
2222 break;
2223 case -EBADE:
2224 error_type = "critical nexus";
2225 break;
2226 case -EIO:
2227 default:
2228 error_type = "I/O";
2229 break;
2230 }
2231 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2232 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2233 (unsigned long long)blk_rq_pos(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 }
2235
Jens Axboebc58ba92009-01-23 10:54:44 +01002236 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 total_bytes = bio_nbytes = 0;
2239 while ((bio = req->bio) != NULL) {
2240 int nbytes;
2241
2242 if (nr_bytes >= bio->bi_size) {
2243 req->bio = bio->bi_next;
2244 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002245 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 next_idx = 0;
2247 bio_nbytes = 0;
2248 } else {
2249 int idx = bio->bi_idx + next_idx;
2250
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002251 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002253 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002254 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 break;
2256 }
2257
2258 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2259 BIO_BUG_ON(nbytes > bio->bi_size);
2260
2261 /*
2262 * not a complete bvec done
2263 */
2264 if (unlikely(nbytes > nr_bytes)) {
2265 bio_nbytes += nr_bytes;
2266 total_bytes += nr_bytes;
2267 break;
2268 }
2269
2270 /*
2271 * advance to the next vector
2272 */
2273 next_idx++;
2274 bio_nbytes += nbytes;
2275 }
2276
2277 total_bytes += nbytes;
2278 nr_bytes -= nbytes;
2279
Jens Axboe6728cb02008-01-31 13:03:55 +01002280 bio = req->bio;
2281 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 /*
2283 * end more in this run, or just return 'not-done'
2284 */
2285 if (unlikely(nr_bytes <= 0))
2286 break;
2287 }
2288 }
2289
2290 /*
2291 * completely done
2292 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002293 if (!req->bio) {
2294 /*
2295 * Reset counters so that the request stacking driver
2296 * can find how many bytes remain in the request
2297 * later.
2298 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002299 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002300 return false;
2301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 /*
2304 * if the request wasn't completed, update state
2305 */
2306 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002307 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 bio->bi_idx += next_idx;
2309 bio_iovec(bio)->bv_offset += nr_bytes;
2310 bio_iovec(bio)->bv_len -= nr_bytes;
2311 }
2312
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002313 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002314 req->buffer = bio_data(req->bio);
2315
2316 /* update sector only for requests with clear definition of sector */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002317 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002318 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002319
Tejun Heo80a761f2009-07-03 17:48:17 +09002320 /* mixed attributes always follow the first bio */
2321 if (req->cmd_flags & REQ_MIXED_MERGE) {
2322 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2323 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2324 }
2325
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002326 /*
2327 * If total number of sectors is less than the first segment
2328 * size, something has gone terribly wrong.
2329 */
2330 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002331 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002332 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002333 }
2334
2335 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002337
Tejun Heo2e60e022009-04-23 11:05:18 +09002338 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339}
Tejun Heo2e60e022009-04-23 11:05:18 +09002340EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Tejun Heo2e60e022009-04-23 11:05:18 +09002342static bool blk_update_bidi_request(struct request *rq, int error,
2343 unsigned int nr_bytes,
2344 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002345{
Tejun Heo2e60e022009-04-23 11:05:18 +09002346 if (blk_update_request(rq, error, nr_bytes))
2347 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002348
Tejun Heo2e60e022009-04-23 11:05:18 +09002349 /* Bidi request must be completed as a whole */
2350 if (unlikely(blk_bidi_rq(rq)) &&
2351 blk_update_request(rq->next_rq, error, bidi_bytes))
2352 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002353
Jens Axboee2e1a142010-06-09 10:42:09 +02002354 if (blk_queue_add_random(rq->q))
2355 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002356
2357 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358}
2359
James Bottomley28018c22010-07-01 19:49:17 +09002360/**
2361 * blk_unprep_request - unprepare a request
2362 * @req: the request
2363 *
2364 * This function makes a request ready for complete resubmission (or
2365 * completion). It happens only after all error handling is complete,
2366 * so represents the appropriate moment to deallocate any resources
2367 * that were allocated to the request in the prep_rq_fn. The queue
2368 * lock is held when calling this.
2369 */
2370void blk_unprep_request(struct request *req)
2371{
2372 struct request_queue *q = req->q;
2373
2374 req->cmd_flags &= ~REQ_DONTPREP;
2375 if (q->unprep_rq_fn)
2376 q->unprep_rq_fn(q, req);
2377}
2378EXPORT_SYMBOL_GPL(blk_unprep_request);
2379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380/*
2381 * queue lock must be held
2382 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002383static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002385 if (blk_rq_tagged(req))
2386 blk_queue_end_tag(req->q, req);
2387
James Bottomleyba396a62009-05-27 14:17:08 +02002388 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002390 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002391 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Mike Andersone78042e2008-10-30 02:16:20 -07002393 blk_delete_timer(req);
2394
James Bottomley28018c22010-07-01 19:49:17 +09002395 if (req->cmd_flags & REQ_DONTPREP)
2396 blk_unprep_request(req);
2397
2398
Jens Axboebc58ba92009-01-23 10:54:44 +01002399 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002400
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002402 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002403 else {
2404 if (blk_bidi_rq(req))
2405 __blk_put_request(req->next_rq->q, req->next_rq);
2406
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 }
2409}
2410
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002411/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002412 * blk_end_bidi_request - Complete a bidi request
2413 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002414 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002415 * @nr_bytes: number of bytes to complete @rq
2416 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002417 *
2418 * Description:
2419 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002420 * Drivers that supports bidi can safely call this member for any
2421 * type of request, bidi or uni. In the later case @bidi_bytes is
2422 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002423 *
2424 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002425 * %false - we are done with this request
2426 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002427 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002428static bool blk_end_bidi_request(struct request *rq, int error,
2429 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002430{
2431 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002432 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002433
Tejun Heo2e60e022009-04-23 11:05:18 +09002434 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2435 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002436
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002437 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002438 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002439 spin_unlock_irqrestore(q->queue_lock, flags);
2440
Tejun Heo2e60e022009-04-23 11:05:18 +09002441 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002442}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002443
2444/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002445 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2446 * @rq: the request to complete
2447 * @error: %0 for success, < %0 for error
2448 * @nr_bytes: number of bytes to complete @rq
2449 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002450 *
2451 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002452 * Identical to blk_end_bidi_request() except that queue lock is
2453 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002454 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002455 * Return:
2456 * %false - we are done with this request
2457 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002458 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002459bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002460 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002461{
Tejun Heo2e60e022009-04-23 11:05:18 +09002462 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2463 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002464
Tejun Heo2e60e022009-04-23 11:05:18 +09002465 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002466
Tejun Heo2e60e022009-04-23 11:05:18 +09002467 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002468}
2469
2470/**
2471 * blk_end_request - Helper function for drivers to complete the request.
2472 * @rq: the request being processed
2473 * @error: %0 for success, < %0 for error
2474 * @nr_bytes: number of bytes to complete
2475 *
2476 * Description:
2477 * Ends I/O on a number of bytes attached to @rq.
2478 * If @rq has leftover, sets it up for the next range of segments.
2479 *
2480 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002481 * %false - we are done with this request
2482 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002483 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002484bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002485{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002486 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002487}
Jens Axboe56ad1742009-07-28 22:11:24 +02002488EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002489
2490/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002491 * blk_end_request_all - Helper function for drives to finish the request.
2492 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002493 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002494 *
2495 * Description:
2496 * Completely finish @rq.
2497 */
2498void blk_end_request_all(struct request *rq, int error)
2499{
2500 bool pending;
2501 unsigned int bidi_bytes = 0;
2502
2503 if (unlikely(blk_bidi_rq(rq)))
2504 bidi_bytes = blk_rq_bytes(rq->next_rq);
2505
2506 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2507 BUG_ON(pending);
2508}
Jens Axboe56ad1742009-07-28 22:11:24 +02002509EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002510
2511/**
2512 * blk_end_request_cur - Helper function to finish the current request chunk.
2513 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002514 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002515 *
2516 * Description:
2517 * Complete the current consecutively mapped chunk from @rq.
2518 *
2519 * Return:
2520 * %false - we are done with this request
2521 * %true - still buffers pending for this request
2522 */
2523bool blk_end_request_cur(struct request *rq, int error)
2524{
2525 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2526}
Jens Axboe56ad1742009-07-28 22:11:24 +02002527EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002528
2529/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002530 * blk_end_request_err - Finish a request till the next failure boundary.
2531 * @rq: the request to finish till the next failure boundary for
2532 * @error: must be negative errno
2533 *
2534 * Description:
2535 * Complete @rq till the next failure boundary.
2536 *
2537 * Return:
2538 * %false - we are done with this request
2539 * %true - still buffers pending for this request
2540 */
2541bool blk_end_request_err(struct request *rq, int error)
2542{
2543 WARN_ON(error >= 0);
2544 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2545}
2546EXPORT_SYMBOL_GPL(blk_end_request_err);
2547
2548/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002549 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002550 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002551 * @error: %0 for success, < %0 for error
2552 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002553 *
2554 * Description:
2555 * Must be called with queue lock held unlike blk_end_request().
2556 *
2557 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002558 * %false - we are done with this request
2559 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002560 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002561bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002562{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002563 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002564}
Jens Axboe56ad1742009-07-28 22:11:24 +02002565EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002566
2567/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002568 * __blk_end_request_all - Helper function for drives to finish the request.
2569 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002570 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002571 *
2572 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002573 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002574 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002575void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002576{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002577 bool pending;
2578 unsigned int bidi_bytes = 0;
2579
2580 if (unlikely(blk_bidi_rq(rq)))
2581 bidi_bytes = blk_rq_bytes(rq->next_rq);
2582
2583 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2584 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002585}
Jens Axboe56ad1742009-07-28 22:11:24 +02002586EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002587
2588/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002589 * __blk_end_request_cur - Helper function to finish the current request chunk.
2590 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002591 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002592 *
2593 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002594 * Complete the current consecutively mapped chunk from @rq. Must
2595 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002596 *
2597 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002598 * %false - we are done with this request
2599 * %true - still buffers pending for this request
2600 */
2601bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002602{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002603 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002604}
Jens Axboe56ad1742009-07-28 22:11:24 +02002605EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002606
Tejun Heo80a761f2009-07-03 17:48:17 +09002607/**
2608 * __blk_end_request_err - Finish a request till the next failure boundary.
2609 * @rq: the request to finish till the next failure boundary for
2610 * @error: must be negative errno
2611 *
2612 * Description:
2613 * Complete @rq till the next failure boundary. Must be called
2614 * with queue lock held.
2615 *
2616 * Return:
2617 * %false - we are done with this request
2618 * %true - still buffers pending for this request
2619 */
2620bool __blk_end_request_err(struct request *rq, int error)
2621{
2622 WARN_ON(error >= 0);
2623 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2624}
2625EXPORT_SYMBOL_GPL(__blk_end_request_err);
2626
Jens Axboe86db1e22008-01-29 14:53:40 +01002627void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2628 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002630 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002631 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
David Woodhousefb2dce82008-08-05 18:01:53 +01002633 if (bio_has_data(bio)) {
2634 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002635 rq->buffer = bio_data(bio);
2636 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002637 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639
NeilBrown66846572007-08-16 13:31:28 +02002640 if (bio->bi_bdev)
2641 rq->rq_disk = bio->bi_bdev->bd_disk;
2642}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002644#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2645/**
2646 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2647 * @rq: the request to be flushed
2648 *
2649 * Description:
2650 * Flush all pages in @rq.
2651 */
2652void rq_flush_dcache_pages(struct request *rq)
2653{
2654 struct req_iterator iter;
2655 struct bio_vec *bvec;
2656
2657 rq_for_each_segment(bvec, rq, iter)
2658 flush_dcache_page(bvec->bv_page);
2659}
2660EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2661#endif
2662
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002663/**
2664 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2665 * @q : the queue of the device being checked
2666 *
2667 * Description:
2668 * Check if underlying low-level drivers of a device are busy.
2669 * If the drivers want to export their busy state, they must set own
2670 * exporting function using blk_queue_lld_busy() first.
2671 *
2672 * Basically, this function is used only by request stacking drivers
2673 * to stop dispatching requests to underlying devices when underlying
2674 * devices are busy. This behavior helps more I/O merging on the queue
2675 * of the request stacking driver and prevents I/O throughput regression
2676 * on burst I/O load.
2677 *
2678 * Return:
2679 * 0 - Not busy (The request stacking driver should dispatch request)
2680 * 1 - Busy (The request stacking driver should stop dispatching request)
2681 */
2682int blk_lld_busy(struct request_queue *q)
2683{
2684 if (q->lld_busy_fn)
2685 return q->lld_busy_fn(q);
2686
2687 return 0;
2688}
2689EXPORT_SYMBOL_GPL(blk_lld_busy);
2690
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002691/**
2692 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2693 * @rq: the clone request to be cleaned up
2694 *
2695 * Description:
2696 * Free all bios in @rq for a cloned request.
2697 */
2698void blk_rq_unprep_clone(struct request *rq)
2699{
2700 struct bio *bio;
2701
2702 while ((bio = rq->bio) != NULL) {
2703 rq->bio = bio->bi_next;
2704
2705 bio_put(bio);
2706 }
2707}
2708EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2709
2710/*
2711 * Copy attributes of the original request to the clone request.
2712 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2713 */
2714static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2715{
2716 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002717 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002718 dst->cmd_type = src->cmd_type;
2719 dst->__sector = blk_rq_pos(src);
2720 dst->__data_len = blk_rq_bytes(src);
2721 dst->nr_phys_segments = src->nr_phys_segments;
2722 dst->ioprio = src->ioprio;
2723 dst->extra_len = src->extra_len;
2724}
2725
2726/**
2727 * blk_rq_prep_clone - Helper function to setup clone request
2728 * @rq: the request to be setup
2729 * @rq_src: original request to be cloned
2730 * @bs: bio_set that bios for clone are allocated from
2731 * @gfp_mask: memory allocation mask for bio
2732 * @bio_ctr: setup function to be called for each clone bio.
2733 * Returns %0 for success, non %0 for failure.
2734 * @data: private data to be passed to @bio_ctr
2735 *
2736 * Description:
2737 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2738 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2739 * are not copied, and copying such parts is the caller's responsibility.
2740 * Also, pages which the original bios are pointing to are not copied
2741 * and the cloned bios just point same pages.
2742 * So cloned bios must be completed before original bios, which means
2743 * the caller must complete @rq before @rq_src.
2744 */
2745int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2746 struct bio_set *bs, gfp_t gfp_mask,
2747 int (*bio_ctr)(struct bio *, struct bio *, void *),
2748 void *data)
2749{
2750 struct bio *bio, *bio_src;
2751
2752 if (!bs)
2753 bs = fs_bio_set;
2754
2755 blk_rq_init(NULL, rq);
2756
2757 __rq_for_each_bio(bio_src, rq_src) {
2758 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2759 if (!bio)
2760 goto free_and_out;
2761
2762 __bio_clone(bio, bio_src);
2763
2764 if (bio_integrity(bio_src) &&
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002765 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002766 goto free_and_out;
2767
2768 if (bio_ctr && bio_ctr(bio, bio_src, data))
2769 goto free_and_out;
2770
2771 if (rq->bio) {
2772 rq->biotail->bi_next = bio;
2773 rq->biotail = bio;
2774 } else
2775 rq->bio = rq->biotail = bio;
2776 }
2777
2778 __blk_rq_prep_clone(rq, rq_src);
2779
2780 return 0;
2781
2782free_and_out:
2783 if (bio)
2784 bio_free(bio, bs);
2785 blk_rq_unprep_clone(rq);
2786
2787 return -ENOMEM;
2788}
2789EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2790
Jens Axboe18887ad2008-07-28 13:08:45 +02002791int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792{
2793 return queue_work(kblockd_workqueue, work);
2794}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795EXPORT_SYMBOL(kblockd_schedule_work);
2796
Vivek Goyale43473b2010-09-15 17:06:35 -04002797int kblockd_schedule_delayed_work(struct request_queue *q,
2798 struct delayed_work *dwork, unsigned long delay)
2799{
2800 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2801}
2802EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2803
Jens Axboe73c10102011-03-08 13:19:51 +01002804#define PLUG_MAGIC 0x91827364
2805
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002806/**
2807 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2808 * @plug: The &struct blk_plug that needs to be initialized
2809 *
2810 * Description:
2811 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2812 * pending I/O should the task end up blocking between blk_start_plug() and
2813 * blk_finish_plug(). This is important from a performance perspective, but
2814 * also ensures that we don't deadlock. For instance, if the task is blocking
2815 * for a memory allocation, memory reclaim could end up wanting to free a
2816 * page belonging to that request that is currently residing in our private
2817 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2818 * this kind of deadlock.
2819 */
Jens Axboe73c10102011-03-08 13:19:51 +01002820void blk_start_plug(struct blk_plug *plug)
2821{
2822 struct task_struct *tsk = current;
2823
2824 plug->magic = PLUG_MAGIC;
2825 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002826 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002827 plug->should_sort = 0;
2828
2829 /*
2830 * If this is a nested plug, don't actually assign it. It will be
2831 * flushed on its own.
2832 */
2833 if (!tsk->plug) {
2834 /*
2835 * Store ordering should not be needed here, since a potential
2836 * preempt will imply a full memory barrier
2837 */
2838 tsk->plug = plug;
2839 }
2840}
2841EXPORT_SYMBOL(blk_start_plug);
2842
2843static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2844{
2845 struct request *rqa = container_of(a, struct request, queuelist);
2846 struct request *rqb = container_of(b, struct request, queuelist);
2847
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002848 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002849}
2850
Jens Axboe49cac012011-04-16 13:51:05 +02002851/*
2852 * If 'from_schedule' is true, then postpone the dispatch of requests
2853 * until a safe kblockd context. We due this to avoid accidental big
2854 * additional stack usage in driver dispatch, in places where the originally
2855 * plugger did not intend it.
2856 */
Jens Axboef6603782011-04-15 15:49:07 +02002857static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002858 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002859 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002860{
Jens Axboe49cac012011-04-16 13:51:05 +02002861 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002862
2863 /*
Tejun Heo8ba61432011-12-14 00:33:37 +01002864 * Don't mess with dead queue.
2865 */
2866 if (unlikely(blk_queue_dead(q))) {
2867 spin_unlock(q->queue_lock);
2868 return;
2869 }
2870
2871 /*
Jens Axboe99e22592011-04-18 09:59:55 +02002872 * If we are punting this to kblockd, then we can safely drop
2873 * the queue_lock before waking kblockd (which needs to take
2874 * this lock).
2875 */
2876 if (from_schedule) {
2877 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002878 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002879 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002880 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002881 spin_unlock(q->queue_lock);
2882 }
2883
Jens Axboe94b5eb22011-04-12 10:12:19 +02002884}
2885
NeilBrown048c9372011-04-18 09:52:22 +02002886static void flush_plug_callbacks(struct blk_plug *plug)
2887{
2888 LIST_HEAD(callbacks);
2889
2890 if (list_empty(&plug->cb_list))
2891 return;
2892
2893 list_splice_init(&plug->cb_list, &callbacks);
2894
2895 while (!list_empty(&callbacks)) {
2896 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2897 struct blk_plug_cb,
2898 list);
2899 list_del(&cb->list);
2900 cb->callback(cb);
2901 }
2902}
2903
Jens Axboe49cac012011-04-16 13:51:05 +02002904void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002905{
2906 struct request_queue *q;
2907 unsigned long flags;
2908 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002909 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002910 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002911
2912 BUG_ON(plug->magic != PLUG_MAGIC);
2913
NeilBrown048c9372011-04-18 09:52:22 +02002914 flush_plug_callbacks(plug);
Jens Axboe73c10102011-03-08 13:19:51 +01002915 if (list_empty(&plug->list))
2916 return;
2917
NeilBrown109b8122011-04-11 14:13:10 +02002918 list_splice_init(&plug->list, &list);
2919
2920 if (plug->should_sort) {
2921 list_sort(NULL, &list, plug_rq_cmp);
2922 plug->should_sort = 0;
2923 }
Jens Axboe73c10102011-03-08 13:19:51 +01002924
2925 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002926 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002927
2928 /*
2929 * Save and disable interrupts here, to avoid doing it for every
2930 * queue lock we have to take.
2931 */
Jens Axboe73c10102011-03-08 13:19:51 +01002932 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002933 while (!list_empty(&list)) {
2934 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002935 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002936 BUG_ON(!rq->q);
2937 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002938 /*
2939 * This drops the queue lock
2940 */
2941 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002942 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002943 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002944 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002945 spin_lock(q->queue_lock);
2946 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002947
2948 /*
2949 * Short-circuit if @q is dead
2950 */
2951 if (unlikely(blk_queue_dead(q))) {
2952 __blk_end_request_all(rq, -ENODEV);
2953 continue;
2954 }
2955
Jens Axboe73c10102011-03-08 13:19:51 +01002956 /*
2957 * rq is already accounted, so use raw insert
2958 */
Jens Axboe401a18e2011-03-25 16:57:52 +01002959 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2960 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2961 else
2962 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002963
2964 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01002965 }
2966
Jens Axboe99e22592011-04-18 09:59:55 +02002967 /*
2968 * This drops the queue lock
2969 */
2970 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002971 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002972
Jens Axboe73c10102011-03-08 13:19:51 +01002973 local_irq_restore(flags);
2974}
Jens Axboe73c10102011-03-08 13:19:51 +01002975
2976void blk_finish_plug(struct blk_plug *plug)
2977{
Jens Axboef6603782011-04-15 15:49:07 +02002978 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02002979
2980 if (plug == current->plug)
2981 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01002982}
2983EXPORT_SYMBOL(blk_finish_plug);
2984
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985int __init blk_dev_init(void)
2986{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02002987 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2988 sizeof(((struct request *)0)->cmd_flags));
2989
Tejun Heo89b90be2011-01-03 15:01:47 +01002990 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2991 kblockd_workqueue = alloc_workqueue("kblockd",
2992 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 if (!kblockd_workqueue)
2994 panic("Failed to create kblockd\n");
2995
2996 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09002997 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
Jens Axboe8324aa92008-01-29 14:51:59 +01002999 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003000 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 return 0;
3003}