blob: a33870b1847bb70c6ef1937f7bef4e93bdf9c980 [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{
Tejun Heo136b5722012-08-21 13:18:24 -0700265 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{
Tejun Heoe7c2f962012-08-21 13:18:24 -0700322 if (likely(!blk_queue_stopped(q)))
323 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324}
Jens Axboec21e6be2011-04-19 13:32:46 +0200325EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200326
327/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200330 *
331 * Description:
332 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900333 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200334 */
335void blk_run_queue(struct request_queue *q)
336{
337 unsigned long flags;
338
339 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200340 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 spin_unlock_irqrestore(q->queue_lock, flags);
342}
343EXPORT_SYMBOL(blk_run_queue);
344
Jens Axboe165125e2007-07-24 09:28:11 +0200345void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500346{
347 kobject_put(&q->kobj);
348}
Jens Axboed86e0e82011-05-27 07:44:43 +0200349EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500350
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200351/**
352 * blk_drain_queue - drain requests from request_queue
353 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200355 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200359 */
Tejun Heoc9a929d2011-10-19 14:42:16 +0200360void blk_drain_queue(struct request_queue *q, bool drain_all)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200361{
Asias He458f27a2012-06-15 08:45:25 +0200362 int i;
363
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200364 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100365 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200366
367 spin_lock_irq(q->queue_lock);
368
Tejun Heob855b042012-03-06 21:24:55 +0100369 /*
370 * The caller might be trying to drain @q before its
371 * elevator is initialized.
372 */
373 if (q->elevator)
374 elv_drain_elevator(q);
375
Tejun Heo5efd6112012-03-05 13:15:12 -0800376 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200377
Tejun Heo4eabc942011-12-15 20:03:04 +0100378 /*
379 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100384 */
Tejun Heob855b042012-03-06 21:24:55 +0100385 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100386 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200387
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700388 drain |= q->nr_rqs_elvpriv;
Tejun Heo481a7d62011-12-14 00:33:37 +0100389
390 /*
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
394 */
395 if (drain_all) {
396 drain |= !list_empty(&q->queue_head);
397 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700398 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100399 drain |= q->in_flight[i];
400 drain |= !list_empty(&q->flush_queue[i]);
401 }
402 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200403
404 spin_unlock_irq(q->queue_lock);
405
Tejun Heo481a7d62011-12-14 00:33:37 +0100406 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200407 break;
408 msleep(10);
409 }
Asias He458f27a2012-06-15 08:45:25 +0200410
411 /*
412 * With queue marked dead, any woken up waiter will fail the
413 * allocation path, so the wakeup chaining is lost and we're
414 * left with hung waiters. We need to wake up those waiters.
415 */
416 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700417 struct request_list *rl;
418
Asias He458f27a2012-06-15 08:45:25 +0200419 spin_lock_irq(q->queue_lock);
Tejun Heoa0516612012-06-26 15:05:44 -0700420
421 blk_queue_for_each_rl(rl, q)
422 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
423 wake_up_all(&rl->wait[i]);
424
Asias He458f27a2012-06-15 08:45:25 +0200425 spin_unlock_irq(q->queue_lock);
426 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200427}
428
Tejun Heoc9a929d2011-10-19 14:42:16 +0200429/**
Tejun Heod7325802012-03-05 13:14:58 -0800430 * blk_queue_bypass_start - enter queue bypass mode
431 * @q: queue of interest
432 *
433 * In bypass mode, only the dispatch FIFO queue of @q is used. This
434 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800435 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700436 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
437 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800438 */
439void blk_queue_bypass_start(struct request_queue *q)
440{
Tejun Heob82d4b12012-04-13 13:11:31 -0700441 bool drain;
442
Tejun Heod7325802012-03-05 13:14:58 -0800443 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700444 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800445 queue_flag_set(QUEUE_FLAG_BYPASS, q);
446 spin_unlock_irq(q->queue_lock);
447
Tejun Heob82d4b12012-04-13 13:11:31 -0700448 if (drain) {
449 blk_drain_queue(q, false);
450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
451 synchronize_rcu();
452 }
Tejun Heod7325802012-03-05 13:14:58 -0800453}
454EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
455
456/**
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
459 *
460 * Leave bypass mode and restore the normal queueing behavior.
461 */
462void blk_queue_bypass_end(struct request_queue *q)
463{
464 spin_lock_irq(q->queue_lock);
465 if (!--q->bypass_depth)
466 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
467 WARN_ON_ONCE(q->bypass_depth < 0);
468 spin_unlock_irq(q->queue_lock);
469}
470EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
471
472/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
475 *
476 * Mark @q DEAD, drain all pending requests, destroy and put it. All
477 * future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500478 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100479void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500480{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200481 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700482
Tejun Heoc9a929d2011-10-19 14:42:16 +0200483 /* mark @q DEAD, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500484 mutex_lock(&q->sysfs_lock);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200485 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200486 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800487
Tejun Heo80fd9972012-04-13 14:50:53 -0700488 /*
489 * Dead queue is permanently in bypass mode till released. Note
490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
496 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800497 q->bypass_depth++;
498 queue_flag_set(QUEUE_FLAG_BYPASS, q);
499
Tejun Heoc9a929d2011-10-19 14:42:16 +0200500 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
502 queue_flag_set(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200503 spin_unlock_irq(lock);
504 mutex_unlock(&q->sysfs_lock);
505
Tejun Heob855b042012-03-06 21:24:55 +0100506 /* drain all requests queued before DEAD marking */
507 blk_drain_queue(q, true);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200508
509 /* @q won't process any more request, flush async actions */
510 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
511 blk_sync_queue(q);
512
Asias He5e5cfac2012-05-24 23:28:52 +0800513 spin_lock_irq(lock);
514 if (q->queue_lock != &q->__queue_lock)
515 q->queue_lock = &q->__queue_lock;
516 spin_unlock_irq(lock);
517
Tejun Heoc9a929d2011-10-19 14:42:16 +0200518 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500519 blk_put_queue(q);
520}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521EXPORT_SYMBOL(blk_cleanup_queue);
522
Tejun Heo5b788ce2012-06-04 20:40:59 -0700523int blk_init_rl(struct request_list *rl, struct request_queue *q,
524 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400526 if (unlikely(rl->rq_pool))
527 return 0;
528
Tejun Heo5b788ce2012-06-04 20:40:59 -0700529 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200530 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
531 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200532 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
533 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Christoph Lameter19460892005-06-23 00:08:19 -0700535 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700536 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700537 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 if (!rl->rq_pool)
539 return -ENOMEM;
540
541 return 0;
542}
543
Tejun Heo5b788ce2012-06-04 20:40:59 -0700544void blk_exit_rl(struct request_list *rl)
545{
546 if (rl->rq_pool)
547 mempool_destroy(rl->rq_pool);
548}
549
Jens Axboe165125e2007-07-24 09:28:11 +0200550struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
Christoph Lameter19460892005-06-23 00:08:19 -0700552 return blk_alloc_queue_node(gfp_mask, -1);
553}
554EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Jens Axboe165125e2007-07-24 09:28:11 +0200556struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700557{
Jens Axboe165125e2007-07-24 09:28:11 +0200558 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700559 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700560
Jens Axboe8324aa92008-01-29 14:51:59 +0100561 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700562 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 if (!q)
564 return NULL;
565
Dan Carpenter00380a42012-03-23 09:58:54 +0100566 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100567 if (q->id < 0)
568 goto fail_q;
569
Jens Axboe0989a022009-06-12 14:42:56 +0200570 q->backing_dev_info.ra_pages =
571 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
572 q->backing_dev_info.state = 0;
573 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200574 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100575 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200576
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700577 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100578 if (err)
579 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700580
Matthew Garrett31373d02010-04-06 14:25:14 +0200581 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
582 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700583 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100584 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700585 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100586 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800587#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800588 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800589#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100590 INIT_LIST_HEAD(&q->flush_queue[0]);
591 INIT_LIST_HEAD(&q->flush_queue[1]);
592 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500593 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500594
Jens Axboe8324aa92008-01-29 14:51:59 +0100595 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Al Viro483f4af2006-03-18 18:34:37 -0500597 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700598 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500599
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500600 /*
601 * By default initialize queue_lock to internal lock and driver can
602 * override it later if need be.
603 */
604 q->queue_lock = &q->__queue_lock;
605
Tejun Heob82d4b12012-04-13 13:11:31 -0700606 /*
607 * A queue starts its life with bypass turned on to avoid
608 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700609 * init. The initial bypass will be finished when the queue is
610 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700611 */
612 q->bypass_depth = 1;
613 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
614
Tejun Heo5efd6112012-03-05 13:15:12 -0800615 if (blkcg_init_queue(q))
Tejun Heof51b8022012-03-05 13:15:05 -0800616 goto fail_id;
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100619
620fail_id:
621 ida_simple_remove(&blk_queue_ida, q->id);
622fail_q:
623 kmem_cache_free(blk_requestq_cachep, q);
624 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
Christoph Lameter19460892005-06-23 00:08:19 -0700626EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628/**
629 * blk_init_queue - prepare a request queue for use with a block device
630 * @rfn: The function to be called to process requests that have been
631 * placed on the queue.
632 * @lock: Request queue spin lock
633 *
634 * Description:
635 * If a block device wishes to use the standard request handling procedures,
636 * which sorts requests and coalesces adjacent requests, then it must
637 * call blk_init_queue(). The function @rfn will be called when there
638 * are requests on the queue that need to be processed. If the device
639 * supports plugging, then @rfn may not be called immediately when requests
640 * are available on the queue, but may be called at some time later instead.
641 * Plugged queues are generally unplugged when a buffer belonging to one
642 * of the requests on the queue is needed, or due to memory pressure.
643 *
644 * @rfn is not required, or even expected, to remove all requests off the
645 * queue, but only as many as it can handle at a time. If it does leave
646 * requests on the queue, it is responsible for arranging that the requests
647 * get dealt with eventually.
648 *
649 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200650 * request queue; this lock will be taken also from interrupt context, so irq
651 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200653 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 * it didn't succeed.
655 *
656 * Note:
657 * blk_init_queue() must be paired with a blk_cleanup_queue() call
658 * when the block device is deactivated (such as at module unload).
659 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700660
Jens Axboe165125e2007-07-24 09:28:11 +0200661struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
Christoph Lameter19460892005-06-23 00:08:19 -0700663 return blk_init_queue_node(rfn, lock, -1);
664}
665EXPORT_SYMBOL(blk_init_queue);
666
Jens Axboe165125e2007-07-24 09:28:11 +0200667struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700668blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
669{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600670 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600672 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
673 if (!uninit_q)
674 return NULL;
675
Mike Snitzer51514122011-11-23 10:59:13 +0100676 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600677 if (!q)
678 blk_cleanup_queue(uninit_q);
679
680 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200681}
682EXPORT_SYMBOL(blk_init_queue_node);
683
684struct request_queue *
685blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
686 spinlock_t *lock)
687{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (!q)
689 return NULL;
690
Tejun Heoa0516612012-06-26 15:05:44 -0700691 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500692 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900696 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700697 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500698
699 /* Override internal queue lock with supplied lock pointer */
700 if (lock)
701 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Jens Axboef3b144a2009-03-06 08:48:33 +0100703 /*
704 * This also sets hw/phys segments, boundary and size
705 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200706 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Alan Stern44ec9542007-02-20 11:01:57 -0500708 q->sg_reserved_size = INT_MAX;
709
Tejun Heob82d4b12012-04-13 13:11:31 -0700710 /* init elevator */
711 if (elevator_init(q, NULL))
712 return NULL;
Tejun Heob82d4b12012-04-13 13:11:31 -0700713 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
Mike Snitzer51514122011-11-23 10:59:13 +0100715EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Tejun Heo09ac46c2011-12-14 00:33:38 +0100717bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Tejun Heo34f60552011-12-14 00:33:37 +0100719 if (likely(!blk_queue_dead(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100720 __blk_get_queue(q);
721 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723
Tejun Heo09ac46c2011-12-14 00:33:38 +0100724 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
Jens Axboed86e0e82011-05-27 07:44:43 +0200726EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Tejun Heo5b788ce2012-06-04 20:40:59 -0700728static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100730 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700731 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100732 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100733 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100734 }
735
Tejun Heo5b788ce2012-06-04 20:40:59 -0700736 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739/*
740 * ioc_batching returns true if the ioc is a valid batching request and
741 * should be given priority access to a request.
742 */
Jens Axboe165125e2007-07-24 09:28:11 +0200743static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 if (!ioc)
746 return 0;
747
748 /*
749 * Make sure the process is able to allocate at least 1 request
750 * even if the batch times out, otherwise we could theoretically
751 * lose wakeups.
752 */
753 return ioc->nr_batch_requests == q->nr_batching ||
754 (ioc->nr_batch_requests > 0
755 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
756}
757
758/*
759 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
760 * will cause the process to be a "batcher" on all queues in the system. This
761 * is the behaviour we want though - once it gets a wakeup it should be given
762 * a nice run.
763 */
Jens Axboe165125e2007-07-24 09:28:11 +0200764static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
766 if (!ioc || ioc_batching(q, ioc))
767 return;
768
769 ioc->nr_batch_requests = q->nr_batching;
770 ioc->last_waited = jiffies;
771}
772
Tejun Heo5b788ce2012-06-04 20:40:59 -0700773static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700775 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Tejun Heoa0516612012-06-26 15:05:44 -0700777 /*
778 * bdi isn't aware of blkcg yet. As all async IOs end up root
779 * blkcg anyway, just use root blkcg state.
780 */
781 if (rl == &q->root_rl &&
782 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200783 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Jens Axboe1faa16d2009-04-06 14:48:01 +0200785 if (rl->count[sync] + 1 <= q->nr_requests) {
786 if (waitqueue_active(&rl->wait[sync]))
787 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Tejun Heo5b788ce2012-06-04 20:40:59 -0700789 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
791}
792
793/*
794 * A request has just been released. Account for it, update the full and
795 * congestion status, wake up any waiters. Called under q->queue_lock.
796 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700797static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700799 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200800 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700802 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200803 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200804 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700805 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Tejun Heo5b788ce2012-06-04 20:40:59 -0700807 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Jens Axboe1faa16d2009-04-06 14:48:01 +0200809 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700810 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100814 * Determine if elevator data should be initialized when allocating the
815 * request associated with @bio.
816 */
817static bool blk_rq_should_init_elevator(struct bio *bio)
818{
819 if (!bio)
820 return true;
821
822 /*
823 * Flush requests do not use the elevator so skip initialization.
824 * This allows a request to share the flush and elevator data.
825 */
826 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
827 return false;
828
829 return true;
830}
831
Tejun Heoda8303c2011-10-19 14:33:05 +0200832/**
Tejun Heo852c7882012-03-05 13:15:27 -0800833 * rq_ioc - determine io_context for request allocation
834 * @bio: request being allocated is for this bio (can be %NULL)
835 *
836 * Determine io_context to use for request allocation for @bio. May return
837 * %NULL if %current->io_context doesn't exist.
838 */
839static struct io_context *rq_ioc(struct bio *bio)
840{
841#ifdef CONFIG_BLK_CGROUP
842 if (bio && bio->bi_ioc)
843 return bio->bi_ioc;
844#endif
845 return current->io_context;
846}
847
848/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700849 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700850 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200851 * @rw_flags: RW and SYNC flags
852 * @bio: bio to allocate request for (can be %NULL)
853 * @gfp_mask: allocation mask
854 *
855 * Get a free request from @q. This function may fail under memory
856 * pressure or if @q is dead.
857 *
858 * Must be callled with @q->queue_lock held and,
859 * Returns %NULL on failure, with @q->queue_lock held.
860 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700862static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700863 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700865 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800866 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700867 struct elevator_type *et = q->elevator->type;
868 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100869 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200870 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200871 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Tejun Heo34f60552011-12-14 00:33:37 +0100873 if (unlikely(blk_queue_dead(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200874 return NULL;
875
Jens Axboe7749a8d2006-12-13 13:02:26 +0100876 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100877 if (may_queue == ELV_MQUEUE_NO)
878 goto rq_starved;
879
Jens Axboe1faa16d2009-04-06 14:48:01 +0200880 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
881 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100882 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100883 * The queue will fill after this allocation, so set
884 * it as full, and mark this process as "batching".
885 * This process will be allowed to complete a batch of
886 * requests, others will be blocked.
887 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700888 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100889 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700890 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100891 } else {
892 if (may_queue != ELV_MQUEUE_MUST
893 && !ioc_batching(q, ioc)) {
894 /*
895 * The queue is full and the allocating
896 * process is not a "batcher", and not
897 * exempted by the IO scheduler
898 */
Tejun Heob6792812012-03-05 13:15:23 -0800899 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100900 }
901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
Tejun Heoa0516612012-06-26 15:05:44 -0700903 /*
904 * bdi isn't aware of blkcg yet. As all async IOs end up
905 * root blkcg anyway, just use root blkcg state.
906 */
907 if (rl == &q->root_rl)
908 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
910
Jens Axboe082cf692005-06-28 16:35:11 +0200911 /*
912 * Only allow batching queuers to allocate up to 50% over the defined
913 * limit of requests, otherwise we could have thousands of requests
914 * allocated with any setting of ->nr_requests
915 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200916 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800917 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100918
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700919 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200920 rl->count[is_sync]++;
921 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200922
Tejun Heof1f8cc92011-12-14 00:33:42 +0100923 /*
924 * Decide whether the new request will be managed by elevator. If
925 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
926 * prevent the current elevator from being destroyed until the new
927 * request is freed. This guarantees icq's won't be destroyed and
928 * makes creating new ones safe.
929 *
930 * Also, lookup icq while holding queue_lock. If it doesn't exist,
931 * it will be created after releasing queue_lock.
932 */
Tejun Heod7325802012-03-05 13:14:58 -0800933 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200934 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700935 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100936 if (et->icq_cache && ioc)
937 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100938 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200939
Jens Axboef253b862010-10-24 22:06:02 +0200940 if (blk_queue_io_stat(q))
941 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 spin_unlock_irq(q->queue_lock);
943
Tejun Heo29e2b092012-04-19 16:29:21 -0700944 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700945 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700946 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800947 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Tejun Heo29e2b092012-04-19 16:29:21 -0700949 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700950 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700951 rq->cmd_flags = rw_flags | REQ_ALLOCED;
952
Tejun Heoaaf7c682012-04-19 16:29:22 -0700953 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700954 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700955 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700956 if (ioc)
957 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700958 if (!icq)
959 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700960 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700961
962 rq->elv.icq = icq;
963 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
964 goto fail_elvpriv;
965
966 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -0700967 if (icq)
968 get_io_context(icq->ioc);
969 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700970out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100971 /*
972 * ioc may be NULL here, and ioc_batching will be false. That's
973 * OK, if the queue is under the request limit then requests need
974 * not count toward the nr_batch_requests limit. There will always
975 * be some limit enforced by BLK_BATCH_TIME.
976 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (ioc_batching(q, ioc))
978 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100979
Jens Axboe1faa16d2009-04-06 14:48:01 +0200980 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return rq;
Tejun Heob6792812012-03-05 13:15:23 -0800982
Tejun Heoaaf7c682012-04-19 16:29:22 -0700983fail_elvpriv:
984 /*
985 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
986 * and may fail indefinitely under memory pressure and thus
987 * shouldn't stall IO. Treat this request as !elvpriv. This will
988 * disturb iosched and blkcg but weird is bettern than dead.
989 */
990 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
991 dev_name(q->backing_dev_info.dev));
992
993 rq->cmd_flags &= ~REQ_ELVPRIV;
994 rq->elv.icq = NULL;
995
996 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700997 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -0700998 spin_unlock_irq(q->queue_lock);
999 goto out;
1000
Tejun Heob6792812012-03-05 13:15:23 -08001001fail_alloc:
1002 /*
1003 * Allocation failed presumably due to memory. Undo anything we
1004 * might have messed up.
1005 *
1006 * Allocating task should really be put onto the front of the wait
1007 * queue, but this is pretty rare.
1008 */
1009 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001010 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001011
1012 /*
1013 * in the very unlikely event that allocation failed and no
1014 * requests for this direction was pending, mark us starved so that
1015 * freeing of a request in the other direction will notice
1016 * us. another possible fix would be to split the rq mempool into
1017 * READ and WRITE
1018 */
1019rq_starved:
1020 if (unlikely(rl->count[is_sync] == 0))
1021 rl->starved[is_sync] = 1;
1022 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023}
1024
Tejun Heoda8303c2011-10-19 14:33:05 +02001025/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001026 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001027 * @q: request_queue to allocate request from
1028 * @rw_flags: RW and SYNC flags
1029 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001030 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001031 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001032 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1033 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001034 *
1035 * Must be callled with @q->queue_lock held and,
1036 * Returns %NULL on failure, with @q->queue_lock held.
1037 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001039static struct request *get_request(struct request_queue *q, int rw_flags,
1040 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001042 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001043 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001044 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001046
1047 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001048retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001049 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001050 if (rq)
1051 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Tejun Heoa0516612012-06-26 15:05:44 -07001053 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dead(q))) {
1054 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001055 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Tejun Heoa06e05e2012-06-04 20:40:55 -07001058 /* wait on @rl and retry */
1059 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1060 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001061
Tejun Heoa06e05e2012-06-04 20:40:55 -07001062 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Tejun Heoa06e05e2012-06-04 20:40:55 -07001064 spin_unlock_irq(q->queue_lock);
1065 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Tejun Heoa06e05e2012-06-04 20:40:55 -07001067 /*
1068 * After sleeping, we become a "batching" process and will be able
1069 * to allocate at least one request, and up to a big batch of them
1070 * for a small period time. See ioc_batching, ioc_set_batching
1071 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001072 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Tejun Heoa06e05e2012-06-04 20:40:55 -07001074 spin_lock_irq(q->queue_lock);
1075 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001076
Tejun Heoa06e05e2012-06-04 20:40:55 -07001077 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Jens Axboe165125e2007-07-24 09:28:11 +02001080struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
1082 struct request *rq;
1083
1084 BUG_ON(rw != READ && rw != WRITE);
1085
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001086 /* create ioc upfront */
1087 create_io_context(gfp_mask, q->node);
1088
Nick Piggind6344532005-06-28 20:45:14 -07001089 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001090 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001091 if (!rq)
1092 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001093 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 return rq;
1096}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097EXPORT_SYMBOL(blk_get_request);
1098
1099/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001100 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001101 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001102 * @bio: The bio describing the memory mappings that will be submitted for IO.
1103 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001104 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001105 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001106 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1107 * type commands. Where the struct request needs to be farther initialized by
1108 * the caller. It is passed a &struct bio, which describes the memory info of
1109 * the I/O transfer.
1110 *
1111 * The caller of blk_make_request must make sure that bi_io_vec
1112 * are set to describe the memory buffers. That bio_data_dir() will return
1113 * the needed direction of the request. (And all bio's in the passed bio-chain
1114 * are properly set accordingly)
1115 *
1116 * If called under none-sleepable conditions, mapped bio buffers must not
1117 * need bouncing, by calling the appropriate masked or flagged allocator,
1118 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1119 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001120 *
1121 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1122 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1123 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1124 * completion of a bio that hasn't been submitted yet, thus resulting in a
1125 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1126 * of bio_alloc(), as that avoids the mempool deadlock.
1127 * If possible a big IO should be split into smaller parts when allocation
1128 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001129 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001130struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1131 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001132{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001133 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1134
1135 if (unlikely(!rq))
1136 return ERR_PTR(-ENOMEM);
1137
1138 for_each_bio(bio) {
1139 struct bio *bounce_bio = bio;
1140 int ret;
1141
1142 blk_queue_bounce(q, &bounce_bio);
1143 ret = blk_rq_append_bio(q, rq, bounce_bio);
1144 if (unlikely(ret)) {
1145 blk_put_request(rq);
1146 return ERR_PTR(ret);
1147 }
1148 }
1149
1150 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001151}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001152EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001153
1154/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 * blk_requeue_request - put a request back on queue
1156 * @q: request queue where request should be inserted
1157 * @rq: request to be inserted
1158 *
1159 * Description:
1160 * Drivers often keep queueing requests until the hardware cannot accept
1161 * more, when that condition happens we need to put the request back
1162 * on the queue. Must be called with queue lock held.
1163 */
Jens Axboe165125e2007-07-24 09:28:11 +02001164void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001166 blk_delete_timer(rq);
1167 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001168 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (blk_rq_tagged(rq))
1171 blk_queue_end_tag(q, rq);
1172
James Bottomleyba396a62009-05-27 14:17:08 +02001173 BUG_ON(blk_queued_rq(rq));
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 elv_requeue_request(q, rq);
1176}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177EXPORT_SYMBOL(blk_requeue_request);
1178
Jens Axboe73c10102011-03-08 13:19:51 +01001179static void add_acct_request(struct request_queue *q, struct request *rq,
1180 int where)
1181{
1182 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001183 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001184}
1185
Tejun Heo074a7ac2008-08-25 19:56:14 +09001186static void part_round_stats_single(int cpu, struct hd_struct *part,
1187 unsigned long now)
1188{
1189 if (now == part->stamp)
1190 return;
1191
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001192 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001193 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001194 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001195 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1196 }
1197 part->stamp = now;
1198}
1199
1200/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001201 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1202 * @cpu: cpu number for stats access
1203 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 *
1205 * The average IO queue length and utilisation statistics are maintained
1206 * by observing the current state of the queue length and the amount of
1207 * time it has been in this state for.
1208 *
1209 * Normally, that accounting is done on IO completion, but that can result
1210 * in more than a second's worth of IO being accounted for within any one
1211 * second, leading to >100% utilisation. To deal with that, we call this
1212 * function to do a round-off before returning the results when reading
1213 * /proc/diskstats. This accounts immediately for all queue usage up to
1214 * the current jiffies and restarts the counters again.
1215 */
Tejun Heoc9959052008-08-25 19:47:21 +09001216void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001217{
1218 unsigned long now = jiffies;
1219
Tejun Heo074a7ac2008-08-25 19:56:14 +09001220 if (part->partno)
1221 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1222 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001223}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001224EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226/*
1227 * queue lock must be held
1228 */
Jens Axboe165125e2007-07-24 09:28:11 +02001229void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 if (unlikely(!q))
1232 return;
1233 if (unlikely(--req->ref_count))
1234 return;
1235
Tejun Heo8922e162005-10-20 16:23:44 +02001236 elv_completed_request(q, req);
1237
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001238 /* this is a bio leak */
1239 WARN_ON(req->bio != NULL);
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 /*
1242 * Request may not have originated from ll_rw_blk. if not,
1243 * it didn't come out of our reserved rq pools
1244 */
Jens Axboe49171e52006-08-10 08:59:11 +02001245 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001246 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001247 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001250 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Tejun Heoa0516612012-06-26 15:05:44 -07001252 blk_free_request(rl, req);
1253 freed_request(rl, flags);
1254 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001257EXPORT_SYMBOL_GPL(__blk_put_request);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259void blk_put_request(struct request *req)
1260{
Tejun Heo8922e162005-10-20 16:23:44 +02001261 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001262 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001264 spin_lock_irqsave(q->queue_lock, flags);
1265 __blk_put_request(q, req);
1266 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268EXPORT_SYMBOL(blk_put_request);
1269
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001270/**
1271 * blk_add_request_payload - add a payload to a request
1272 * @rq: request to update
1273 * @page: page backing the payload
1274 * @len: length of the payload.
1275 *
1276 * This allows to later add a payload to an already submitted request by
1277 * a block driver. The driver needs to take care of freeing the payload
1278 * itself.
1279 *
1280 * Note that this is a quite horrible hack and nothing but handling of
1281 * discard requests should ever use it.
1282 */
1283void blk_add_request_payload(struct request *rq, struct page *page,
1284 unsigned int len)
1285{
1286 struct bio *bio = rq->bio;
1287
1288 bio->bi_io_vec->bv_page = page;
1289 bio->bi_io_vec->bv_offset = 0;
1290 bio->bi_io_vec->bv_len = len;
1291
1292 bio->bi_size = len;
1293 bio->bi_vcnt = 1;
1294 bio->bi_phys_segments = 1;
1295
1296 rq->__data_len = rq->resid_len = len;
1297 rq->nr_phys_segments = 1;
1298 rq->buffer = bio_data(bio);
1299}
1300EXPORT_SYMBOL_GPL(blk_add_request_payload);
1301
Jens Axboe73c10102011-03-08 13:19:51 +01001302static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1303 struct bio *bio)
1304{
1305 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1306
Jens Axboe73c10102011-03-08 13:19:51 +01001307 if (!ll_back_merge_fn(q, req, bio))
1308 return false;
1309
1310 trace_block_bio_backmerge(q, bio);
1311
1312 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1313 blk_rq_set_mixed_merge(req);
1314
1315 req->biotail->bi_next = bio;
1316 req->biotail = bio;
1317 req->__data_len += bio->bi_size;
1318 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1319
1320 drive_stat_acct(req, 0);
1321 return true;
1322}
1323
1324static bool bio_attempt_front_merge(struct request_queue *q,
1325 struct request *req, struct bio *bio)
1326{
1327 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001328
Jens Axboe73c10102011-03-08 13:19:51 +01001329 if (!ll_front_merge_fn(q, req, bio))
1330 return false;
1331
1332 trace_block_bio_frontmerge(q, bio);
1333
1334 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1335 blk_rq_set_mixed_merge(req);
1336
Jens Axboe73c10102011-03-08 13:19:51 +01001337 bio->bi_next = req->bio;
1338 req->bio = bio;
1339
1340 /*
1341 * may not be valid. if the low level driver said
1342 * it didn't need a bounce buffer then it better
1343 * not touch req->buffer either...
1344 */
1345 req->buffer = bio_data(bio);
1346 req->__sector = bio->bi_sector;
1347 req->__data_len += bio->bi_size;
1348 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1349
1350 drive_stat_acct(req, 0);
1351 return true;
1352}
1353
Tejun Heobd87b582011-10-19 14:33:08 +02001354/**
1355 * attempt_plug_merge - try to merge with %current's plugged list
1356 * @q: request_queue new bio is being queued at
1357 * @bio: new bio being queued
1358 * @request_count: out parameter for number of traversed plugged requests
1359 *
1360 * Determine whether @bio being queued on @q can be merged with a request
1361 * on %current's plugged list. Returns %true if merge was successful,
1362 * otherwise %false.
1363 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001364 * Plugging coalesces IOs from the same issuer for the same purpose without
1365 * going through @q->queue_lock. As such it's more of an issuing mechanism
1366 * than scheduling, and the request, while may have elvpriv data, is not
1367 * added on the elevator at this point. In addition, we don't have
1368 * reliable access to the elevator outside queue lock. Only check basic
1369 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001370 */
Tejun Heobd87b582011-10-19 14:33:08 +02001371static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1372 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001373{
1374 struct blk_plug *plug;
1375 struct request *rq;
1376 bool ret = false;
1377
Tejun Heobd87b582011-10-19 14:33:08 +02001378 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001379 if (!plug)
1380 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001381 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001382
1383 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1384 int el_ret;
1385
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001386 if (rq->q == q)
1387 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001388
Tejun Heo07c2bd32012-02-08 09:19:42 +01001389 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001390 continue;
1391
Tejun Heo050c8ea2012-02-08 09:19:38 +01001392 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001393 if (el_ret == ELEVATOR_BACK_MERGE) {
1394 ret = bio_attempt_back_merge(q, rq, bio);
1395 if (ret)
1396 break;
1397 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1398 ret = bio_attempt_front_merge(q, rq, bio);
1399 if (ret)
1400 break;
1401 }
1402 }
1403out:
1404 return ret;
1405}
1406
Jens Axboe86db1e22008-01-29 14:53:40 +01001407void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001408{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001409 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001410
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001411 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1412 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001413 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001414
Tejun Heo52d9e672006-01-06 09:49:58 +01001415 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001416 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001417 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001418 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001419}
1420
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001421void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001423 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001424 struct blk_plug *plug;
1425 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1426 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001427 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 /*
1430 * low level driver can indicate that it wants pages above a
1431 * certain limit bounced to low memory (ie for highmem, or even
1432 * ISA dma in theory)
1433 */
1434 blk_queue_bounce(q, &bio);
1435
Tejun Heo4fed9472010-09-03 11:56:17 +02001436 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001437 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001438 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001439 goto get_rq;
1440 }
1441
Jens Axboe73c10102011-03-08 13:19:51 +01001442 /*
1443 * Check if we can merge with the plugged list before grabbing
1444 * any locks.
1445 */
Tejun Heobd87b582011-10-19 14:33:08 +02001446 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001447 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001448
1449 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
1451 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001452 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001453 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001454 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001455 if (!attempt_back_merge(q, req))
1456 elv_merged_request(q, req, el_ret);
1457 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001458 }
Jens Axboe73c10102011-03-08 13:19:51 +01001459 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001460 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001461 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001462 if (!attempt_front_merge(q, req))
1463 elv_merged_request(q, req, el_ret);
1464 goto out_unlock;
1465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 }
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001469 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001470 * This sync check and mask will be re-done in init_request_from_bio(),
1471 * but we need to set it earlier to expose the sync flag to the
1472 * rq allocator and io schedulers.
1473 */
1474 rw_flags = bio_data_dir(bio);
1475 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001476 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001477
1478 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001479 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001480 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001481 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001482 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001483 if (unlikely(!req)) {
1484 bio_endio(bio, -ENODEV); /* @q is dead */
1485 goto out_unlock;
1486 }
Nick Piggind6344532005-06-28 20:45:14 -07001487
Nick Piggin450991b2005-06-28 20:45:13 -07001488 /*
1489 * After dropping the lock and possibly sleeping here, our request
1490 * may now be mergeable after it had proven unmergeable (above).
1491 * We don't worry about that case for efficiency. It won't happen
1492 * often, and the elevators are able to handle it.
1493 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001494 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
Tao Ma9562ad92011-10-24 16:11:30 +02001496 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001497 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001498
Jens Axboe73c10102011-03-08 13:19:51 +01001499 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001500 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001501 /*
1502 * If this is the first request added after a plug, fire
1503 * of a plug trace. If others have been added before, check
1504 * if we have multiple devices in this plug. If so, make a
1505 * note to sort the list before dispatch.
1506 */
1507 if (list_empty(&plug->list))
1508 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001509 else {
1510 if (!plug->should_sort) {
1511 struct request *__rq;
Jens Axboe73c10102011-03-08 13:19:51 +01001512
Shaohua Li3540d5e2011-11-16 09:21:50 +01001513 __rq = list_entry_rq(plug->list.prev);
1514 if (__rq->q != q)
1515 plug->should_sort = 1;
1516 }
Shaohua Li019ceb72011-11-16 09:21:50 +01001517 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001518 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001519 trace_block_plug(q);
1520 }
Jens Axboe73c10102011-03-08 13:19:51 +01001521 }
Shaohua Lia6327162011-08-24 16:04:32 +02001522 list_add_tail(&req->queuelist, &plug->list);
1523 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001524 } else {
1525 spin_lock_irq(q->queue_lock);
1526 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001527 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001528out_unlock:
1529 spin_unlock_irq(q->queue_lock);
1530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
Jens Axboec20e8de2011-09-12 12:03:37 +02001532EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
1534/*
1535 * If bio->bi_dev is a partition, remap the location
1536 */
1537static inline void blk_partition_remap(struct bio *bio)
1538{
1539 struct block_device *bdev = bio->bi_bdev;
1540
Jens Axboebf2de6f2007-09-27 13:01:25 +02001541 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 bio->bi_sector += p->start_sect;
1545 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001546
Mike Snitzerd07335e2010-11-16 12:52:38 +01001547 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1548 bdev->bd_dev,
1549 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 }
1551}
1552
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553static void handle_bad_sector(struct bio *bio)
1554{
1555 char b[BDEVNAME_SIZE];
1556
1557 printk(KERN_INFO "attempt to access beyond end of device\n");
1558 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1559 bdevname(bio->bi_bdev, b),
1560 bio->bi_rw,
1561 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001562 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
1564 set_bit(BIO_EOF, &bio->bi_flags);
1565}
1566
Akinobu Mitac17bb492006-12-08 02:39:46 -08001567#ifdef CONFIG_FAIL_MAKE_REQUEST
1568
1569static DECLARE_FAULT_ATTR(fail_make_request);
1570
1571static int __init setup_fail_make_request(char *str)
1572{
1573 return setup_fault_attr(&fail_make_request, str);
1574}
1575__setup("fail_make_request=", setup_fail_make_request);
1576
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001577static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001578{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001579 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001580}
1581
1582static int __init fail_make_request_debugfs(void)
1583{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001584 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1585 NULL, &fail_make_request);
1586
1587 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001588}
1589
1590late_initcall(fail_make_request_debugfs);
1591
1592#else /* CONFIG_FAIL_MAKE_REQUEST */
1593
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001594static inline bool should_fail_request(struct hd_struct *part,
1595 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001596{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001597 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001598}
1599
1600#endif /* CONFIG_FAIL_MAKE_REQUEST */
1601
Jens Axboec07e2b42007-07-18 13:27:58 +02001602/*
1603 * Check whether this bio extends beyond the end of the device.
1604 */
1605static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1606{
1607 sector_t maxsector;
1608
1609 if (!nr_sectors)
1610 return 0;
1611
1612 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001613 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001614 if (maxsector) {
1615 sector_t sector = bio->bi_sector;
1616
1617 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1618 /*
1619 * This may well happen - the kernel calls bread()
1620 * without checking the size of the device, e.g., when
1621 * mounting a device.
1622 */
1623 handle_bad_sector(bio);
1624 return 1;
1625 }
1626 }
1627
1628 return 0;
1629}
1630
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001631static noinline_for_stack bool
1632generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633{
Jens Axboe165125e2007-07-24 09:28:11 +02001634 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001635 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001636 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001637 char b[BDEVNAME_SIZE];
1638 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
1640 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Jens Axboec07e2b42007-07-18 13:27:58 +02001642 if (bio_check_eod(bio, nr_sectors))
1643 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001645 q = bdev_get_queue(bio->bi_bdev);
1646 if (unlikely(!q)) {
1647 printk(KERN_ERR
1648 "generic_make_request: Trying to access "
1649 "nonexistent block-device %s (%Lu)\n",
1650 bdevname(bio->bi_bdev, b),
1651 (long long) bio->bi_sector);
1652 goto end_io;
1653 }
1654
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001655 if (likely(bio_is_rw(bio) &&
1656 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001657 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1658 bdevname(bio->bi_bdev, b),
1659 bio_sectors(bio),
1660 queue_max_hw_sectors(q));
1661 goto end_io;
1662 }
1663
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001664 part = bio->bi_bdev->bd_part;
1665 if (should_fail_request(part, bio->bi_size) ||
1666 should_fail_request(&part_to_disk(part)->part0,
1667 bio->bi_size))
1668 goto end_io;
1669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001671 * If this device has partitions, remap block n
1672 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001674 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001676 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1677 goto end_io;
1678
1679 if (bio_check_eod(bio, nr_sectors))
1680 goto end_io;
1681
1682 /*
1683 * Filter flush bio's early so that make_request based
1684 * drivers without flush support don't have to worry
1685 * about them.
1686 */
1687 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1688 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1689 if (!nr_sectors) {
1690 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001691 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001695 if ((bio->bi_rw & REQ_DISCARD) &&
1696 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001697 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001698 err = -EOPNOTSUPP;
1699 goto end_io;
1700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001702 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 err = -EOPNOTSUPP;
1704 goto end_io;
1705 }
1706
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001707 /*
1708 * Various block parts want %current->io_context and lazy ioc
1709 * allocation ends up trading a lot of pain for a small amount of
1710 * memory. Just allocate it upfront. This may fail and block
1711 * layer knows how to live with it.
1712 */
1713 create_io_context(GFP_ATOMIC, q->node);
1714
Tejun Heobc16a4f2011-10-19 14:33:01 +02001715 if (blk_throtl_bio(q, bio))
1716 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001717
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001718 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001719 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001720
1721end_io:
1722 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001723 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001726/**
1727 * generic_make_request - hand a buffer to its device driver for I/O
1728 * @bio: The bio describing the location in memory and on the device.
1729 *
1730 * generic_make_request() is used to make I/O requests of block
1731 * devices. It is passed a &struct bio, which describes the I/O that needs
1732 * to be done.
1733 *
1734 * generic_make_request() does not return any status. The
1735 * success/failure status of the request, along with notification of
1736 * completion, is delivered asynchronously through the bio->bi_end_io
1737 * function described (one day) else where.
1738 *
1739 * The caller of generic_make_request must make sure that bi_io_vec
1740 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1741 * set to describe the device address, and the
1742 * bi_end_io and optionally bi_private are set to describe how
1743 * completion notification should be signaled.
1744 *
1745 * generic_make_request and the drivers it calls may use bi_next if this
1746 * bio happens to be merged with someone else, and may resubmit the bio to
1747 * a lower device by calling into generic_make_request recursively, which
1748 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001749 */
1750void generic_make_request(struct bio *bio)
1751{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001752 struct bio_list bio_list_on_stack;
1753
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001754 if (!generic_make_request_checks(bio))
1755 return;
1756
1757 /*
1758 * We only want one ->make_request_fn to be active at a time, else
1759 * stack usage with stacked devices could be a problem. So use
1760 * current->bio_list to keep a list of requests submited by a
1761 * make_request_fn function. current->bio_list is also used as a
1762 * flag to say if generic_make_request is currently active in this
1763 * task or not. If it is NULL, then no make_request is active. If
1764 * it is non-NULL, then a make_request is active, and new requests
1765 * should be added at the tail
1766 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001767 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001768 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001769 return;
1770 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001771
Neil Brownd89d8792007-05-01 09:53:42 +02001772 /* following loop may be a bit non-obvious, and so deserves some
1773 * explanation.
1774 * Before entering the loop, bio->bi_next is NULL (as all callers
1775 * ensure that) so we have a list with a single bio.
1776 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001777 * we assign bio_list to a pointer to the bio_list_on_stack,
1778 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001779 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001780 * through a recursive call to generic_make_request. If it
1781 * did, we find a non-NULL value in bio_list and re-enter the loop
1782 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001783 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001784 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001785 */
1786 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001787 bio_list_init(&bio_list_on_stack);
1788 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001789 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001790 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1791
1792 q->make_request_fn(q, bio);
1793
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001794 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001795 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001796 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001797}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798EXPORT_SYMBOL(generic_make_request);
1799
1800/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001801 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1803 * @bio: The &struct bio which describes the I/O
1804 *
1805 * submit_bio() is very similar in purpose to generic_make_request(), and
1806 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001807 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 *
1809 */
1810void submit_bio(int rw, struct bio *bio)
1811{
Jens Axboe22e2c502005-06-27 10:55:12 +02001812 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Jens Axboebf2de6f2007-09-27 13:01:25 +02001814 /*
1815 * If it's a regular read/write or a barrier with data attached,
1816 * go through the normal accounting stuff before submission.
1817 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001818 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001819 unsigned int count;
1820
1821 if (unlikely(rw & REQ_WRITE_SAME))
1822 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1823 else
1824 count = bio_sectors(bio);
1825
Jens Axboebf2de6f2007-09-27 13:01:25 +02001826 if (rw & WRITE) {
1827 count_vm_events(PGPGOUT, count);
1828 } else {
1829 task_io_account_read(bio->bi_size);
1830 count_vm_events(PGPGIN, count);
1831 }
1832
1833 if (unlikely(block_dump)) {
1834 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001835 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001836 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001837 (rw & WRITE) ? "WRITE" : "READ",
1838 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001839 bdevname(bio->bi_bdev, b),
1840 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 }
1843
1844 generic_make_request(bio);
1845}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846EXPORT_SYMBOL(submit_bio);
1847
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001848/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001849 * blk_rq_check_limits - Helper function to check a request for the queue limit
1850 * @q: the queue
1851 * @rq: the request being checked
1852 *
1853 * Description:
1854 * @rq may have been made based on weaker limitations of upper-level queues
1855 * in request stacking drivers, and it may violate the limitation of @q.
1856 * Since the block layer and the underlying device driver trust @rq
1857 * after it is inserted to @q, it should be checked against @q before
1858 * the insertion using this generic function.
1859 *
1860 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001861 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001862 * Request stacking drivers like request-based dm may change the queue
1863 * limits while requests are in the queue (e.g. dm's table swapping).
1864 * Such request stacking drivers should check those requests agaist
1865 * the new queue limits again when they dispatch those requests,
1866 * although such checkings are also done against the old queue limits
1867 * when submitting requests.
1868 */
1869int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1870{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001871 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001872 return 0;
1873
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001874 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001875 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1876 return -EIO;
1877 }
1878
1879 /*
1880 * queue's settings related to segment counting like q->bounce_pfn
1881 * may differ from that of other stacking queues.
1882 * Recalculate it to check the request correctly on this queue's
1883 * limitation.
1884 */
1885 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001886 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001887 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1888 return -EIO;
1889 }
1890
1891 return 0;
1892}
1893EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1894
1895/**
1896 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1897 * @q: the queue to submit the request
1898 * @rq: the request being queued
1899 */
1900int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1901{
1902 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001903 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001904
1905 if (blk_rq_check_limits(q, rq))
1906 return -EIO;
1907
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001908 if (rq->rq_disk &&
1909 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001910 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001911
1912 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo8ba61432011-12-14 00:33:37 +01001913 if (unlikely(blk_queue_dead(q))) {
1914 spin_unlock_irqrestore(q->queue_lock, flags);
1915 return -ENODEV;
1916 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001917
1918 /*
1919 * Submitting request must be dequeued before calling this function
1920 * because it will be linked to another request_queue
1921 */
1922 BUG_ON(blk_queued_rq(rq));
1923
Jeff Moyer4853aba2011-08-15 21:37:25 +02001924 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1925 where = ELEVATOR_INSERT_FLUSH;
1926
1927 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001928 if (where == ELEVATOR_INSERT_FLUSH)
1929 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001930 spin_unlock_irqrestore(q->queue_lock, flags);
1931
1932 return 0;
1933}
1934EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1935
Tejun Heo80a761f2009-07-03 17:48:17 +09001936/**
1937 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1938 * @rq: request to examine
1939 *
1940 * Description:
1941 * A request could be merge of IOs which require different failure
1942 * handling. This function determines the number of bytes which
1943 * can be failed from the beginning of the request without
1944 * crossing into area which need to be retried further.
1945 *
1946 * Return:
1947 * The number of bytes to fail.
1948 *
1949 * Context:
1950 * queue_lock must be held.
1951 */
1952unsigned int blk_rq_err_bytes(const struct request *rq)
1953{
1954 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1955 unsigned int bytes = 0;
1956 struct bio *bio;
1957
1958 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1959 return blk_rq_bytes(rq);
1960
1961 /*
1962 * Currently the only 'mixing' which can happen is between
1963 * different fastfail types. We can safely fail portions
1964 * which have all the failfast bits that the first one has -
1965 * the ones which are at least as eager to fail as the first
1966 * one.
1967 */
1968 for (bio = rq->bio; bio; bio = bio->bi_next) {
1969 if ((bio->bi_rw & ff) != ff)
1970 break;
1971 bytes += bio->bi_size;
1972 }
1973
1974 /* this could lead to infinite loop */
1975 BUG_ON(blk_rq_bytes(rq) && !bytes);
1976 return bytes;
1977}
1978EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1979
Jens Axboebc58ba92009-01-23 10:54:44 +01001980static void blk_account_io_completion(struct request *req, unsigned int bytes)
1981{
Jens Axboec2553b52009-04-24 08:10:11 +02001982 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01001983 const int rw = rq_data_dir(req);
1984 struct hd_struct *part;
1985 int cpu;
1986
1987 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001988 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001989 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1990 part_stat_unlock();
1991 }
1992}
1993
1994static void blk_account_io_done(struct request *req)
1995{
Jens Axboebc58ba92009-01-23 10:54:44 +01001996 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001997 * Account IO completion. flush_rq isn't accounted as a
1998 * normal IO on queueing nor completion. Accounting the
1999 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002000 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002001 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002002 unsigned long duration = jiffies - req->start_time;
2003 const int rw = rq_data_dir(req);
2004 struct hd_struct *part;
2005 int cpu;
2006
2007 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002008 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002009
2010 part_stat_inc(cpu, part, ios[rw]);
2011 part_stat_add(cpu, part, ticks[rw], duration);
2012 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002013 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002014
Jens Axboe6c23a962011-01-07 08:43:37 +01002015 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002016 part_stat_unlock();
2017 }
2018}
2019
Tejun Heo53a08802008-12-03 12:41:26 +01002020/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002021 * blk_peek_request - peek at the top of a request queue
2022 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002023 *
2024 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002025 * Return the request at the top of @q. The returned request
2026 * should be started using blk_start_request() before LLD starts
2027 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002028 *
2029 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002030 * Pointer to the request at the top of @q if available. Null
2031 * otherwise.
2032 *
2033 * Context:
2034 * queue_lock must be held.
2035 */
2036struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002037{
2038 struct request *rq;
2039 int ret;
2040
2041 while ((rq = __elv_next_request(q)) != NULL) {
2042 if (!(rq->cmd_flags & REQ_STARTED)) {
2043 /*
2044 * This is the first time the device driver
2045 * sees this request (possibly after
2046 * requeueing). Notify IO scheduler.
2047 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002048 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002049 elv_activate_rq(q, rq);
2050
2051 /*
2052 * just mark as started even if we don't start
2053 * it, a request that has been delayed should
2054 * not be passed by new incoming requests
2055 */
2056 rq->cmd_flags |= REQ_STARTED;
2057 trace_block_rq_issue(q, rq);
2058 }
2059
2060 if (!q->boundary_rq || q->boundary_rq == rq) {
2061 q->end_sector = rq_end_sector(rq);
2062 q->boundary_rq = NULL;
2063 }
2064
2065 if (rq->cmd_flags & REQ_DONTPREP)
2066 break;
2067
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002068 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002069 /*
2070 * make sure space for the drain appears we
2071 * know we can do this because max_hw_segments
2072 * has been adjusted to be one fewer than the
2073 * device can handle
2074 */
2075 rq->nr_phys_segments++;
2076 }
2077
2078 if (!q->prep_rq_fn)
2079 break;
2080
2081 ret = q->prep_rq_fn(q, rq);
2082 if (ret == BLKPREP_OK) {
2083 break;
2084 } else if (ret == BLKPREP_DEFER) {
2085 /*
2086 * the request may have been (partially) prepped.
2087 * we need to keep this request in the front to
2088 * avoid resource deadlock. REQ_STARTED will
2089 * prevent other fs requests from passing this one.
2090 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002091 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002092 !(rq->cmd_flags & REQ_DONTPREP)) {
2093 /*
2094 * remove the space for the drain we added
2095 * so that we don't add it again
2096 */
2097 --rq->nr_phys_segments;
2098 }
2099
2100 rq = NULL;
2101 break;
2102 } else if (ret == BLKPREP_KILL) {
2103 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002104 /*
2105 * Mark this request as started so we don't trigger
2106 * any debug logic in the end I/O path.
2107 */
2108 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002109 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002110 } else {
2111 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2112 break;
2113 }
2114 }
2115
2116 return rq;
2117}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002118EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002119
Tejun Heo9934c8c2009-05-08 11:54:16 +09002120void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002121{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002122 struct request_queue *q = rq->q;
2123
Tejun Heo158dbda2009-04-23 11:05:18 +09002124 BUG_ON(list_empty(&rq->queuelist));
2125 BUG_ON(ELV_ON_HASH(rq));
2126
2127 list_del_init(&rq->queuelist);
2128
2129 /*
2130 * the time frame between a request being removed from the lists
2131 * and to it is freed is accounted as io that is in progress at
2132 * the driver side.
2133 */
Divyesh Shah91952912010-04-01 15:01:41 -07002134 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002135 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002136 set_io_start_time_ns(rq);
2137 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002138}
2139
Tejun Heo5efccd12009-04-23 11:05:18 +09002140/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002141 * blk_start_request - start request processing on the driver
2142 * @req: request to dequeue
2143 *
2144 * Description:
2145 * Dequeue @req and start timeout timer on it. This hands off the
2146 * request to the driver.
2147 *
2148 * Block internal functions which don't want to start timer should
2149 * call blk_dequeue_request().
2150 *
2151 * Context:
2152 * queue_lock must be held.
2153 */
2154void blk_start_request(struct request *req)
2155{
2156 blk_dequeue_request(req);
2157
2158 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002159 * We are now handing the request to the hardware, initialize
2160 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002161 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002162 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002163 if (unlikely(blk_bidi_rq(req)))
2164 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2165
Tejun Heo9934c8c2009-05-08 11:54:16 +09002166 blk_add_timer(req);
2167}
2168EXPORT_SYMBOL(blk_start_request);
2169
2170/**
2171 * blk_fetch_request - fetch a request from a request queue
2172 * @q: request queue to fetch a request from
2173 *
2174 * Description:
2175 * Return the request at the top of @q. The request is started on
2176 * return and LLD can start processing it immediately.
2177 *
2178 * Return:
2179 * Pointer to the request at the top of @q if available. Null
2180 * otherwise.
2181 *
2182 * Context:
2183 * queue_lock must be held.
2184 */
2185struct request *blk_fetch_request(struct request_queue *q)
2186{
2187 struct request *rq;
2188
2189 rq = blk_peek_request(q);
2190 if (rq)
2191 blk_start_request(rq);
2192 return rq;
2193}
2194EXPORT_SYMBOL(blk_fetch_request);
2195
2196/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002197 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002198 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002199 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002200 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002201 *
2202 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002203 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2204 * the request structure even if @req doesn't have leftover.
2205 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002206 *
2207 * This special helper function is only for request stacking drivers
2208 * (e.g. request-based dm) so that they can handle partial completion.
2209 * Actual device drivers should use blk_end_request instead.
2210 *
2211 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2212 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002213 *
2214 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002215 * %false - this request doesn't have any more data
2216 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002217 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002218bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002220 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 struct bio *bio;
2222
Tejun Heo2e60e022009-04-23 11:05:18 +09002223 if (!req->bio)
2224 return false;
2225
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002226 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002229 * For fs requests, rq is just carrier of independent bio's
2230 * and each partial completion should be handled separately.
2231 * Reset per-request error on each partial completion.
2232 *
2233 * TODO: tj: This is too subtle. It would be better to let
2234 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002236 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 req->errors = 0;
2238
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002239 if (error && req->cmd_type == REQ_TYPE_FS &&
2240 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002241 char *error_type;
2242
2243 switch (error) {
2244 case -ENOLINK:
2245 error_type = "recoverable transport";
2246 break;
2247 case -EREMOTEIO:
2248 error_type = "critical target";
2249 break;
2250 case -EBADE:
2251 error_type = "critical nexus";
2252 break;
2253 case -EIO:
2254 default:
2255 error_type = "I/O";
2256 break;
2257 }
Yi Zou37d7b342012-08-30 16:26:25 -07002258 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2259 error_type, req->rq_disk ?
2260 req->rq_disk->disk_name : "?",
2261 (unsigned long long)blk_rq_pos(req));
2262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 }
2264
Jens Axboebc58ba92009-01-23 10:54:44 +01002265 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002266
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 total_bytes = bio_nbytes = 0;
2268 while ((bio = req->bio) != NULL) {
2269 int nbytes;
2270
2271 if (nr_bytes >= bio->bi_size) {
2272 req->bio = bio->bi_next;
2273 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002274 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 next_idx = 0;
2276 bio_nbytes = 0;
2277 } else {
2278 int idx = bio->bi_idx + next_idx;
2279
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002280 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002282 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002283 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 break;
2285 }
2286
2287 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2288 BIO_BUG_ON(nbytes > bio->bi_size);
2289
2290 /*
2291 * not a complete bvec done
2292 */
2293 if (unlikely(nbytes > nr_bytes)) {
2294 bio_nbytes += nr_bytes;
2295 total_bytes += nr_bytes;
2296 break;
2297 }
2298
2299 /*
2300 * advance to the next vector
2301 */
2302 next_idx++;
2303 bio_nbytes += nbytes;
2304 }
2305
2306 total_bytes += nbytes;
2307 nr_bytes -= nbytes;
2308
Jens Axboe6728cb02008-01-31 13:03:55 +01002309 bio = req->bio;
2310 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 /*
2312 * end more in this run, or just return 'not-done'
2313 */
2314 if (unlikely(nr_bytes <= 0))
2315 break;
2316 }
2317 }
2318
2319 /*
2320 * completely done
2321 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002322 if (!req->bio) {
2323 /*
2324 * Reset counters so that the request stacking driver
2325 * can find how many bytes remain in the request
2326 * later.
2327 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002328 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002329 return false;
2330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
2332 /*
2333 * if the request wasn't completed, update state
2334 */
2335 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002336 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 bio->bi_idx += next_idx;
2338 bio_iovec(bio)->bv_offset += nr_bytes;
2339 bio_iovec(bio)->bv_len -= nr_bytes;
2340 }
2341
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002342 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002343 req->buffer = bio_data(req->bio);
2344
2345 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002346 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002347 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002348
Tejun Heo80a761f2009-07-03 17:48:17 +09002349 /* mixed attributes always follow the first bio */
2350 if (req->cmd_flags & REQ_MIXED_MERGE) {
2351 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2352 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2353 }
2354
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002355 /*
2356 * If total number of sectors is less than the first segment
2357 * size, something has gone terribly wrong.
2358 */
2359 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002360 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002361 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002362 }
2363
2364 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002366
Tejun Heo2e60e022009-04-23 11:05:18 +09002367 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368}
Tejun Heo2e60e022009-04-23 11:05:18 +09002369EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370
Tejun Heo2e60e022009-04-23 11:05:18 +09002371static bool blk_update_bidi_request(struct request *rq, int error,
2372 unsigned int nr_bytes,
2373 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002374{
Tejun Heo2e60e022009-04-23 11:05:18 +09002375 if (blk_update_request(rq, error, nr_bytes))
2376 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002377
Tejun Heo2e60e022009-04-23 11:05:18 +09002378 /* Bidi request must be completed as a whole */
2379 if (unlikely(blk_bidi_rq(rq)) &&
2380 blk_update_request(rq->next_rq, error, bidi_bytes))
2381 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002382
Jens Axboee2e1a142010-06-09 10:42:09 +02002383 if (blk_queue_add_random(rq->q))
2384 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002385
2386 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387}
2388
James Bottomley28018c22010-07-01 19:49:17 +09002389/**
2390 * blk_unprep_request - unprepare a request
2391 * @req: the request
2392 *
2393 * This function makes a request ready for complete resubmission (or
2394 * completion). It happens only after all error handling is complete,
2395 * so represents the appropriate moment to deallocate any resources
2396 * that were allocated to the request in the prep_rq_fn. The queue
2397 * lock is held when calling this.
2398 */
2399void blk_unprep_request(struct request *req)
2400{
2401 struct request_queue *q = req->q;
2402
2403 req->cmd_flags &= ~REQ_DONTPREP;
2404 if (q->unprep_rq_fn)
2405 q->unprep_rq_fn(q, req);
2406}
2407EXPORT_SYMBOL_GPL(blk_unprep_request);
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409/*
2410 * queue lock must be held
2411 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002412static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002414 if (blk_rq_tagged(req))
2415 blk_queue_end_tag(req->q, req);
2416
James Bottomleyba396a62009-05-27 14:17:08 +02002417 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002419 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002420 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Mike Andersone78042e2008-10-30 02:16:20 -07002422 blk_delete_timer(req);
2423
James Bottomley28018c22010-07-01 19:49:17 +09002424 if (req->cmd_flags & REQ_DONTPREP)
2425 blk_unprep_request(req);
2426
2427
Jens Axboebc58ba92009-01-23 10:54:44 +01002428 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002431 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002432 else {
2433 if (blk_bidi_rq(req))
2434 __blk_put_request(req->next_rq->q, req->next_rq);
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 }
2438}
2439
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002440/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002441 * blk_end_bidi_request - Complete a bidi request
2442 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002443 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002444 * @nr_bytes: number of bytes to complete @rq
2445 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002446 *
2447 * Description:
2448 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002449 * Drivers that supports bidi can safely call this member for any
2450 * type of request, bidi or uni. In the later case @bidi_bytes is
2451 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002452 *
2453 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002454 * %false - we are done with this request
2455 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002456 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002457static bool blk_end_bidi_request(struct request *rq, int error,
2458 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002459{
2460 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002461 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002462
Tejun Heo2e60e022009-04-23 11:05:18 +09002463 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2464 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002465
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002466 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002467 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002468 spin_unlock_irqrestore(q->queue_lock, flags);
2469
Tejun Heo2e60e022009-04-23 11:05:18 +09002470 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002471}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002472
2473/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002474 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2475 * @rq: the request to complete
2476 * @error: %0 for success, < %0 for error
2477 * @nr_bytes: number of bytes to complete @rq
2478 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002479 *
2480 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002481 * Identical to blk_end_bidi_request() except that queue lock is
2482 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002483 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002484 * Return:
2485 * %false - we are done with this request
2486 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002487 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002488bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002489 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002490{
Tejun Heo2e60e022009-04-23 11:05:18 +09002491 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2492 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002493
Tejun Heo2e60e022009-04-23 11:05:18 +09002494 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002495
Tejun Heo2e60e022009-04-23 11:05:18 +09002496 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002497}
2498
2499/**
2500 * blk_end_request - Helper function for drivers to complete the request.
2501 * @rq: the request being processed
2502 * @error: %0 for success, < %0 for error
2503 * @nr_bytes: number of bytes to complete
2504 *
2505 * Description:
2506 * Ends I/O on a number of bytes attached to @rq.
2507 * If @rq has leftover, sets it up for the next range of segments.
2508 *
2509 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002510 * %false - we are done with this request
2511 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002512 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002513bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002514{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002515 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002516}
Jens Axboe56ad1742009-07-28 22:11:24 +02002517EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002518
2519/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002520 * blk_end_request_all - Helper function for drives to finish the request.
2521 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002522 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002523 *
2524 * Description:
2525 * Completely finish @rq.
2526 */
2527void blk_end_request_all(struct request *rq, int error)
2528{
2529 bool pending;
2530 unsigned int bidi_bytes = 0;
2531
2532 if (unlikely(blk_bidi_rq(rq)))
2533 bidi_bytes = blk_rq_bytes(rq->next_rq);
2534
2535 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2536 BUG_ON(pending);
2537}
Jens Axboe56ad1742009-07-28 22:11:24 +02002538EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002539
2540/**
2541 * blk_end_request_cur - Helper function to finish the current request chunk.
2542 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002543 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002544 *
2545 * Description:
2546 * Complete the current consecutively mapped chunk from @rq.
2547 *
2548 * Return:
2549 * %false - we are done with this request
2550 * %true - still buffers pending for this request
2551 */
2552bool blk_end_request_cur(struct request *rq, int error)
2553{
2554 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2555}
Jens Axboe56ad1742009-07-28 22:11:24 +02002556EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002557
2558/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002559 * blk_end_request_err - Finish a request till the next failure boundary.
2560 * @rq: the request to finish till the next failure boundary for
2561 * @error: must be negative errno
2562 *
2563 * Description:
2564 * Complete @rq till the next failure boundary.
2565 *
2566 * Return:
2567 * %false - we are done with this request
2568 * %true - still buffers pending for this request
2569 */
2570bool blk_end_request_err(struct request *rq, int error)
2571{
2572 WARN_ON(error >= 0);
2573 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2574}
2575EXPORT_SYMBOL_GPL(blk_end_request_err);
2576
2577/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002578 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002579 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002580 * @error: %0 for success, < %0 for error
2581 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002582 *
2583 * Description:
2584 * Must be called with queue lock held unlike blk_end_request().
2585 *
2586 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002587 * %false - we are done with this request
2588 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002589 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002590bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002591{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002592 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002593}
Jens Axboe56ad1742009-07-28 22:11:24 +02002594EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002595
2596/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002597 * __blk_end_request_all - Helper function for drives to finish the request.
2598 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002599 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002600 *
2601 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002602 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002603 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002604void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002605{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002606 bool pending;
2607 unsigned int bidi_bytes = 0;
2608
2609 if (unlikely(blk_bidi_rq(rq)))
2610 bidi_bytes = blk_rq_bytes(rq->next_rq);
2611
2612 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2613 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002614}
Jens Axboe56ad1742009-07-28 22:11:24 +02002615EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002616
2617/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002618 * __blk_end_request_cur - Helper function to finish the current request chunk.
2619 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002620 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002621 *
2622 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002623 * Complete the current consecutively mapped chunk from @rq. Must
2624 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002625 *
2626 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002627 * %false - we are done with this request
2628 * %true - still buffers pending for this request
2629 */
2630bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002631{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002632 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002633}
Jens Axboe56ad1742009-07-28 22:11:24 +02002634EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002635
Tejun Heo80a761f2009-07-03 17:48:17 +09002636/**
2637 * __blk_end_request_err - Finish a request till the next failure boundary.
2638 * @rq: the request to finish till the next failure boundary for
2639 * @error: must be negative errno
2640 *
2641 * Description:
2642 * Complete @rq till the next failure boundary. Must be called
2643 * with queue lock held.
2644 *
2645 * Return:
2646 * %false - we are done with this request
2647 * %true - still buffers pending for this request
2648 */
2649bool __blk_end_request_err(struct request *rq, int error)
2650{
2651 WARN_ON(error >= 0);
2652 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2653}
2654EXPORT_SYMBOL_GPL(__blk_end_request_err);
2655
Jens Axboe86db1e22008-01-29 14:53:40 +01002656void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2657 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002659 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002660 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
David Woodhousefb2dce82008-08-05 18:01:53 +01002662 if (bio_has_data(bio)) {
2663 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002664 rq->buffer = bio_data(bio);
2665 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002666 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
NeilBrown66846572007-08-16 13:31:28 +02002669 if (bio->bi_bdev)
2670 rq->rq_disk = bio->bi_bdev->bd_disk;
2671}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002673#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2674/**
2675 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2676 * @rq: the request to be flushed
2677 *
2678 * Description:
2679 * Flush all pages in @rq.
2680 */
2681void rq_flush_dcache_pages(struct request *rq)
2682{
2683 struct req_iterator iter;
2684 struct bio_vec *bvec;
2685
2686 rq_for_each_segment(bvec, rq, iter)
2687 flush_dcache_page(bvec->bv_page);
2688}
2689EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2690#endif
2691
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002692/**
2693 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2694 * @q : the queue of the device being checked
2695 *
2696 * Description:
2697 * Check if underlying low-level drivers of a device are busy.
2698 * If the drivers want to export their busy state, they must set own
2699 * exporting function using blk_queue_lld_busy() first.
2700 *
2701 * Basically, this function is used only by request stacking drivers
2702 * to stop dispatching requests to underlying devices when underlying
2703 * devices are busy. This behavior helps more I/O merging on the queue
2704 * of the request stacking driver and prevents I/O throughput regression
2705 * on burst I/O load.
2706 *
2707 * Return:
2708 * 0 - Not busy (The request stacking driver should dispatch request)
2709 * 1 - Busy (The request stacking driver should stop dispatching request)
2710 */
2711int blk_lld_busy(struct request_queue *q)
2712{
2713 if (q->lld_busy_fn)
2714 return q->lld_busy_fn(q);
2715
2716 return 0;
2717}
2718EXPORT_SYMBOL_GPL(blk_lld_busy);
2719
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002720/**
2721 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2722 * @rq: the clone request to be cleaned up
2723 *
2724 * Description:
2725 * Free all bios in @rq for a cloned request.
2726 */
2727void blk_rq_unprep_clone(struct request *rq)
2728{
2729 struct bio *bio;
2730
2731 while ((bio = rq->bio) != NULL) {
2732 rq->bio = bio->bi_next;
2733
2734 bio_put(bio);
2735 }
2736}
2737EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2738
2739/*
2740 * Copy attributes of the original request to the clone request.
2741 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2742 */
2743static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2744{
2745 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002746 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002747 dst->cmd_type = src->cmd_type;
2748 dst->__sector = blk_rq_pos(src);
2749 dst->__data_len = blk_rq_bytes(src);
2750 dst->nr_phys_segments = src->nr_phys_segments;
2751 dst->ioprio = src->ioprio;
2752 dst->extra_len = src->extra_len;
2753}
2754
2755/**
2756 * blk_rq_prep_clone - Helper function to setup clone request
2757 * @rq: the request to be setup
2758 * @rq_src: original request to be cloned
2759 * @bs: bio_set that bios for clone are allocated from
2760 * @gfp_mask: memory allocation mask for bio
2761 * @bio_ctr: setup function to be called for each clone bio.
2762 * Returns %0 for success, non %0 for failure.
2763 * @data: private data to be passed to @bio_ctr
2764 *
2765 * Description:
2766 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2767 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2768 * are not copied, and copying such parts is the caller's responsibility.
2769 * Also, pages which the original bios are pointing to are not copied
2770 * and the cloned bios just point same pages.
2771 * So cloned bios must be completed before original bios, which means
2772 * the caller must complete @rq before @rq_src.
2773 */
2774int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2775 struct bio_set *bs, gfp_t gfp_mask,
2776 int (*bio_ctr)(struct bio *, struct bio *, void *),
2777 void *data)
2778{
2779 struct bio *bio, *bio_src;
2780
2781 if (!bs)
2782 bs = fs_bio_set;
2783
2784 blk_rq_init(NULL, rq);
2785
2786 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002787 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002788 if (!bio)
2789 goto free_and_out;
2790
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002791 if (bio_ctr && bio_ctr(bio, bio_src, data))
2792 goto free_and_out;
2793
2794 if (rq->bio) {
2795 rq->biotail->bi_next = bio;
2796 rq->biotail = bio;
2797 } else
2798 rq->bio = rq->biotail = bio;
2799 }
2800
2801 __blk_rq_prep_clone(rq, rq_src);
2802
2803 return 0;
2804
2805free_and_out:
2806 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002807 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002808 blk_rq_unprep_clone(rq);
2809
2810 return -ENOMEM;
2811}
2812EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2813
Jens Axboe18887ad2008-07-28 13:08:45 +02002814int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815{
2816 return queue_work(kblockd_workqueue, work);
2817}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818EXPORT_SYMBOL(kblockd_schedule_work);
2819
Vivek Goyale43473b2010-09-15 17:06:35 -04002820int kblockd_schedule_delayed_work(struct request_queue *q,
2821 struct delayed_work *dwork, unsigned long delay)
2822{
2823 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2824}
2825EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2826
Jens Axboe73c10102011-03-08 13:19:51 +01002827#define PLUG_MAGIC 0x91827364
2828
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002829/**
2830 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2831 * @plug: The &struct blk_plug that needs to be initialized
2832 *
2833 * Description:
2834 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2835 * pending I/O should the task end up blocking between blk_start_plug() and
2836 * blk_finish_plug(). This is important from a performance perspective, but
2837 * also ensures that we don't deadlock. For instance, if the task is blocking
2838 * for a memory allocation, memory reclaim could end up wanting to free a
2839 * page belonging to that request that is currently residing in our private
2840 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2841 * this kind of deadlock.
2842 */
Jens Axboe73c10102011-03-08 13:19:51 +01002843void blk_start_plug(struct blk_plug *plug)
2844{
2845 struct task_struct *tsk = current;
2846
2847 plug->magic = PLUG_MAGIC;
2848 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002849 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002850 plug->should_sort = 0;
2851
2852 /*
2853 * If this is a nested plug, don't actually assign it. It will be
2854 * flushed on its own.
2855 */
2856 if (!tsk->plug) {
2857 /*
2858 * Store ordering should not be needed here, since a potential
2859 * preempt will imply a full memory barrier
2860 */
2861 tsk->plug = plug;
2862 }
2863}
2864EXPORT_SYMBOL(blk_start_plug);
2865
2866static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2867{
2868 struct request *rqa = container_of(a, struct request, queuelist);
2869 struct request *rqb = container_of(b, struct request, queuelist);
2870
Konstantin Khlebnikovf83e8262011-04-04 00:15:02 +02002871 return !(rqa->q <= rqb->q);
Jens Axboe73c10102011-03-08 13:19:51 +01002872}
2873
Jens Axboe49cac012011-04-16 13:51:05 +02002874/*
2875 * If 'from_schedule' is true, then postpone the dispatch of requests
2876 * until a safe kblockd context. We due this to avoid accidental big
2877 * additional stack usage in driver dispatch, in places where the originally
2878 * plugger did not intend it.
2879 */
Jens Axboef6603782011-04-15 15:49:07 +02002880static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002881 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002882 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002883{
Jens Axboe49cac012011-04-16 13:51:05 +02002884 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002885
2886 /*
Tejun Heo8ba61432011-12-14 00:33:37 +01002887 * Don't mess with dead queue.
2888 */
2889 if (unlikely(blk_queue_dead(q))) {
2890 spin_unlock(q->queue_lock);
2891 return;
2892 }
2893
2894 /*
Jens Axboe99e22592011-04-18 09:59:55 +02002895 * If we are punting this to kblockd, then we can safely drop
2896 * the queue_lock before waking kblockd (which needs to take
2897 * this lock).
2898 */
2899 if (from_schedule) {
2900 spin_unlock(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002901 blk_run_queue_async(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002902 } else {
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002903 __blk_run_queue(q);
Jens Axboe99e22592011-04-18 09:59:55 +02002904 spin_unlock(q->queue_lock);
2905 }
2906
Jens Axboe94b5eb22011-04-12 10:12:19 +02002907}
2908
NeilBrown74018dc2012-07-31 09:08:15 +02002909static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002910{
2911 LIST_HEAD(callbacks);
2912
Shaohua Li2a7d5552012-07-31 09:08:15 +02002913 while (!list_empty(&plug->cb_list)) {
2914 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002915
Shaohua Li2a7d5552012-07-31 09:08:15 +02002916 while (!list_empty(&callbacks)) {
2917 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002918 struct blk_plug_cb,
2919 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002920 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02002921 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002922 }
NeilBrown048c9372011-04-18 09:52:22 +02002923 }
2924}
2925
NeilBrown9cbb1752012-07-31 09:08:14 +02002926struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2927 int size)
2928{
2929 struct blk_plug *plug = current->plug;
2930 struct blk_plug_cb *cb;
2931
2932 if (!plug)
2933 return NULL;
2934
2935 list_for_each_entry(cb, &plug->cb_list, list)
2936 if (cb->callback == unplug && cb->data == data)
2937 return cb;
2938
2939 /* Not currently on the callback list */
2940 BUG_ON(size < sizeof(*cb));
2941 cb = kzalloc(size, GFP_ATOMIC);
2942 if (cb) {
2943 cb->data = data;
2944 cb->callback = unplug;
2945 list_add(&cb->list, &plug->cb_list);
2946 }
2947 return cb;
2948}
2949EXPORT_SYMBOL(blk_check_plugged);
2950
Jens Axboe49cac012011-04-16 13:51:05 +02002951void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002952{
2953 struct request_queue *q;
2954 unsigned long flags;
2955 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002956 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002957 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002958
2959 BUG_ON(plug->magic != PLUG_MAGIC);
2960
NeilBrown74018dc2012-07-31 09:08:15 +02002961 flush_plug_callbacks(plug, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002962 if (list_empty(&plug->list))
2963 return;
2964
NeilBrown109b8122011-04-11 14:13:10 +02002965 list_splice_init(&plug->list, &list);
2966
2967 if (plug->should_sort) {
2968 list_sort(NULL, &list, plug_rq_cmp);
2969 plug->should_sort = 0;
2970 }
Jens Axboe73c10102011-03-08 13:19:51 +01002971
2972 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002973 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002974
2975 /*
2976 * Save and disable interrupts here, to avoid doing it for every
2977 * queue lock we have to take.
2978 */
Jens Axboe73c10102011-03-08 13:19:51 +01002979 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002980 while (!list_empty(&list)) {
2981 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002982 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002983 BUG_ON(!rq->q);
2984 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002985 /*
2986 * This drops the queue lock
2987 */
2988 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002989 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002990 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002991 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002992 spin_lock(q->queue_lock);
2993 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002994
2995 /*
2996 * Short-circuit if @q is dead
2997 */
2998 if (unlikely(blk_queue_dead(q))) {
2999 __blk_end_request_all(rq, -ENODEV);
3000 continue;
3001 }
3002
Jens Axboe73c10102011-03-08 13:19:51 +01003003 /*
3004 * rq is already accounted, so use raw insert
3005 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003006 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3007 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3008 else
3009 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003010
3011 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003012 }
3013
Jens Axboe99e22592011-04-18 09:59:55 +02003014 /*
3015 * This drops the queue lock
3016 */
3017 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003018 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003019
Jens Axboe73c10102011-03-08 13:19:51 +01003020 local_irq_restore(flags);
3021}
Jens Axboe73c10102011-03-08 13:19:51 +01003022
3023void blk_finish_plug(struct blk_plug *plug)
3024{
Jens Axboef6603782011-04-15 15:49:07 +02003025 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003026
3027 if (plug == current->plug)
3028 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003029}
3030EXPORT_SYMBOL(blk_finish_plug);
3031
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032int __init blk_dev_init(void)
3033{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003034 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3035 sizeof(((struct request *)0)->cmd_flags));
3036
Tejun Heo89b90be2011-01-03 15:01:47 +01003037 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3038 kblockd_workqueue = alloc_workqueue("kblockd",
3039 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 if (!kblockd_workqueue)
3041 panic("Failed to create kblockd\n");
3042
3043 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003044 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
Jens Axboe8324aa92008-01-29 14:51:59 +01003046 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003047 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 return 0;
3050}