blob: 33c33bc99ddd5546e6ba30ce267cb436d0328c51 [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>
Lin Ming6c954662013-03-23 11:42:26 +080033#include <linux/pm_runtime.h>
Li Zefan55782132009-06-09 13:43:05 +080034
35#define CREATE_TRACE_POINTS
36#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Jens Axboe8324aa92008-01-29 14:51:59 +010038#include "blk.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080039#include "blk-cgroup.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010040
Mike Snitzerd07335e2010-11-16 12:52:38 +010041EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020042EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070043EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
NeilBrowncbae8d42012-12-14 20:49:27 +010044EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010045
Tejun Heoa73f7302011-12-14 00:33:37 +010046DEFINE_IDA(blk_queue_ida);
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * For the allocated request tables
50 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010051static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/*
54 * For queue allocation
55 */
Jens Axboe6728cb02008-01-31 13:03:55 +010056struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * Controlling structure to kblockd
60 */
Jens Axboeff856ba2006-01-09 16:02:34 +010061static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Jens Axboe26b82562008-01-29 13:54:41 +010063static void drive_stat_acct(struct request *rq, int new_io)
64{
Jens Axboe28f13702008-05-07 10:15:46 +020065 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010066 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090067 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010068
Jens Axboec2553b52009-04-24 08:10:11 +020069 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010070 return;
71
Tejun Heo074a7ac2008-08-25 19:56:14 +090072 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090073
Jerome Marchand09e099d2011-01-05 16:57:38 +010074 if (!new_io) {
75 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090076 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010077 } else {
78 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010079 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010080 /*
81 * The partition is already being removed,
82 * the request will be accounted on the disk only
83 *
84 * We take a reference on disk->part0 although that
85 * partition will never be deleted, so we can treat
86 * it as any other partition.
87 */
88 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010089 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010090 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090091 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020092 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010093 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010094 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020095
Tejun Heo074a7ac2008-08-25 19:56:14 +090096 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010097}
98
Jens Axboe8324aa92008-01-29 14:51:59 +010099void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
101 int nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) + 1;
104 if (nr > q->nr_requests)
105 nr = q->nr_requests;
106 q->nr_congestion_on = nr;
107
108 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
109 if (nr < 1)
110 nr = 1;
111 q->nr_congestion_off = nr;
112}
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/**
115 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * @bdev: device
117 *
118 * Locates the passed device's request queue and returns the address of its
119 * backing_dev_info
120 *
121 * Will return NULL if the request queue cannot be located.
122 */
123struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
124{
125 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200126 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
128 if (q)
129 ret = &q->backing_dev_info;
130 return ret;
131}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132EXPORT_SYMBOL(blk_get_backing_dev_info);
133
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200134void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200136 memset(rq, 0, sizeof(*rq));
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700139 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200140 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100141 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900142 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200143 INIT_HLIST_NODE(&rq->hash);
144 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200145 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800146 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100147 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100148 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900149 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700150 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100151 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200153EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
NeilBrown5bb23a62007-09-27 12:46:13 +0200155static void req_bio_endio(struct request *rq, struct bio *bio,
156 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100157{
Tejun Heo143a87f2011-01-25 12:43:52 +0100158 if (error)
159 clear_bit(BIO_UPTODATE, &bio->bi_flags);
160 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
161 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100162
Tejun Heo143a87f2011-01-25 12:43:52 +0100163 if (unlikely(rq->cmd_flags & REQ_QUIET))
164 set_bit(BIO_QUIET, &bio->bi_flags);
165
Kent Overstreetf79ea412012-09-20 16:38:30 -0700166 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100167
168 /* don't actually finish bio if it's part of flush sequence */
169 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
170 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100171}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173void blk_dump_rq_flags(struct request *rq, char *msg)
174{
175 int bit;
176
Jens Axboe6728cb02008-01-31 13:03:55 +0100177 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200178 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
179 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Tejun Heo83096eb2009-05-07 22:24:39 +0900181 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
182 (unsigned long long)blk_rq_pos(rq),
183 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900185 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200187 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100188 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200189 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 printk("%02x ", rq->cmd[bit]);
191 printk("\n");
192 }
193}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194EXPORT_SYMBOL(blk_dump_rq_flags);
195
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500196static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200197{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500198 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200199
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500200 q = container_of(work, struct request_queue, delay_work.work);
201 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200202 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500203 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500207 * blk_delay_queue - restart queueing after defined interval
208 * @q: The &struct request_queue in question
209 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 *
211 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500212 * Sometimes queueing needs to be postponed for a little while, to allow
213 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100214 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500215 */
216void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
Bart Van Assche70460572012-11-28 13:45:56 +0100218 if (likely(!blk_queue_dead(q)))
219 queue_delayed_work(kblockd_workqueue, &q->delay_work,
220 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500222EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224/**
225 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200226 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 *
228 * Description:
229 * blk_start_queue() will clear the stop flag on the queue, and call
230 * the request_fn for the queue if it was in a stopped state when
231 * entered. Also see blk_stop_queue(). Queue lock must be held.
232 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200233void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200235 WARN_ON(!irqs_disabled());
236
Nick Piggin75ad23b2008-04-29 14:48:33 +0200237 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200238 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240EXPORT_SYMBOL(blk_start_queue);
241
242/**
243 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200244 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 *
246 * Description:
247 * The Linux block layer assumes that a block driver will consume all
248 * entries on the request queue when the request_fn strategy is called.
249 * Often this will not happen, because of hardware limitations (queue
250 * depth settings). If a device driver gets a 'queue full' response,
251 * or if it simply chooses not to queue more I/O at one point, it can
252 * call this function to prevent the request_fn from being called until
253 * the driver has signalled it's ready to go again. This happens by calling
254 * blk_start_queue() to restart queue operations. Queue lock must be held.
255 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200256void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Tejun Heo136b5722012-08-21 13:18:24 -0700258 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200259 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
261EXPORT_SYMBOL(blk_stop_queue);
262
263/**
264 * blk_sync_queue - cancel any pending callbacks on a queue
265 * @q: the queue
266 *
267 * Description:
268 * The block layer may perform asynchronous callback activity
269 * on a queue, such as calling the unplug function after a timeout.
270 * A block device may call blk_sync_queue to ensure that any
271 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200272 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 * that its ->make_request_fn will not re-add plugging prior to calling
274 * this function.
275 *
Vivek Goyalda527772011-03-02 19:05:33 -0500276 * This function does not cancel any asynchronous activity arising
277 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800278 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500279 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 */
281void blk_sync_queue(struct request_queue *q)
282{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100283 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500284 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285}
286EXPORT_SYMBOL(blk_sync_queue);
287
288/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100289 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
290 * @q: The queue to run
291 *
292 * Description:
293 * Invoke request handling on a queue if there are any pending requests.
294 * May be used to restart request handling after a request has completed.
295 * This variant runs the queue whether or not the queue has been
296 * stopped. Must be called with the queue lock held and interrupts
297 * disabled. See also @blk_run_queue.
298 */
299inline void __blk_run_queue_uncond(struct request_queue *q)
300{
301 if (unlikely(blk_queue_dead(q)))
302 return;
303
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100304 /*
305 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
306 * the queue lock internally. As a result multiple threads may be
307 * running such a request function concurrently. Keep track of the
308 * number of active request_fn invocations such that blk_drain_queue()
309 * can wait until all these request_fn calls have finished.
310 */
311 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100312 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100313 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100314}
315
316/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200317 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200319 *
320 * Description:
321 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200322 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Tejun Heoa538cd02009-04-23 11:05:17 +0900326 if (unlikely(blk_queue_stopped(q)))
327 return;
328
Bart Van Asschec246e802012-12-06 14:32:01 +0100329 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200330}
331EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200332
Nick Piggin75ad23b2008-04-29 14:48:33 +0200333/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200334 * blk_run_queue_async - run a single device queue in workqueue context
335 * @q: The queue to run
336 *
337 * Description:
338 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100339 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200340 */
341void blk_run_queue_async(struct request_queue *q)
342{
Bart Van Assche70460572012-11-28 13:45:56 +0100343 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700344 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200345}
Jens Axboec21e6be2011-04-19 13:32:46 +0200346EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200347
348/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200349 * blk_run_queue - run a single device queue
350 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200351 *
352 * Description:
353 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900354 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200355 */
356void blk_run_queue(struct request_queue *q)
357{
358 unsigned long flags;
359
360 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200361 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 spin_unlock_irqrestore(q->queue_lock, flags);
363}
364EXPORT_SYMBOL(blk_run_queue);
365
Jens Axboe165125e2007-07-24 09:28:11 +0200366void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500367{
368 kobject_put(&q->kobj);
369}
Jens Axboed86e0e82011-05-27 07:44:43 +0200370EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500371
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200372/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100373 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200374 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200375 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200376 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200377 * Drain requests from @q. If @drain_all is set, all requests are drained.
378 * If not, only ELVPRIV requests are drained. The caller is responsible
379 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200380 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100381static void __blk_drain_queue(struct request_queue *q, bool drain_all)
382 __releases(q->queue_lock)
383 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200384{
Asias He458f27a2012-06-15 08:45:25 +0200385 int i;
386
Bart Van Assche807592a2012-11-28 13:43:38 +0100387 lockdep_assert_held(q->queue_lock);
388
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200389 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100390 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200391
Tejun Heob855b042012-03-06 21:24:55 +0100392 /*
393 * The caller might be trying to drain @q before its
394 * elevator is initialized.
395 */
396 if (q->elevator)
397 elv_drain_elevator(q);
398
Tejun Heo5efd6112012-03-05 13:15:12 -0800399 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200400
Tejun Heo4eabc942011-12-15 20:03:04 +0100401 /*
402 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100403 * driver init after queue creation or is not yet fully
404 * active yet. Some drivers (e.g. fd and loop) get unhappy
405 * in such cases. Kick queue iff dispatch queue has
406 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100407 */
Tejun Heob855b042012-03-06 21:24:55 +0100408 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100409 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200410
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700411 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100412 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100413
414 /*
415 * Unfortunately, requests are queued at and tracked from
416 * multiple places and there's no single counter which can
417 * be drained. Check all the queues and counters.
418 */
419 if (drain_all) {
420 drain |= !list_empty(&q->queue_head);
421 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700422 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100423 drain |= q->in_flight[i];
424 drain |= !list_empty(&q->flush_queue[i]);
425 }
426 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200427
Tejun Heo481a7d62011-12-14 00:33:37 +0100428 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200429 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100430
431 spin_unlock_irq(q->queue_lock);
432
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200433 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100434
435 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200436 }
Asias He458f27a2012-06-15 08:45:25 +0200437
438 /*
439 * With queue marked dead, any woken up waiter will fail the
440 * allocation path, so the wakeup chaining is lost and we're
441 * left with hung waiters. We need to wake up those waiters.
442 */
443 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700444 struct request_list *rl;
445
Tejun Heoa0516612012-06-26 15:05:44 -0700446 blk_queue_for_each_rl(rl, q)
447 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
448 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200449 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200450}
451
Tejun Heoc9a929d2011-10-19 14:42:16 +0200452/**
Tejun Heod7325802012-03-05 13:14:58 -0800453 * blk_queue_bypass_start - enter queue bypass mode
454 * @q: queue of interest
455 *
456 * In bypass mode, only the dispatch FIFO queue of @q is used. This
457 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800458 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700459 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
460 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800461 */
462void blk_queue_bypass_start(struct request_queue *q)
463{
Tejun Heob82d4b12012-04-13 13:11:31 -0700464 bool drain;
465
Tejun Heod7325802012-03-05 13:14:58 -0800466 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700467 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800468 queue_flag_set(QUEUE_FLAG_BYPASS, q);
469 spin_unlock_irq(q->queue_lock);
470
Tejun Heob82d4b12012-04-13 13:11:31 -0700471 if (drain) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100472 spin_lock_irq(q->queue_lock);
473 __blk_drain_queue(q, false);
474 spin_unlock_irq(q->queue_lock);
475
Tejun Heob82d4b12012-04-13 13:11:31 -0700476 /* ensure blk_queue_bypass() is %true inside RCU read lock */
477 synchronize_rcu();
478 }
Tejun Heod7325802012-03-05 13:14:58 -0800479}
480EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
481
482/**
483 * blk_queue_bypass_end - leave queue bypass mode
484 * @q: queue of interest
485 *
486 * Leave bypass mode and restore the normal queueing behavior.
487 */
488void blk_queue_bypass_end(struct request_queue *q)
489{
490 spin_lock_irq(q->queue_lock);
491 if (!--q->bypass_depth)
492 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
493 WARN_ON_ONCE(q->bypass_depth < 0);
494 spin_unlock_irq(q->queue_lock);
495}
496EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
497
498/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200499 * blk_cleanup_queue - shutdown a request queue
500 * @q: request queue to shutdown
501 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100502 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
503 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500504 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100505void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500506{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200507 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700508
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100509 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500510 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100511 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200512 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800513
Tejun Heo80fd9972012-04-13 14:50:53 -0700514 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100515 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700516 * that, unlike blk_queue_bypass_start(), we aren't performing
517 * synchronize_rcu() after entering bypass mode to avoid the delay
518 * as some drivers create and destroy a lot of queues while
519 * probing. This is still safe because blk_release_queue() will be
520 * called only after the queue refcnt drops to zero and nothing,
521 * RCU or not, would be traversing the queue by then.
522 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800523 q->bypass_depth++;
524 queue_flag_set(QUEUE_FLAG_BYPASS, q);
525
Tejun Heoc9a929d2011-10-19 14:42:16 +0200526 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
527 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100528 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200529 spin_unlock_irq(lock);
530 mutex_unlock(&q->sysfs_lock);
531
Bart Van Asschec246e802012-12-06 14:32:01 +0100532 /*
533 * Drain all requests queued before DYING marking. Set DEAD flag to
534 * prevent that q->request_fn() gets invoked after draining finished.
535 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100536 spin_lock_irq(lock);
537 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100538 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100539 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200540
541 /* @q won't process any more request, flush async actions */
542 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
543 blk_sync_queue(q);
544
Asias He5e5cfac2012-05-24 23:28:52 +0800545 spin_lock_irq(lock);
546 if (q->queue_lock != &q->__queue_lock)
547 q->queue_lock = &q->__queue_lock;
548 spin_unlock_irq(lock);
549
Tejun Heoc9a929d2011-10-19 14:42:16 +0200550 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500551 blk_put_queue(q);
552}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553EXPORT_SYMBOL(blk_cleanup_queue);
554
Tejun Heo5b788ce2012-06-04 20:40:59 -0700555int blk_init_rl(struct request_list *rl, struct request_queue *q,
556 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400558 if (unlikely(rl->rq_pool))
559 return 0;
560
Tejun Heo5b788ce2012-06-04 20:40:59 -0700561 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200562 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
563 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200564 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
565 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Christoph Lameter19460892005-06-23 00:08:19 -0700567 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700568 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700569 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (!rl->rq_pool)
571 return -ENOMEM;
572
573 return 0;
574}
575
Tejun Heo5b788ce2012-06-04 20:40:59 -0700576void blk_exit_rl(struct request_list *rl)
577{
578 if (rl->rq_pool)
579 mempool_destroy(rl->rq_pool);
580}
581
Jens Axboe165125e2007-07-24 09:28:11 +0200582struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100584 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700585}
586EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Jens Axboe165125e2007-07-24 09:28:11 +0200588struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700589{
Jens Axboe165125e2007-07-24 09:28:11 +0200590 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700591 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700592
Jens Axboe8324aa92008-01-29 14:51:59 +0100593 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700594 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if (!q)
596 return NULL;
597
Dan Carpenter00380a42012-03-23 09:58:54 +0100598 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100599 if (q->id < 0)
600 goto fail_q;
601
Jens Axboe0989a022009-06-12 14:42:56 +0200602 q->backing_dev_info.ra_pages =
603 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
604 q->backing_dev_info.state = 0;
605 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200606 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100607 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200608
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700609 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100610 if (err)
611 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700612
Matthew Garrett31373d02010-04-06 14:25:14 +0200613 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
614 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700615 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100616 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700617 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100618 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800619#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800620 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800621#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100622 INIT_LIST_HEAD(&q->flush_queue[0]);
623 INIT_LIST_HEAD(&q->flush_queue[1]);
624 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500625 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500626
Jens Axboe8324aa92008-01-29 14:51:59 +0100627 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Al Viro483f4af2006-03-18 18:34:37 -0500629 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700630 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500631
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500632 /*
633 * By default initialize queue_lock to internal lock and driver can
634 * override it later if need be.
635 */
636 q->queue_lock = &q->__queue_lock;
637
Tejun Heob82d4b12012-04-13 13:11:31 -0700638 /*
639 * A queue starts its life with bypass turned on to avoid
640 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700641 * init. The initial bypass will be finished when the queue is
642 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700643 */
644 q->bypass_depth = 1;
645 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
646
Tejun Heo5efd6112012-03-05 13:15:12 -0800647 if (blkcg_init_queue(q))
Tejun Heof51b8022012-03-05 13:15:05 -0800648 goto fail_id;
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100651
652fail_id:
653 ida_simple_remove(&blk_queue_ida, q->id);
654fail_q:
655 kmem_cache_free(blk_requestq_cachep, q);
656 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
Christoph Lameter19460892005-06-23 00:08:19 -0700658EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660/**
661 * blk_init_queue - prepare a request queue for use with a block device
662 * @rfn: The function to be called to process requests that have been
663 * placed on the queue.
664 * @lock: Request queue spin lock
665 *
666 * Description:
667 * If a block device wishes to use the standard request handling procedures,
668 * which sorts requests and coalesces adjacent requests, then it must
669 * call blk_init_queue(). The function @rfn will be called when there
670 * are requests on the queue that need to be processed. If the device
671 * supports plugging, then @rfn may not be called immediately when requests
672 * are available on the queue, but may be called at some time later instead.
673 * Plugged queues are generally unplugged when a buffer belonging to one
674 * of the requests on the queue is needed, or due to memory pressure.
675 *
676 * @rfn is not required, or even expected, to remove all requests off the
677 * queue, but only as many as it can handle at a time. If it does leave
678 * requests on the queue, it is responsible for arranging that the requests
679 * get dealt with eventually.
680 *
681 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200682 * request queue; this lock will be taken also from interrupt context, so irq
683 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200685 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 * it didn't succeed.
687 *
688 * Note:
689 * blk_init_queue() must be paired with a blk_cleanup_queue() call
690 * when the block device is deactivated (such as at module unload).
691 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700692
Jens Axboe165125e2007-07-24 09:28:11 +0200693struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100695 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700696}
697EXPORT_SYMBOL(blk_init_queue);
698
Jens Axboe165125e2007-07-24 09:28:11 +0200699struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700700blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
701{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600702 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600704 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
705 if (!uninit_q)
706 return NULL;
707
Mike Snitzer51514122011-11-23 10:59:13 +0100708 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600709 if (!q)
710 blk_cleanup_queue(uninit_q);
711
712 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200713}
714EXPORT_SYMBOL(blk_init_queue_node);
715
716struct request_queue *
717blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
718 spinlock_t *lock)
719{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (!q)
721 return NULL;
722
Tejun Heoa0516612012-06-26 15:05:44 -0700723 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500724 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900728 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700729 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500730
731 /* Override internal queue lock with supplied lock pointer */
732 if (lock)
733 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Jens Axboef3b144a2009-03-06 08:48:33 +0100735 /*
736 * This also sets hw/phys segments, boundary and size
737 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200738 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Alan Stern44ec9542007-02-20 11:01:57 -0500740 q->sg_reserved_size = INT_MAX;
741
Tejun Heob82d4b12012-04-13 13:11:31 -0700742 /* init elevator */
743 if (elevator_init(q, NULL))
744 return NULL;
Tejun Heob82d4b12012-04-13 13:11:31 -0700745 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
Mike Snitzer51514122011-11-23 10:59:13 +0100747EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Tejun Heo09ac46c2011-12-14 00:33:38 +0100749bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100751 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100752 __blk_get_queue(q);
753 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
Tejun Heo09ac46c2011-12-14 00:33:38 +0100756 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
Jens Axboed86e0e82011-05-27 07:44:43 +0200758EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Tejun Heo5b788ce2012-06-04 20:40:59 -0700760static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100762 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700763 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100764 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100765 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100766 }
767
Tejun Heo5b788ce2012-06-04 20:40:59 -0700768 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771/*
772 * ioc_batching returns true if the ioc is a valid batching request and
773 * should be given priority access to a request.
774 */
Jens Axboe165125e2007-07-24 09:28:11 +0200775static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
777 if (!ioc)
778 return 0;
779
780 /*
781 * Make sure the process is able to allocate at least 1 request
782 * even if the batch times out, otherwise we could theoretically
783 * lose wakeups.
784 */
785 return ioc->nr_batch_requests == q->nr_batching ||
786 (ioc->nr_batch_requests > 0
787 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
788}
789
790/*
791 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
792 * will cause the process to be a "batcher" on all queues in the system. This
793 * is the behaviour we want though - once it gets a wakeup it should be given
794 * a nice run.
795 */
Jens Axboe165125e2007-07-24 09:28:11 +0200796static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
798 if (!ioc || ioc_batching(q, ioc))
799 return;
800
801 ioc->nr_batch_requests = q->nr_batching;
802 ioc->last_waited = jiffies;
803}
804
Tejun Heo5b788ce2012-06-04 20:40:59 -0700805static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700807 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Tejun Heoa0516612012-06-26 15:05:44 -0700809 /*
810 * bdi isn't aware of blkcg yet. As all async IOs end up root
811 * blkcg anyway, just use root blkcg state.
812 */
813 if (rl == &q->root_rl &&
814 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200815 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Jens Axboe1faa16d2009-04-06 14:48:01 +0200817 if (rl->count[sync] + 1 <= q->nr_requests) {
818 if (waitqueue_active(&rl->wait[sync]))
819 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Tejun Heo5b788ce2012-06-04 20:40:59 -0700821 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823}
824
825/*
826 * A request has just been released. Account for it, update the full and
827 * congestion status, wake up any waiters. Called under q->queue_lock.
828 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700829static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700831 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200832 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700834 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200835 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200836 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700837 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Tejun Heo5b788ce2012-06-04 20:40:59 -0700839 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Jens Axboe1faa16d2009-04-06 14:48:01 +0200841 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700842 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100846 * Determine if elevator data should be initialized when allocating the
847 * request associated with @bio.
848 */
849static bool blk_rq_should_init_elevator(struct bio *bio)
850{
851 if (!bio)
852 return true;
853
854 /*
855 * Flush requests do not use the elevator so skip initialization.
856 * This allows a request to share the flush and elevator data.
857 */
858 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
859 return false;
860
861 return true;
862}
863
Tejun Heoda8303c2011-10-19 14:33:05 +0200864/**
Tejun Heo852c7882012-03-05 13:15:27 -0800865 * rq_ioc - determine io_context for request allocation
866 * @bio: request being allocated is for this bio (can be %NULL)
867 *
868 * Determine io_context to use for request allocation for @bio. May return
869 * %NULL if %current->io_context doesn't exist.
870 */
871static struct io_context *rq_ioc(struct bio *bio)
872{
873#ifdef CONFIG_BLK_CGROUP
874 if (bio && bio->bi_ioc)
875 return bio->bi_ioc;
876#endif
877 return current->io_context;
878}
879
880/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700881 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700882 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200883 * @rw_flags: RW and SYNC flags
884 * @bio: bio to allocate request for (can be %NULL)
885 * @gfp_mask: allocation mask
886 *
887 * Get a free request from @q. This function may fail under memory
888 * pressure or if @q is dead.
889 *
890 * Must be callled with @q->queue_lock held and,
891 * Returns %NULL on failure, with @q->queue_lock held.
892 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700894static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700895 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700897 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800898 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700899 struct elevator_type *et = q->elevator->type;
900 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100901 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200902 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200903 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100905 if (unlikely(blk_queue_dying(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200906 return NULL;
907
Jens Axboe7749a8d2006-12-13 13:02:26 +0100908 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100909 if (may_queue == ELV_MQUEUE_NO)
910 goto rq_starved;
911
Jens Axboe1faa16d2009-04-06 14:48:01 +0200912 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
913 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100914 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100915 * The queue will fill after this allocation, so set
916 * it as full, and mark this process as "batching".
917 * This process will be allowed to complete a batch of
918 * requests, others will be blocked.
919 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700920 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100921 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700922 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100923 } else {
924 if (may_queue != ELV_MQUEUE_MUST
925 && !ioc_batching(q, ioc)) {
926 /*
927 * The queue is full and the allocating
928 * process is not a "batcher", and not
929 * exempted by the IO scheduler
930 */
Tejun Heob6792812012-03-05 13:15:23 -0800931 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100932 }
933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 }
Tejun Heoa0516612012-06-26 15:05:44 -0700935 /*
936 * bdi isn't aware of blkcg yet. As all async IOs end up
937 * root blkcg anyway, just use root blkcg state.
938 */
939 if (rl == &q->root_rl)
940 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 }
942
Jens Axboe082cf692005-06-28 16:35:11 +0200943 /*
944 * Only allow batching queuers to allocate up to 50% over the defined
945 * limit of requests, otherwise we could have thousands of requests
946 * allocated with any setting of ->nr_requests
947 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200948 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800949 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100950
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700951 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200952 rl->count[is_sync]++;
953 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200954
Tejun Heof1f8cc92011-12-14 00:33:42 +0100955 /*
956 * Decide whether the new request will be managed by elevator. If
957 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
958 * prevent the current elevator from being destroyed until the new
959 * request is freed. This guarantees icq's won't be destroyed and
960 * makes creating new ones safe.
961 *
962 * Also, lookup icq while holding queue_lock. If it doesn't exist,
963 * it will be created after releasing queue_lock.
964 */
Tejun Heod7325802012-03-05 13:14:58 -0800965 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200966 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700967 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100968 if (et->icq_cache && ioc)
969 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100970 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200971
Jens Axboef253b862010-10-24 22:06:02 +0200972 if (blk_queue_io_stat(q))
973 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 spin_unlock_irq(q->queue_lock);
975
Tejun Heo29e2b092012-04-19 16:29:21 -0700976 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700977 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700978 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800979 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Tejun Heo29e2b092012-04-19 16:29:21 -0700981 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700982 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700983 rq->cmd_flags = rw_flags | REQ_ALLOCED;
984
Tejun Heoaaf7c682012-04-19 16:29:22 -0700985 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700986 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700987 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700988 if (ioc)
989 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700990 if (!icq)
991 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700992 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700993
994 rq->elv.icq = icq;
995 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
996 goto fail_elvpriv;
997
998 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -0700999 if (icq)
1000 get_io_context(icq->ioc);
1001 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001002out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001003 /*
1004 * ioc may be NULL here, and ioc_batching will be false. That's
1005 * OK, if the queue is under the request limit then requests need
1006 * not count toward the nr_batch_requests limit. There will always
1007 * be some limit enforced by BLK_BATCH_TIME.
1008 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (ioc_batching(q, ioc))
1010 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001011
Jens Axboe1faa16d2009-04-06 14:48:01 +02001012 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001014
Tejun Heoaaf7c682012-04-19 16:29:22 -07001015fail_elvpriv:
1016 /*
1017 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1018 * and may fail indefinitely under memory pressure and thus
1019 * shouldn't stall IO. Treat this request as !elvpriv. This will
1020 * disturb iosched and blkcg but weird is bettern than dead.
1021 */
1022 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1023 dev_name(q->backing_dev_info.dev));
1024
1025 rq->cmd_flags &= ~REQ_ELVPRIV;
1026 rq->elv.icq = NULL;
1027
1028 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001029 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001030 spin_unlock_irq(q->queue_lock);
1031 goto out;
1032
Tejun Heob6792812012-03-05 13:15:23 -08001033fail_alloc:
1034 /*
1035 * Allocation failed presumably due to memory. Undo anything we
1036 * might have messed up.
1037 *
1038 * Allocating task should really be put onto the front of the wait
1039 * queue, but this is pretty rare.
1040 */
1041 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001042 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001043
1044 /*
1045 * in the very unlikely event that allocation failed and no
1046 * requests for this direction was pending, mark us starved so that
1047 * freeing of a request in the other direction will notice
1048 * us. another possible fix would be to split the rq mempool into
1049 * READ and WRITE
1050 */
1051rq_starved:
1052 if (unlikely(rl->count[is_sync] == 0))
1053 rl->starved[is_sync] = 1;
1054 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055}
1056
Tejun Heoda8303c2011-10-19 14:33:05 +02001057/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001058 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001059 * @q: request_queue to allocate request from
1060 * @rw_flags: RW and SYNC flags
1061 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001062 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001063 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001064 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1065 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001066 *
1067 * Must be callled with @q->queue_lock held and,
1068 * Returns %NULL on failure, with @q->queue_lock held.
1069 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001071static struct request *get_request(struct request_queue *q, int rw_flags,
1072 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001074 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001075 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001076 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001078
1079 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001080retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001081 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001082 if (rq)
1083 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001085 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001086 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001087 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Tejun Heoa06e05e2012-06-04 20:40:55 -07001090 /* wait on @rl and retry */
1091 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1092 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001093
Tejun Heoa06e05e2012-06-04 20:40:55 -07001094 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Tejun Heoa06e05e2012-06-04 20:40:55 -07001096 spin_unlock_irq(q->queue_lock);
1097 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Tejun Heoa06e05e2012-06-04 20:40:55 -07001099 /*
1100 * After sleeping, we become a "batching" process and will be able
1101 * to allocate at least one request, and up to a big batch of them
1102 * for a small period time. See ioc_batching, ioc_set_batching
1103 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001104 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Tejun Heoa06e05e2012-06-04 20:40:55 -07001106 spin_lock_irq(q->queue_lock);
1107 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001108
Tejun Heoa06e05e2012-06-04 20:40:55 -07001109 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110}
1111
Jens Axboe165125e2007-07-24 09:28:11 +02001112struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
1114 struct request *rq;
1115
1116 BUG_ON(rw != READ && rw != WRITE);
1117
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001118 /* create ioc upfront */
1119 create_io_context(gfp_mask, q->node);
1120
Nick Piggind6344532005-06-28 20:45:14 -07001121 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001122 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001123 if (!rq)
1124 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001125 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 return rq;
1128}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129EXPORT_SYMBOL(blk_get_request);
1130
1131/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001132 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001133 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001134 * @bio: The bio describing the memory mappings that will be submitted for IO.
1135 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001136 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001137 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001138 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1139 * type commands. Where the struct request needs to be farther initialized by
1140 * the caller. It is passed a &struct bio, which describes the memory info of
1141 * the I/O transfer.
1142 *
1143 * The caller of blk_make_request must make sure that bi_io_vec
1144 * are set to describe the memory buffers. That bio_data_dir() will return
1145 * the needed direction of the request. (And all bio's in the passed bio-chain
1146 * are properly set accordingly)
1147 *
1148 * If called under none-sleepable conditions, mapped bio buffers must not
1149 * need bouncing, by calling the appropriate masked or flagged allocator,
1150 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1151 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001152 *
1153 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1154 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1155 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1156 * completion of a bio that hasn't been submitted yet, thus resulting in a
1157 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1158 * of bio_alloc(), as that avoids the mempool deadlock.
1159 * If possible a big IO should be split into smaller parts when allocation
1160 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001161 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001162struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1163 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001164{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001165 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1166
1167 if (unlikely(!rq))
1168 return ERR_PTR(-ENOMEM);
1169
1170 for_each_bio(bio) {
1171 struct bio *bounce_bio = bio;
1172 int ret;
1173
1174 blk_queue_bounce(q, &bounce_bio);
1175 ret = blk_rq_append_bio(q, rq, bounce_bio);
1176 if (unlikely(ret)) {
1177 blk_put_request(rq);
1178 return ERR_PTR(ret);
1179 }
1180 }
1181
1182 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001183}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001184EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001185
1186/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 * blk_requeue_request - put a request back on queue
1188 * @q: request queue where request should be inserted
1189 * @rq: request to be inserted
1190 *
1191 * Description:
1192 * Drivers often keep queueing requests until the hardware cannot accept
1193 * more, when that condition happens we need to put the request back
1194 * on the queue. Must be called with queue lock held.
1195 */
Jens Axboe165125e2007-07-24 09:28:11 +02001196void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001198 blk_delete_timer(rq);
1199 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001200 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (blk_rq_tagged(rq))
1203 blk_queue_end_tag(q, rq);
1204
James Bottomleyba396a62009-05-27 14:17:08 +02001205 BUG_ON(blk_queued_rq(rq));
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 elv_requeue_request(q, rq);
1208}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209EXPORT_SYMBOL(blk_requeue_request);
1210
Jens Axboe73c10102011-03-08 13:19:51 +01001211static void add_acct_request(struct request_queue *q, struct request *rq,
1212 int where)
1213{
1214 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001215 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001216}
1217
Tejun Heo074a7ac2008-08-25 19:56:14 +09001218static void part_round_stats_single(int cpu, struct hd_struct *part,
1219 unsigned long now)
1220{
1221 if (now == part->stamp)
1222 return;
1223
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001224 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001225 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001226 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001227 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1228 }
1229 part->stamp = now;
1230}
1231
1232/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001233 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1234 * @cpu: cpu number for stats access
1235 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 *
1237 * The average IO queue length and utilisation statistics are maintained
1238 * by observing the current state of the queue length and the amount of
1239 * time it has been in this state for.
1240 *
1241 * Normally, that accounting is done on IO completion, but that can result
1242 * in more than a second's worth of IO being accounted for within any one
1243 * second, leading to >100% utilisation. To deal with that, we call this
1244 * function to do a round-off before returning the results when reading
1245 * /proc/diskstats. This accounts immediately for all queue usage up to
1246 * the current jiffies and restarts the counters again.
1247 */
Tejun Heoc9959052008-08-25 19:47:21 +09001248void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001249{
1250 unsigned long now = jiffies;
1251
Tejun Heo074a7ac2008-08-25 19:56:14 +09001252 if (part->partno)
1253 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1254 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001255}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001256EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001257
Lin Mingc8158812013-03-23 11:42:27 +08001258#ifdef CONFIG_PM_RUNTIME
1259static void blk_pm_put_request(struct request *rq)
1260{
1261 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1262 pm_runtime_mark_last_busy(rq->q->dev);
1263}
1264#else
1265static inline void blk_pm_put_request(struct request *rq) {}
1266#endif
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268/*
1269 * queue lock must be held
1270 */
Jens Axboe165125e2007-07-24 09:28:11 +02001271void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (unlikely(!q))
1274 return;
1275 if (unlikely(--req->ref_count))
1276 return;
1277
Lin Mingc8158812013-03-23 11:42:27 +08001278 blk_pm_put_request(req);
1279
Tejun Heo8922e162005-10-20 16:23:44 +02001280 elv_completed_request(q, req);
1281
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001282 /* this is a bio leak */
1283 WARN_ON(req->bio != NULL);
1284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 /*
1286 * Request may not have originated from ll_rw_blk. if not,
1287 * it didn't come out of our reserved rq pools
1288 */
Jens Axboe49171e52006-08-10 08:59:11 +02001289 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001290 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001291 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001294 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Tejun Heoa0516612012-06-26 15:05:44 -07001296 blk_free_request(rl, req);
1297 freed_request(rl, flags);
1298 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 }
1300}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001301EXPORT_SYMBOL_GPL(__blk_put_request);
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303void blk_put_request(struct request *req)
1304{
Tejun Heo8922e162005-10-20 16:23:44 +02001305 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001306 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001308 spin_lock_irqsave(q->queue_lock, flags);
1309 __blk_put_request(q, req);
1310 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312EXPORT_SYMBOL(blk_put_request);
1313
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001314/**
1315 * blk_add_request_payload - add a payload to a request
1316 * @rq: request to update
1317 * @page: page backing the payload
1318 * @len: length of the payload.
1319 *
1320 * This allows to later add a payload to an already submitted request by
1321 * a block driver. The driver needs to take care of freeing the payload
1322 * itself.
1323 *
1324 * Note that this is a quite horrible hack and nothing but handling of
1325 * discard requests should ever use it.
1326 */
1327void blk_add_request_payload(struct request *rq, struct page *page,
1328 unsigned int len)
1329{
1330 struct bio *bio = rq->bio;
1331
1332 bio->bi_io_vec->bv_page = page;
1333 bio->bi_io_vec->bv_offset = 0;
1334 bio->bi_io_vec->bv_len = len;
1335
1336 bio->bi_size = len;
1337 bio->bi_vcnt = 1;
1338 bio->bi_phys_segments = 1;
1339
1340 rq->__data_len = rq->resid_len = len;
1341 rq->nr_phys_segments = 1;
1342 rq->buffer = bio_data(bio);
1343}
1344EXPORT_SYMBOL_GPL(blk_add_request_payload);
1345
Jens Axboe73c10102011-03-08 13:19:51 +01001346static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1347 struct bio *bio)
1348{
1349 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1350
Jens Axboe73c10102011-03-08 13:19:51 +01001351 if (!ll_back_merge_fn(q, req, bio))
1352 return false;
1353
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001354 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001355
1356 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1357 blk_rq_set_mixed_merge(req);
1358
1359 req->biotail->bi_next = bio;
1360 req->biotail = bio;
1361 req->__data_len += bio->bi_size;
1362 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1363
1364 drive_stat_acct(req, 0);
1365 return true;
1366}
1367
1368static bool bio_attempt_front_merge(struct request_queue *q,
1369 struct request *req, struct bio *bio)
1370{
1371 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001372
Jens Axboe73c10102011-03-08 13:19:51 +01001373 if (!ll_front_merge_fn(q, req, bio))
1374 return false;
1375
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001376 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001377
1378 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1379 blk_rq_set_mixed_merge(req);
1380
Jens Axboe73c10102011-03-08 13:19:51 +01001381 bio->bi_next = req->bio;
1382 req->bio = bio;
1383
1384 /*
1385 * may not be valid. if the low level driver said
1386 * it didn't need a bounce buffer then it better
1387 * not touch req->buffer either...
1388 */
1389 req->buffer = bio_data(bio);
1390 req->__sector = bio->bi_sector;
1391 req->__data_len += bio->bi_size;
1392 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1393
1394 drive_stat_acct(req, 0);
1395 return true;
1396}
1397
Tejun Heobd87b582011-10-19 14:33:08 +02001398/**
1399 * attempt_plug_merge - try to merge with %current's plugged list
1400 * @q: request_queue new bio is being queued at
1401 * @bio: new bio being queued
1402 * @request_count: out parameter for number of traversed plugged requests
1403 *
1404 * Determine whether @bio being queued on @q can be merged with a request
1405 * on %current's plugged list. Returns %true if merge was successful,
1406 * otherwise %false.
1407 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001408 * Plugging coalesces IOs from the same issuer for the same purpose without
1409 * going through @q->queue_lock. As such it's more of an issuing mechanism
1410 * than scheduling, and the request, while may have elvpriv data, is not
1411 * added on the elevator at this point. In addition, we don't have
1412 * reliable access to the elevator outside queue lock. Only check basic
1413 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001414 */
Tejun Heobd87b582011-10-19 14:33:08 +02001415static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1416 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001417{
1418 struct blk_plug *plug;
1419 struct request *rq;
1420 bool ret = false;
1421
Tejun Heobd87b582011-10-19 14:33:08 +02001422 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001423 if (!plug)
1424 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001425 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001426
1427 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1428 int el_ret;
1429
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001430 if (rq->q == q)
1431 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001432
Tejun Heo07c2bd32012-02-08 09:19:42 +01001433 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001434 continue;
1435
Tejun Heo050c8ea2012-02-08 09:19:38 +01001436 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001437 if (el_ret == ELEVATOR_BACK_MERGE) {
1438 ret = bio_attempt_back_merge(q, rq, bio);
1439 if (ret)
1440 break;
1441 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1442 ret = bio_attempt_front_merge(q, rq, bio);
1443 if (ret)
1444 break;
1445 }
1446 }
1447out:
1448 return ret;
1449}
1450
Jens Axboe86db1e22008-01-29 14:53:40 +01001451void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001452{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001453 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001454
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001455 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1456 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001457 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001458
Tejun Heo52d9e672006-01-06 09:49:58 +01001459 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001460 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001461 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001462 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001463}
1464
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001465void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001467 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001468 struct blk_plug *plug;
1469 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1470 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001471 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 /*
1474 * low level driver can indicate that it wants pages above a
1475 * certain limit bounced to low memory (ie for highmem, or even
1476 * ISA dma in theory)
1477 */
1478 blk_queue_bounce(q, &bio);
1479
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001480 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1481 bio_endio(bio, -EIO);
1482 return;
1483 }
1484
Tejun Heo4fed9472010-09-03 11:56:17 +02001485 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001486 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001487 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001488 goto get_rq;
1489 }
1490
Jens Axboe73c10102011-03-08 13:19:51 +01001491 /*
1492 * Check if we can merge with the plugged list before grabbing
1493 * any locks.
1494 */
Tejun Heobd87b582011-10-19 14:33:08 +02001495 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001496 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001497
1498 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001501 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001502 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001503 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001504 if (!attempt_back_merge(q, req))
1505 elv_merged_request(q, req, el_ret);
1506 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001507 }
Jens Axboe73c10102011-03-08 13:19:51 +01001508 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001509 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001510 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001511 if (!attempt_front_merge(q, req))
1512 elv_merged_request(q, req, el_ret);
1513 goto out_unlock;
1514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 }
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001518 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001519 * This sync check and mask will be re-done in init_request_from_bio(),
1520 * but we need to set it earlier to expose the sync flag to the
1521 * rq allocator and io schedulers.
1522 */
1523 rw_flags = bio_data_dir(bio);
1524 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001525 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001526
1527 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001528 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001529 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001530 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001531 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001532 if (unlikely(!req)) {
1533 bio_endio(bio, -ENODEV); /* @q is dead */
1534 goto out_unlock;
1535 }
Nick Piggind6344532005-06-28 20:45:14 -07001536
Nick Piggin450991b2005-06-28 20:45:13 -07001537 /*
1538 * After dropping the lock and possibly sleeping here, our request
1539 * may now be mergeable after it had proven unmergeable (above).
1540 * We don't worry about that case for efficiency. It won't happen
1541 * often, and the elevators are able to handle it.
1542 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001543 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Tao Ma9562ad92011-10-24 16:11:30 +02001545 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001546 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001547
Jens Axboe73c10102011-03-08 13:19:51 +01001548 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001549 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001550 /*
1551 * If this is the first request added after a plug, fire
1552 * of a plug trace. If others have been added before, check
1553 * if we have multiple devices in this plug. If so, make a
1554 * note to sort the list before dispatch.
1555 */
1556 if (list_empty(&plug->list))
1557 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001558 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001559 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001560 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001561 trace_block_plug(q);
1562 }
Jens Axboe73c10102011-03-08 13:19:51 +01001563 }
Shaohua Lia6327162011-08-24 16:04:32 +02001564 list_add_tail(&req->queuelist, &plug->list);
1565 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001566 } else {
1567 spin_lock_irq(q->queue_lock);
1568 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001569 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001570out_unlock:
1571 spin_unlock_irq(q->queue_lock);
1572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573}
Jens Axboec20e8de2011-09-12 12:03:37 +02001574EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576/*
1577 * If bio->bi_dev is a partition, remap the location
1578 */
1579static inline void blk_partition_remap(struct bio *bio)
1580{
1581 struct block_device *bdev = bio->bi_bdev;
1582
Jens Axboebf2de6f2007-09-27 13:01:25 +02001583 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 bio->bi_sector += p->start_sect;
1587 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001588
Mike Snitzerd07335e2010-11-16 12:52:38 +01001589 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1590 bdev->bd_dev,
1591 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
1593}
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595static void handle_bad_sector(struct bio *bio)
1596{
1597 char b[BDEVNAME_SIZE];
1598
1599 printk(KERN_INFO "attempt to access beyond end of device\n");
1600 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1601 bdevname(bio->bi_bdev, b),
1602 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001603 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001604 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 set_bit(BIO_EOF, &bio->bi_flags);
1607}
1608
Akinobu Mitac17bb492006-12-08 02:39:46 -08001609#ifdef CONFIG_FAIL_MAKE_REQUEST
1610
1611static DECLARE_FAULT_ATTR(fail_make_request);
1612
1613static int __init setup_fail_make_request(char *str)
1614{
1615 return setup_fault_attr(&fail_make_request, str);
1616}
1617__setup("fail_make_request=", setup_fail_make_request);
1618
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001619static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001620{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001621 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001622}
1623
1624static int __init fail_make_request_debugfs(void)
1625{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001626 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1627 NULL, &fail_make_request);
1628
1629 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001630}
1631
1632late_initcall(fail_make_request_debugfs);
1633
1634#else /* CONFIG_FAIL_MAKE_REQUEST */
1635
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001636static inline bool should_fail_request(struct hd_struct *part,
1637 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001638{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001639 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001640}
1641
1642#endif /* CONFIG_FAIL_MAKE_REQUEST */
1643
Jens Axboec07e2b42007-07-18 13:27:58 +02001644/*
1645 * Check whether this bio extends beyond the end of the device.
1646 */
1647static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1648{
1649 sector_t maxsector;
1650
1651 if (!nr_sectors)
1652 return 0;
1653
1654 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001655 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001656 if (maxsector) {
1657 sector_t sector = bio->bi_sector;
1658
1659 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1660 /*
1661 * This may well happen - the kernel calls bread()
1662 * without checking the size of the device, e.g., when
1663 * mounting a device.
1664 */
1665 handle_bad_sector(bio);
1666 return 1;
1667 }
1668 }
1669
1670 return 0;
1671}
1672
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001673static noinline_for_stack bool
1674generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Jens Axboe165125e2007-07-24 09:28:11 +02001676 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001677 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001678 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001679 char b[BDEVNAME_SIZE];
1680 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
1682 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Jens Axboec07e2b42007-07-18 13:27:58 +02001684 if (bio_check_eod(bio, nr_sectors))
1685 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001687 q = bdev_get_queue(bio->bi_bdev);
1688 if (unlikely(!q)) {
1689 printk(KERN_ERR
1690 "generic_make_request: Trying to access "
1691 "nonexistent block-device %s (%Lu)\n",
1692 bdevname(bio->bi_bdev, b),
1693 (long long) bio->bi_sector);
1694 goto end_io;
1695 }
1696
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001697 if (likely(bio_is_rw(bio) &&
1698 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001699 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1700 bdevname(bio->bi_bdev, b),
1701 bio_sectors(bio),
1702 queue_max_hw_sectors(q));
1703 goto end_io;
1704 }
1705
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001706 part = bio->bi_bdev->bd_part;
1707 if (should_fail_request(part, bio->bi_size) ||
1708 should_fail_request(&part_to_disk(part)->part0,
1709 bio->bi_size))
1710 goto end_io;
1711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001713 * If this device has partitions, remap block n
1714 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001716 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001718 if (bio_check_eod(bio, nr_sectors))
1719 goto end_io;
1720
1721 /*
1722 * Filter flush bio's early so that make_request based
1723 * drivers without flush support don't have to worry
1724 * about them.
1725 */
1726 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1727 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1728 if (!nr_sectors) {
1729 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001730 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001734 if ((bio->bi_rw & REQ_DISCARD) &&
1735 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001736 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001737 err = -EOPNOTSUPP;
1738 goto end_io;
1739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001741 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 err = -EOPNOTSUPP;
1743 goto end_io;
1744 }
1745
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001746 /*
1747 * Various block parts want %current->io_context and lazy ioc
1748 * allocation ends up trading a lot of pain for a small amount of
1749 * memory. Just allocate it upfront. This may fail and block
1750 * layer knows how to live with it.
1751 */
1752 create_io_context(GFP_ATOMIC, q->node);
1753
Tejun Heobc16a4f2011-10-19 14:33:01 +02001754 if (blk_throtl_bio(q, bio))
1755 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001756
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001757 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001758 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001759
1760end_io:
1761 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001762 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
1764
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001765/**
1766 * generic_make_request - hand a buffer to its device driver for I/O
1767 * @bio: The bio describing the location in memory and on the device.
1768 *
1769 * generic_make_request() is used to make I/O requests of block
1770 * devices. It is passed a &struct bio, which describes the I/O that needs
1771 * to be done.
1772 *
1773 * generic_make_request() does not return any status. The
1774 * success/failure status of the request, along with notification of
1775 * completion, is delivered asynchronously through the bio->bi_end_io
1776 * function described (one day) else where.
1777 *
1778 * The caller of generic_make_request must make sure that bi_io_vec
1779 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1780 * set to describe the device address, and the
1781 * bi_end_io and optionally bi_private are set to describe how
1782 * completion notification should be signaled.
1783 *
1784 * generic_make_request and the drivers it calls may use bi_next if this
1785 * bio happens to be merged with someone else, and may resubmit the bio to
1786 * a lower device by calling into generic_make_request recursively, which
1787 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001788 */
1789void generic_make_request(struct bio *bio)
1790{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001791 struct bio_list bio_list_on_stack;
1792
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001793 if (!generic_make_request_checks(bio))
1794 return;
1795
1796 /*
1797 * We only want one ->make_request_fn to be active at a time, else
1798 * stack usage with stacked devices could be a problem. So use
1799 * current->bio_list to keep a list of requests submited by a
1800 * make_request_fn function. current->bio_list is also used as a
1801 * flag to say if generic_make_request is currently active in this
1802 * task or not. If it is NULL, then no make_request is active. If
1803 * it is non-NULL, then a make_request is active, and new requests
1804 * should be added at the tail
1805 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001806 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001807 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001808 return;
1809 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001810
Neil Brownd89d8792007-05-01 09:53:42 +02001811 /* following loop may be a bit non-obvious, and so deserves some
1812 * explanation.
1813 * Before entering the loop, bio->bi_next is NULL (as all callers
1814 * ensure that) so we have a list with a single bio.
1815 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001816 * we assign bio_list to a pointer to the bio_list_on_stack,
1817 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001818 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001819 * through a recursive call to generic_make_request. If it
1820 * did, we find a non-NULL value in bio_list and re-enter the loop
1821 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001822 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001823 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001824 */
1825 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001826 bio_list_init(&bio_list_on_stack);
1827 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001828 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001829 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1830
1831 q->make_request_fn(q, bio);
1832
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001833 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001834 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001835 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001836}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837EXPORT_SYMBOL(generic_make_request);
1838
1839/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001840 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1842 * @bio: The &struct bio which describes the I/O
1843 *
1844 * submit_bio() is very similar in purpose to generic_make_request(), and
1845 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001846 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 *
1848 */
1849void submit_bio(int rw, struct bio *bio)
1850{
Jens Axboe22e2c502005-06-27 10:55:12 +02001851 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Jens Axboebf2de6f2007-09-27 13:01:25 +02001853 /*
1854 * If it's a regular read/write or a barrier with data attached,
1855 * go through the normal accounting stuff before submission.
1856 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001857 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001858 unsigned int count;
1859
1860 if (unlikely(rw & REQ_WRITE_SAME))
1861 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1862 else
1863 count = bio_sectors(bio);
1864
Jens Axboebf2de6f2007-09-27 13:01:25 +02001865 if (rw & WRITE) {
1866 count_vm_events(PGPGOUT, count);
1867 } else {
1868 task_io_account_read(bio->bi_size);
1869 count_vm_events(PGPGIN, count);
1870 }
1871
1872 if (unlikely(block_dump)) {
1873 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001874 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001875 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001876 (rw & WRITE) ? "WRITE" : "READ",
1877 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001878 bdevname(bio->bi_bdev, b),
1879 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882
1883 generic_make_request(bio);
1884}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885EXPORT_SYMBOL(submit_bio);
1886
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001887/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001888 * blk_rq_check_limits - Helper function to check a request for the queue limit
1889 * @q: the queue
1890 * @rq: the request being checked
1891 *
1892 * Description:
1893 * @rq may have been made based on weaker limitations of upper-level queues
1894 * in request stacking drivers, and it may violate the limitation of @q.
1895 * Since the block layer and the underlying device driver trust @rq
1896 * after it is inserted to @q, it should be checked against @q before
1897 * the insertion using this generic function.
1898 *
1899 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001900 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001901 * Request stacking drivers like request-based dm may change the queue
1902 * limits while requests are in the queue (e.g. dm's table swapping).
1903 * Such request stacking drivers should check those requests agaist
1904 * the new queue limits again when they dispatch those requests,
1905 * although such checkings are also done against the old queue limits
1906 * when submitting requests.
1907 */
1908int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1909{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001910 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001911 return 0;
1912
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001913 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001914 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1915 return -EIO;
1916 }
1917
1918 /*
1919 * queue's settings related to segment counting like q->bounce_pfn
1920 * may differ from that of other stacking queues.
1921 * Recalculate it to check the request correctly on this queue's
1922 * limitation.
1923 */
1924 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001925 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001926 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1927 return -EIO;
1928 }
1929
1930 return 0;
1931}
1932EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1933
1934/**
1935 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1936 * @q: the queue to submit the request
1937 * @rq: the request being queued
1938 */
1939int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1940{
1941 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001942 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001943
1944 if (blk_rq_check_limits(q, rq))
1945 return -EIO;
1946
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001947 if (rq->rq_disk &&
1948 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001949 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001950
1951 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001952 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01001953 spin_unlock_irqrestore(q->queue_lock, flags);
1954 return -ENODEV;
1955 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001956
1957 /*
1958 * Submitting request must be dequeued before calling this function
1959 * because it will be linked to another request_queue
1960 */
1961 BUG_ON(blk_queued_rq(rq));
1962
Jeff Moyer4853aba2011-08-15 21:37:25 +02001963 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1964 where = ELEVATOR_INSERT_FLUSH;
1965
1966 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001967 if (where == ELEVATOR_INSERT_FLUSH)
1968 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001969 spin_unlock_irqrestore(q->queue_lock, flags);
1970
1971 return 0;
1972}
1973EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1974
Tejun Heo80a761f2009-07-03 17:48:17 +09001975/**
1976 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1977 * @rq: request to examine
1978 *
1979 * Description:
1980 * A request could be merge of IOs which require different failure
1981 * handling. This function determines the number of bytes which
1982 * can be failed from the beginning of the request without
1983 * crossing into area which need to be retried further.
1984 *
1985 * Return:
1986 * The number of bytes to fail.
1987 *
1988 * Context:
1989 * queue_lock must be held.
1990 */
1991unsigned int blk_rq_err_bytes(const struct request *rq)
1992{
1993 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1994 unsigned int bytes = 0;
1995 struct bio *bio;
1996
1997 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1998 return blk_rq_bytes(rq);
1999
2000 /*
2001 * Currently the only 'mixing' which can happen is between
2002 * different fastfail types. We can safely fail portions
2003 * which have all the failfast bits that the first one has -
2004 * the ones which are at least as eager to fail as the first
2005 * one.
2006 */
2007 for (bio = rq->bio; bio; bio = bio->bi_next) {
2008 if ((bio->bi_rw & ff) != ff)
2009 break;
2010 bytes += bio->bi_size;
2011 }
2012
2013 /* this could lead to infinite loop */
2014 BUG_ON(blk_rq_bytes(rq) && !bytes);
2015 return bytes;
2016}
2017EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2018
Jens Axboebc58ba92009-01-23 10:54:44 +01002019static void blk_account_io_completion(struct request *req, unsigned int bytes)
2020{
Jens Axboec2553b52009-04-24 08:10:11 +02002021 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002022 const int rw = rq_data_dir(req);
2023 struct hd_struct *part;
2024 int cpu;
2025
2026 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002027 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002028 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2029 part_stat_unlock();
2030 }
2031}
2032
2033static void blk_account_io_done(struct request *req)
2034{
Jens Axboebc58ba92009-01-23 10:54:44 +01002035 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002036 * Account IO completion. flush_rq isn't accounted as a
2037 * normal IO on queueing nor completion. Accounting the
2038 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002039 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002040 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002041 unsigned long duration = jiffies - req->start_time;
2042 const int rw = rq_data_dir(req);
2043 struct hd_struct *part;
2044 int cpu;
2045
2046 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002047 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002048
2049 part_stat_inc(cpu, part, ios[rw]);
2050 part_stat_add(cpu, part, ticks[rw], duration);
2051 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002052 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002053
Jens Axboe6c23a962011-01-07 08:43:37 +01002054 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002055 part_stat_unlock();
2056 }
2057}
2058
Lin Mingc8158812013-03-23 11:42:27 +08002059#ifdef CONFIG_PM_RUNTIME
2060/*
2061 * Don't process normal requests when queue is suspended
2062 * or in the process of suspending/resuming
2063 */
2064static struct request *blk_pm_peek_request(struct request_queue *q,
2065 struct request *rq)
2066{
2067 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2068 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2069 return NULL;
2070 else
2071 return rq;
2072}
2073#else
2074static inline struct request *blk_pm_peek_request(struct request_queue *q,
2075 struct request *rq)
2076{
2077 return rq;
2078}
2079#endif
2080
Tejun Heo53a08802008-12-03 12:41:26 +01002081/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002082 * blk_peek_request - peek at the top of a request queue
2083 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002084 *
2085 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002086 * Return the request at the top of @q. The returned request
2087 * should be started using blk_start_request() before LLD starts
2088 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002089 *
2090 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002091 * Pointer to the request at the top of @q if available. Null
2092 * otherwise.
2093 *
2094 * Context:
2095 * queue_lock must be held.
2096 */
2097struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002098{
2099 struct request *rq;
2100 int ret;
2101
2102 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002103
2104 rq = blk_pm_peek_request(q, rq);
2105 if (!rq)
2106 break;
2107
Tejun Heo158dbda2009-04-23 11:05:18 +09002108 if (!(rq->cmd_flags & REQ_STARTED)) {
2109 /*
2110 * This is the first time the device driver
2111 * sees this request (possibly after
2112 * requeueing). Notify IO scheduler.
2113 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002114 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002115 elv_activate_rq(q, rq);
2116
2117 /*
2118 * just mark as started even if we don't start
2119 * it, a request that has been delayed should
2120 * not be passed by new incoming requests
2121 */
2122 rq->cmd_flags |= REQ_STARTED;
2123 trace_block_rq_issue(q, rq);
2124 }
2125
2126 if (!q->boundary_rq || q->boundary_rq == rq) {
2127 q->end_sector = rq_end_sector(rq);
2128 q->boundary_rq = NULL;
2129 }
2130
2131 if (rq->cmd_flags & REQ_DONTPREP)
2132 break;
2133
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002134 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002135 /*
2136 * make sure space for the drain appears we
2137 * know we can do this because max_hw_segments
2138 * has been adjusted to be one fewer than the
2139 * device can handle
2140 */
2141 rq->nr_phys_segments++;
2142 }
2143
2144 if (!q->prep_rq_fn)
2145 break;
2146
2147 ret = q->prep_rq_fn(q, rq);
2148 if (ret == BLKPREP_OK) {
2149 break;
2150 } else if (ret == BLKPREP_DEFER) {
2151 /*
2152 * the request may have been (partially) prepped.
2153 * we need to keep this request in the front to
2154 * avoid resource deadlock. REQ_STARTED will
2155 * prevent other fs requests from passing this one.
2156 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002157 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002158 !(rq->cmd_flags & REQ_DONTPREP)) {
2159 /*
2160 * remove the space for the drain we added
2161 * so that we don't add it again
2162 */
2163 --rq->nr_phys_segments;
2164 }
2165
2166 rq = NULL;
2167 break;
2168 } else if (ret == BLKPREP_KILL) {
2169 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002170 /*
2171 * Mark this request as started so we don't trigger
2172 * any debug logic in the end I/O path.
2173 */
2174 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002175 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002176 } else {
2177 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2178 break;
2179 }
2180 }
2181
2182 return rq;
2183}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002184EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002185
Tejun Heo9934c8c2009-05-08 11:54:16 +09002186void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002187{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002188 struct request_queue *q = rq->q;
2189
Tejun Heo158dbda2009-04-23 11:05:18 +09002190 BUG_ON(list_empty(&rq->queuelist));
2191 BUG_ON(ELV_ON_HASH(rq));
2192
2193 list_del_init(&rq->queuelist);
2194
2195 /*
2196 * the time frame between a request being removed from the lists
2197 * and to it is freed is accounted as io that is in progress at
2198 * the driver side.
2199 */
Divyesh Shah91952912010-04-01 15:01:41 -07002200 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002201 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002202 set_io_start_time_ns(rq);
2203 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002204}
2205
Tejun Heo5efccd12009-04-23 11:05:18 +09002206/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002207 * blk_start_request - start request processing on the driver
2208 * @req: request to dequeue
2209 *
2210 * Description:
2211 * Dequeue @req and start timeout timer on it. This hands off the
2212 * request to the driver.
2213 *
2214 * Block internal functions which don't want to start timer should
2215 * call blk_dequeue_request().
2216 *
2217 * Context:
2218 * queue_lock must be held.
2219 */
2220void blk_start_request(struct request *req)
2221{
2222 blk_dequeue_request(req);
2223
2224 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002225 * We are now handing the request to the hardware, initialize
2226 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002227 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002228 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002229 if (unlikely(blk_bidi_rq(req)))
2230 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2231
Tejun Heo9934c8c2009-05-08 11:54:16 +09002232 blk_add_timer(req);
2233}
2234EXPORT_SYMBOL(blk_start_request);
2235
2236/**
2237 * blk_fetch_request - fetch a request from a request queue
2238 * @q: request queue to fetch a request from
2239 *
2240 * Description:
2241 * Return the request at the top of @q. The request is started on
2242 * return and LLD can start processing it immediately.
2243 *
2244 * Return:
2245 * Pointer to the request at the top of @q if available. Null
2246 * otherwise.
2247 *
2248 * Context:
2249 * queue_lock must be held.
2250 */
2251struct request *blk_fetch_request(struct request_queue *q)
2252{
2253 struct request *rq;
2254
2255 rq = blk_peek_request(q);
2256 if (rq)
2257 blk_start_request(rq);
2258 return rq;
2259}
2260EXPORT_SYMBOL(blk_fetch_request);
2261
2262/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002263 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002264 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002265 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002266 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002267 *
2268 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002269 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2270 * the request structure even if @req doesn't have leftover.
2271 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002272 *
2273 * This special helper function is only for request stacking drivers
2274 * (e.g. request-based dm) so that they can handle partial completion.
2275 * Actual device drivers should use blk_end_request instead.
2276 *
2277 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2278 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002279 *
2280 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002281 * %false - this request doesn't have any more data
2282 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002283 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002284bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002286 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Tejun Heo2e60e022009-04-23 11:05:18 +09002288 if (!req->bio)
2289 return false;
2290
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002291 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002294 * For fs requests, rq is just carrier of independent bio's
2295 * and each partial completion should be handled separately.
2296 * Reset per-request error on each partial completion.
2297 *
2298 * TODO: tj: This is too subtle. It would be better to let
2299 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002301 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 req->errors = 0;
2303
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002304 if (error && req->cmd_type == REQ_TYPE_FS &&
2305 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002306 char *error_type;
2307
2308 switch (error) {
2309 case -ENOLINK:
2310 error_type = "recoverable transport";
2311 break;
2312 case -EREMOTEIO:
2313 error_type = "critical target";
2314 break;
2315 case -EBADE:
2316 error_type = "critical nexus";
2317 break;
2318 case -EIO:
2319 default:
2320 error_type = "I/O";
2321 break;
2322 }
Yi Zou37d7b342012-08-30 16:26:25 -07002323 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2324 error_type, req->rq_disk ?
2325 req->rq_disk->disk_name : "?",
2326 (unsigned long long)blk_rq_pos(req));
2327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 }
2329
Jens Axboebc58ba92009-01-23 10:54:44 +01002330 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002331
Kent Overstreetf79ea412012-09-20 16:38:30 -07002332 total_bytes = 0;
2333 while (req->bio) {
2334 struct bio *bio = req->bio;
2335 unsigned bio_bytes = min(bio->bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Kent Overstreetf79ea412012-09-20 16:38:30 -07002337 if (bio_bytes == bio->bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Kent Overstreetf79ea412012-09-20 16:38:30 -07002340 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Kent Overstreetf79ea412012-09-20 16:38:30 -07002342 total_bytes += bio_bytes;
2343 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Kent Overstreetf79ea412012-09-20 16:38:30 -07002345 if (!nr_bytes)
2346 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 }
2348
2349 /*
2350 * completely done
2351 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002352 if (!req->bio) {
2353 /*
2354 * Reset counters so that the request stacking driver
2355 * can find how many bytes remain in the request
2356 * later.
2357 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002358 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002359 return false;
2360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002362 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002363 req->buffer = bio_data(req->bio);
2364
2365 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002366 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002367 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002368
Tejun Heo80a761f2009-07-03 17:48:17 +09002369 /* mixed attributes always follow the first bio */
2370 if (req->cmd_flags & REQ_MIXED_MERGE) {
2371 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2372 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2373 }
2374
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002375 /*
2376 * If total number of sectors is less than the first segment
2377 * size, something has gone terribly wrong.
2378 */
2379 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002380 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002381 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002382 }
2383
2384 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002386
Tejun Heo2e60e022009-04-23 11:05:18 +09002387 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
Tejun Heo2e60e022009-04-23 11:05:18 +09002389EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
Tejun Heo2e60e022009-04-23 11:05:18 +09002391static bool blk_update_bidi_request(struct request *rq, int error,
2392 unsigned int nr_bytes,
2393 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002394{
Tejun Heo2e60e022009-04-23 11:05:18 +09002395 if (blk_update_request(rq, error, nr_bytes))
2396 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002397
Tejun Heo2e60e022009-04-23 11:05:18 +09002398 /* Bidi request must be completed as a whole */
2399 if (unlikely(blk_bidi_rq(rq)) &&
2400 blk_update_request(rq->next_rq, error, bidi_bytes))
2401 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002402
Jens Axboee2e1a142010-06-09 10:42:09 +02002403 if (blk_queue_add_random(rq->q))
2404 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002405
2406 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407}
2408
James Bottomley28018c22010-07-01 19:49:17 +09002409/**
2410 * blk_unprep_request - unprepare a request
2411 * @req: the request
2412 *
2413 * This function makes a request ready for complete resubmission (or
2414 * completion). It happens only after all error handling is complete,
2415 * so represents the appropriate moment to deallocate any resources
2416 * that were allocated to the request in the prep_rq_fn. The queue
2417 * lock is held when calling this.
2418 */
2419void blk_unprep_request(struct request *req)
2420{
2421 struct request_queue *q = req->q;
2422
2423 req->cmd_flags &= ~REQ_DONTPREP;
2424 if (q->unprep_rq_fn)
2425 q->unprep_rq_fn(q, req);
2426}
2427EXPORT_SYMBOL_GPL(blk_unprep_request);
2428
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429/*
2430 * queue lock must be held
2431 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002432static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002434 if (blk_rq_tagged(req))
2435 blk_queue_end_tag(req->q, req);
2436
James Bottomleyba396a62009-05-27 14:17:08 +02002437 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002439 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002440 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Mike Andersone78042e2008-10-30 02:16:20 -07002442 blk_delete_timer(req);
2443
James Bottomley28018c22010-07-01 19:49:17 +09002444 if (req->cmd_flags & REQ_DONTPREP)
2445 blk_unprep_request(req);
2446
2447
Jens Axboebc58ba92009-01-23 10:54:44 +01002448 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002451 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002452 else {
2453 if (blk_bidi_rq(req))
2454 __blk_put_request(req->next_rq->q, req->next_rq);
2455
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458}
2459
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002460/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002461 * blk_end_bidi_request - Complete a bidi request
2462 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002463 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002464 * @nr_bytes: number of bytes to complete @rq
2465 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002466 *
2467 * Description:
2468 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002469 * Drivers that supports bidi can safely call this member for any
2470 * type of request, bidi or uni. In the later case @bidi_bytes is
2471 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002472 *
2473 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002474 * %false - we are done with this request
2475 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002476 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002477static bool blk_end_bidi_request(struct request *rq, int error,
2478 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002479{
2480 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002481 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002482
Tejun Heo2e60e022009-04-23 11:05:18 +09002483 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2484 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002485
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002486 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002487 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002488 spin_unlock_irqrestore(q->queue_lock, flags);
2489
Tejun Heo2e60e022009-04-23 11:05:18 +09002490 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002491}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002492
2493/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002494 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2495 * @rq: the request to complete
2496 * @error: %0 for success, < %0 for error
2497 * @nr_bytes: number of bytes to complete @rq
2498 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002499 *
2500 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002501 * Identical to blk_end_bidi_request() except that queue lock is
2502 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002503 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002504 * Return:
2505 * %false - we are done with this request
2506 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002507 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002508bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002509 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002510{
Tejun Heo2e60e022009-04-23 11:05:18 +09002511 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2512 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002513
Tejun Heo2e60e022009-04-23 11:05:18 +09002514 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002515
Tejun Heo2e60e022009-04-23 11:05:18 +09002516 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002517}
2518
2519/**
2520 * blk_end_request - Helper function for drivers to complete the request.
2521 * @rq: the request being processed
2522 * @error: %0 for success, < %0 for error
2523 * @nr_bytes: number of bytes to complete
2524 *
2525 * Description:
2526 * Ends I/O on a number of bytes attached to @rq.
2527 * If @rq has leftover, sets it up for the next range of segments.
2528 *
2529 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002530 * %false - we are done with this request
2531 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002532 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002533bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002534{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002535 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002536}
Jens Axboe56ad1742009-07-28 22:11:24 +02002537EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002538
2539/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002540 * blk_end_request_all - Helper function for drives to finish the request.
2541 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002542 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002543 *
2544 * Description:
2545 * Completely finish @rq.
2546 */
2547void blk_end_request_all(struct request *rq, int error)
2548{
2549 bool pending;
2550 unsigned int bidi_bytes = 0;
2551
2552 if (unlikely(blk_bidi_rq(rq)))
2553 bidi_bytes = blk_rq_bytes(rq->next_rq);
2554
2555 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2556 BUG_ON(pending);
2557}
Jens Axboe56ad1742009-07-28 22:11:24 +02002558EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002559
2560/**
2561 * blk_end_request_cur - Helper function to finish the current request chunk.
2562 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002563 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002564 *
2565 * Description:
2566 * Complete the current consecutively mapped chunk from @rq.
2567 *
2568 * Return:
2569 * %false - we are done with this request
2570 * %true - still buffers pending for this request
2571 */
2572bool blk_end_request_cur(struct request *rq, int error)
2573{
2574 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2575}
Jens Axboe56ad1742009-07-28 22:11:24 +02002576EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002577
2578/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002579 * blk_end_request_err - Finish a request till the next failure boundary.
2580 * @rq: the request to finish till the next failure boundary for
2581 * @error: must be negative errno
2582 *
2583 * Description:
2584 * Complete @rq till the next failure boundary.
2585 *
2586 * Return:
2587 * %false - we are done with this request
2588 * %true - still buffers pending for this request
2589 */
2590bool blk_end_request_err(struct request *rq, int error)
2591{
2592 WARN_ON(error >= 0);
2593 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2594}
2595EXPORT_SYMBOL_GPL(blk_end_request_err);
2596
2597/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002598 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002599 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002600 * @error: %0 for success, < %0 for error
2601 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002602 *
2603 * Description:
2604 * Must be called with queue lock held unlike blk_end_request().
2605 *
2606 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002607 * %false - we are done with this request
2608 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002609 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002610bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002611{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002612 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002613}
Jens Axboe56ad1742009-07-28 22:11:24 +02002614EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002615
2616/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002617 * __blk_end_request_all - Helper function for drives to finish the request.
2618 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002619 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002620 *
2621 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002622 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002623 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002624void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002625{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002626 bool pending;
2627 unsigned int bidi_bytes = 0;
2628
2629 if (unlikely(blk_bidi_rq(rq)))
2630 bidi_bytes = blk_rq_bytes(rq->next_rq);
2631
2632 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2633 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002634}
Jens Axboe56ad1742009-07-28 22:11:24 +02002635EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002636
2637/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002638 * __blk_end_request_cur - Helper function to finish the current request chunk.
2639 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002640 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002641 *
2642 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002643 * Complete the current consecutively mapped chunk from @rq. Must
2644 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002645 *
2646 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002647 * %false - we are done with this request
2648 * %true - still buffers pending for this request
2649 */
2650bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002651{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002652 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002653}
Jens Axboe56ad1742009-07-28 22:11:24 +02002654EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002655
Tejun Heo80a761f2009-07-03 17:48:17 +09002656/**
2657 * __blk_end_request_err - Finish a request till the next failure boundary.
2658 * @rq: the request to finish till the next failure boundary for
2659 * @error: must be negative errno
2660 *
2661 * Description:
2662 * Complete @rq till the next failure boundary. Must be called
2663 * with queue lock held.
2664 *
2665 * Return:
2666 * %false - we are done with this request
2667 * %true - still buffers pending for this request
2668 */
2669bool __blk_end_request_err(struct request *rq, int error)
2670{
2671 WARN_ON(error >= 0);
2672 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2673}
2674EXPORT_SYMBOL_GPL(__blk_end_request_err);
2675
Jens Axboe86db1e22008-01-29 14:53:40 +01002676void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2677 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002679 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002680 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
David Woodhousefb2dce82008-08-05 18:01:53 +01002682 if (bio_has_data(bio)) {
2683 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002684 rq->buffer = bio_data(bio);
2685 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002686 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
NeilBrown66846572007-08-16 13:31:28 +02002689 if (bio->bi_bdev)
2690 rq->rq_disk = bio->bi_bdev->bd_disk;
2691}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002693#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2694/**
2695 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2696 * @rq: the request to be flushed
2697 *
2698 * Description:
2699 * Flush all pages in @rq.
2700 */
2701void rq_flush_dcache_pages(struct request *rq)
2702{
2703 struct req_iterator iter;
2704 struct bio_vec *bvec;
2705
2706 rq_for_each_segment(bvec, rq, iter)
2707 flush_dcache_page(bvec->bv_page);
2708}
2709EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2710#endif
2711
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002712/**
2713 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2714 * @q : the queue of the device being checked
2715 *
2716 * Description:
2717 * Check if underlying low-level drivers of a device are busy.
2718 * If the drivers want to export their busy state, they must set own
2719 * exporting function using blk_queue_lld_busy() first.
2720 *
2721 * Basically, this function is used only by request stacking drivers
2722 * to stop dispatching requests to underlying devices when underlying
2723 * devices are busy. This behavior helps more I/O merging on the queue
2724 * of the request stacking driver and prevents I/O throughput regression
2725 * on burst I/O load.
2726 *
2727 * Return:
2728 * 0 - Not busy (The request stacking driver should dispatch request)
2729 * 1 - Busy (The request stacking driver should stop dispatching request)
2730 */
2731int blk_lld_busy(struct request_queue *q)
2732{
2733 if (q->lld_busy_fn)
2734 return q->lld_busy_fn(q);
2735
2736 return 0;
2737}
2738EXPORT_SYMBOL_GPL(blk_lld_busy);
2739
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002740/**
2741 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2742 * @rq: the clone request to be cleaned up
2743 *
2744 * Description:
2745 * Free all bios in @rq for a cloned request.
2746 */
2747void blk_rq_unprep_clone(struct request *rq)
2748{
2749 struct bio *bio;
2750
2751 while ((bio = rq->bio) != NULL) {
2752 rq->bio = bio->bi_next;
2753
2754 bio_put(bio);
2755 }
2756}
2757EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2758
2759/*
2760 * Copy attributes of the original request to the clone request.
2761 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2762 */
2763static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2764{
2765 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002766 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002767 dst->cmd_type = src->cmd_type;
2768 dst->__sector = blk_rq_pos(src);
2769 dst->__data_len = blk_rq_bytes(src);
2770 dst->nr_phys_segments = src->nr_phys_segments;
2771 dst->ioprio = src->ioprio;
2772 dst->extra_len = src->extra_len;
2773}
2774
2775/**
2776 * blk_rq_prep_clone - Helper function to setup clone request
2777 * @rq: the request to be setup
2778 * @rq_src: original request to be cloned
2779 * @bs: bio_set that bios for clone are allocated from
2780 * @gfp_mask: memory allocation mask for bio
2781 * @bio_ctr: setup function to be called for each clone bio.
2782 * Returns %0 for success, non %0 for failure.
2783 * @data: private data to be passed to @bio_ctr
2784 *
2785 * Description:
2786 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2787 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2788 * are not copied, and copying such parts is the caller's responsibility.
2789 * Also, pages which the original bios are pointing to are not copied
2790 * and the cloned bios just point same pages.
2791 * So cloned bios must be completed before original bios, which means
2792 * the caller must complete @rq before @rq_src.
2793 */
2794int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2795 struct bio_set *bs, gfp_t gfp_mask,
2796 int (*bio_ctr)(struct bio *, struct bio *, void *),
2797 void *data)
2798{
2799 struct bio *bio, *bio_src;
2800
2801 if (!bs)
2802 bs = fs_bio_set;
2803
2804 blk_rq_init(NULL, rq);
2805
2806 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002807 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002808 if (!bio)
2809 goto free_and_out;
2810
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002811 if (bio_ctr && bio_ctr(bio, bio_src, data))
2812 goto free_and_out;
2813
2814 if (rq->bio) {
2815 rq->biotail->bi_next = bio;
2816 rq->biotail = bio;
2817 } else
2818 rq->bio = rq->biotail = bio;
2819 }
2820
2821 __blk_rq_prep_clone(rq, rq_src);
2822
2823 return 0;
2824
2825free_and_out:
2826 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002827 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002828 blk_rq_unprep_clone(rq);
2829
2830 return -ENOMEM;
2831}
2832EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2833
Jens Axboe18887ad2008-07-28 13:08:45 +02002834int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835{
2836 return queue_work(kblockd_workqueue, work);
2837}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838EXPORT_SYMBOL(kblockd_schedule_work);
2839
Vivek Goyale43473b2010-09-15 17:06:35 -04002840int kblockd_schedule_delayed_work(struct request_queue *q,
2841 struct delayed_work *dwork, unsigned long delay)
2842{
2843 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2844}
2845EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2846
Jens Axboe73c10102011-03-08 13:19:51 +01002847#define PLUG_MAGIC 0x91827364
2848
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002849/**
2850 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2851 * @plug: The &struct blk_plug that needs to be initialized
2852 *
2853 * Description:
2854 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2855 * pending I/O should the task end up blocking between blk_start_plug() and
2856 * blk_finish_plug(). This is important from a performance perspective, but
2857 * also ensures that we don't deadlock. For instance, if the task is blocking
2858 * for a memory allocation, memory reclaim could end up wanting to free a
2859 * page belonging to that request that is currently residing in our private
2860 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2861 * this kind of deadlock.
2862 */
Jens Axboe73c10102011-03-08 13:19:51 +01002863void blk_start_plug(struct blk_plug *plug)
2864{
2865 struct task_struct *tsk = current;
2866
2867 plug->magic = PLUG_MAGIC;
2868 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002869 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002870
2871 /*
2872 * If this is a nested plug, don't actually assign it. It will be
2873 * flushed on its own.
2874 */
2875 if (!tsk->plug) {
2876 /*
2877 * Store ordering should not be needed here, since a potential
2878 * preempt will imply a full memory barrier
2879 */
2880 tsk->plug = plug;
2881 }
2882}
2883EXPORT_SYMBOL(blk_start_plug);
2884
2885static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2886{
2887 struct request *rqa = container_of(a, struct request, queuelist);
2888 struct request *rqb = container_of(b, struct request, queuelist);
2889
Jianpeng Ma975927b2012-10-25 21:58:17 +02002890 return !(rqa->q < rqb->q ||
2891 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01002892}
2893
Jens Axboe49cac012011-04-16 13:51:05 +02002894/*
2895 * If 'from_schedule' is true, then postpone the dispatch of requests
2896 * until a safe kblockd context. We due this to avoid accidental big
2897 * additional stack usage in driver dispatch, in places where the originally
2898 * plugger did not intend it.
2899 */
Jens Axboef6603782011-04-15 15:49:07 +02002900static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002901 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002902 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002903{
Jens Axboe49cac012011-04-16 13:51:05 +02002904 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002905
Bart Van Assche70460572012-11-28 13:45:56 +01002906 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002907 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002908 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002909 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002910 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002911}
2912
NeilBrown74018dc2012-07-31 09:08:15 +02002913static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002914{
2915 LIST_HEAD(callbacks);
2916
Shaohua Li2a7d5552012-07-31 09:08:15 +02002917 while (!list_empty(&plug->cb_list)) {
2918 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002919
Shaohua Li2a7d5552012-07-31 09:08:15 +02002920 while (!list_empty(&callbacks)) {
2921 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002922 struct blk_plug_cb,
2923 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002924 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02002925 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002926 }
NeilBrown048c9372011-04-18 09:52:22 +02002927 }
2928}
2929
NeilBrown9cbb1752012-07-31 09:08:14 +02002930struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2931 int size)
2932{
2933 struct blk_plug *plug = current->plug;
2934 struct blk_plug_cb *cb;
2935
2936 if (!plug)
2937 return NULL;
2938
2939 list_for_each_entry(cb, &plug->cb_list, list)
2940 if (cb->callback == unplug && cb->data == data)
2941 return cb;
2942
2943 /* Not currently on the callback list */
2944 BUG_ON(size < sizeof(*cb));
2945 cb = kzalloc(size, GFP_ATOMIC);
2946 if (cb) {
2947 cb->data = data;
2948 cb->callback = unplug;
2949 list_add(&cb->list, &plug->cb_list);
2950 }
2951 return cb;
2952}
2953EXPORT_SYMBOL(blk_check_plugged);
2954
Jens Axboe49cac012011-04-16 13:51:05 +02002955void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002956{
2957 struct request_queue *q;
2958 unsigned long flags;
2959 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002960 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002961 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002962
2963 BUG_ON(plug->magic != PLUG_MAGIC);
2964
NeilBrown74018dc2012-07-31 09:08:15 +02002965 flush_plug_callbacks(plug, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002966 if (list_empty(&plug->list))
2967 return;
2968
NeilBrown109b8122011-04-11 14:13:10 +02002969 list_splice_init(&plug->list, &list);
2970
Jianpeng Ma422765c2013-01-11 14:46:09 +01002971 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01002972
2973 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002974 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002975
2976 /*
2977 * Save and disable interrupts here, to avoid doing it for every
2978 * queue lock we have to take.
2979 */
Jens Axboe73c10102011-03-08 13:19:51 +01002980 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002981 while (!list_empty(&list)) {
2982 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002983 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002984 BUG_ON(!rq->q);
2985 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02002986 /*
2987 * This drops the queue lock
2988 */
2989 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02002990 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002991 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002992 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01002993 spin_lock(q->queue_lock);
2994 }
Tejun Heo8ba61432011-12-14 00:33:37 +01002995
2996 /*
2997 * Short-circuit if @q is dead
2998 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002999 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003000 __blk_end_request_all(rq, -ENODEV);
3001 continue;
3002 }
3003
Jens Axboe73c10102011-03-08 13:19:51 +01003004 /*
3005 * rq is already accounted, so use raw insert
3006 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003007 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3008 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3009 else
3010 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003011
3012 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003013 }
3014
Jens Axboe99e22592011-04-18 09:59:55 +02003015 /*
3016 * This drops the queue lock
3017 */
3018 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003019 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003020
Jens Axboe73c10102011-03-08 13:19:51 +01003021 local_irq_restore(flags);
3022}
Jens Axboe73c10102011-03-08 13:19:51 +01003023
3024void blk_finish_plug(struct blk_plug *plug)
3025{
Jens Axboef6603782011-04-15 15:49:07 +02003026 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003027
3028 if (plug == current->plug)
3029 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003030}
3031EXPORT_SYMBOL(blk_finish_plug);
3032
Lin Ming6c954662013-03-23 11:42:26 +08003033#ifdef CONFIG_PM_RUNTIME
3034/**
3035 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3036 * @q: the queue of the device
3037 * @dev: the device the queue belongs to
3038 *
3039 * Description:
3040 * Initialize runtime-PM-related fields for @q and start auto suspend for
3041 * @dev. Drivers that want to take advantage of request-based runtime PM
3042 * should call this function after @dev has been initialized, and its
3043 * request queue @q has been allocated, and runtime PM for it can not happen
3044 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3045 * cases, driver should call this function before any I/O has taken place.
3046 *
3047 * This function takes care of setting up using auto suspend for the device,
3048 * the autosuspend delay is set to -1 to make runtime suspend impossible
3049 * until an updated value is either set by user or by driver. Drivers do
3050 * not need to touch other autosuspend settings.
3051 *
3052 * The block layer runtime PM is request based, so only works for drivers
3053 * that use request as their IO unit instead of those directly use bio's.
3054 */
3055void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3056{
3057 q->dev = dev;
3058 q->rpm_status = RPM_ACTIVE;
3059 pm_runtime_set_autosuspend_delay(q->dev, -1);
3060 pm_runtime_use_autosuspend(q->dev);
3061}
3062EXPORT_SYMBOL(blk_pm_runtime_init);
3063
3064/**
3065 * blk_pre_runtime_suspend - Pre runtime suspend check
3066 * @q: the queue of the device
3067 *
3068 * Description:
3069 * This function will check if runtime suspend is allowed for the device
3070 * by examining if there are any requests pending in the queue. If there
3071 * are requests pending, the device can not be runtime suspended; otherwise,
3072 * the queue's status will be updated to SUSPENDING and the driver can
3073 * proceed to suspend the device.
3074 *
3075 * For the not allowed case, we mark last busy for the device so that
3076 * runtime PM core will try to autosuspend it some time later.
3077 *
3078 * This function should be called near the start of the device's
3079 * runtime_suspend callback.
3080 *
3081 * Return:
3082 * 0 - OK to runtime suspend the device
3083 * -EBUSY - Device should not be runtime suspended
3084 */
3085int blk_pre_runtime_suspend(struct request_queue *q)
3086{
3087 int ret = 0;
3088
3089 spin_lock_irq(q->queue_lock);
3090 if (q->nr_pending) {
3091 ret = -EBUSY;
3092 pm_runtime_mark_last_busy(q->dev);
3093 } else {
3094 q->rpm_status = RPM_SUSPENDING;
3095 }
3096 spin_unlock_irq(q->queue_lock);
3097 return ret;
3098}
3099EXPORT_SYMBOL(blk_pre_runtime_suspend);
3100
3101/**
3102 * blk_post_runtime_suspend - Post runtime suspend processing
3103 * @q: the queue of the device
3104 * @err: return value of the device's runtime_suspend function
3105 *
3106 * Description:
3107 * Update the queue's runtime status according to the return value of the
3108 * device's runtime suspend function and mark last busy for the device so
3109 * that PM core will try to auto suspend the device at a later time.
3110 *
3111 * This function should be called near the end of the device's
3112 * runtime_suspend callback.
3113 */
3114void blk_post_runtime_suspend(struct request_queue *q, int err)
3115{
3116 spin_lock_irq(q->queue_lock);
3117 if (!err) {
3118 q->rpm_status = RPM_SUSPENDED;
3119 } else {
3120 q->rpm_status = RPM_ACTIVE;
3121 pm_runtime_mark_last_busy(q->dev);
3122 }
3123 spin_unlock_irq(q->queue_lock);
3124}
3125EXPORT_SYMBOL(blk_post_runtime_suspend);
3126
3127/**
3128 * blk_pre_runtime_resume - Pre runtime resume processing
3129 * @q: the queue of the device
3130 *
3131 * Description:
3132 * Update the queue's runtime status to RESUMING in preparation for the
3133 * runtime resume of the device.
3134 *
3135 * This function should be called near the start of the device's
3136 * runtime_resume callback.
3137 */
3138void blk_pre_runtime_resume(struct request_queue *q)
3139{
3140 spin_lock_irq(q->queue_lock);
3141 q->rpm_status = RPM_RESUMING;
3142 spin_unlock_irq(q->queue_lock);
3143}
3144EXPORT_SYMBOL(blk_pre_runtime_resume);
3145
3146/**
3147 * blk_post_runtime_resume - Post runtime resume processing
3148 * @q: the queue of the device
3149 * @err: return value of the device's runtime_resume function
3150 *
3151 * Description:
3152 * Update the queue's runtime status according to the return value of the
3153 * device's runtime_resume function. If it is successfully resumed, process
3154 * the requests that are queued into the device's queue when it is resuming
3155 * and then mark last busy and initiate autosuspend for it.
3156 *
3157 * This function should be called near the end of the device's
3158 * runtime_resume callback.
3159 */
3160void blk_post_runtime_resume(struct request_queue *q, int err)
3161{
3162 spin_lock_irq(q->queue_lock);
3163 if (!err) {
3164 q->rpm_status = RPM_ACTIVE;
3165 __blk_run_queue(q);
3166 pm_runtime_mark_last_busy(q->dev);
3167 pm_runtime_autosuspend(q->dev);
3168 } else {
3169 q->rpm_status = RPM_SUSPENDED;
3170 }
3171 spin_unlock_irq(q->queue_lock);
3172}
3173EXPORT_SYMBOL(blk_post_runtime_resume);
3174#endif
3175
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176int __init blk_dev_init(void)
3177{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003178 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3179 sizeof(((struct request *)0)->cmd_flags));
3180
Tejun Heo89b90be2011-01-03 15:01:47 +01003181 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3182 kblockd_workqueue = alloc_workqueue("kblockd",
3183 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 if (!kblockd_workqueue)
3185 panic("Failed to create kblockd\n");
3186
3187 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003188 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189
Jens Axboe8324aa92008-01-29 14:51:59 +01003190 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003191 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 return 0;
3194}