blob: 074b758efc42cf116d61f4755f0107f491890980 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080029#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010030#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020031#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070032#include <linux/ratelimit.h>
Li Zefan55782132009-06-09 13:43:05 +080033
34#define CREATE_TRACE_POINTS
35#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jens Axboe8324aa92008-01-29 14:51:59 +010037#include "blk.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080038#include "blk-cgroup.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010039
Mike Snitzerd07335e2010-11-16 12:52:38 +010040EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020041EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
NeilBrowncbae8d42012-12-14 20:49:27 +010042EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010043
Tejun Heoa73f7302011-12-14 00:33:37 +010044DEFINE_IDA(blk_queue_ida);
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/*
47 * For the allocated request tables
48 */
Adrian Bunk5ece6c52008-02-18 13:45:51 +010049static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51/*
52 * For queue allocation
53 */
Jens Axboe6728cb02008-01-31 13:03:55 +010054struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 * Controlling structure to kblockd
58 */
Jens Axboeff856ba2006-01-09 16:02:34 +010059static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Jens Axboe26b82562008-01-29 13:54:41 +010061static void drive_stat_acct(struct request *rq, int new_io)
62{
Jens Axboe28f13702008-05-07 10:15:46 +020063 struct hd_struct *part;
Jens Axboe26b82562008-01-29 13:54:41 +010064 int rw = rq_data_dir(rq);
Tejun Heoc9959052008-08-25 19:47:21 +090065 int cpu;
Jens Axboe26b82562008-01-29 13:54:41 +010066
Jens Axboec2553b52009-04-24 08:10:11 +020067 if (!blk_do_io_stat(rq))
Jens Axboe26b82562008-01-29 13:54:41 +010068 return;
69
Tejun Heo074a7ac2008-08-25 19:56:14 +090070 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +090071
Jerome Marchand09e099d2011-01-05 16:57:38 +010072 if (!new_io) {
73 part = rq->part;
Tejun Heo074a7ac2008-08-25 19:56:14 +090074 part_stat_inc(cpu, part, merges[rw]);
Jerome Marchand09e099d2011-01-05 16:57:38 +010075 } else {
76 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
Jens Axboe6c23a962011-01-07 08:43:37 +010077 if (!hd_struct_try_get(part)) {
Jerome Marchand09e099d2011-01-05 16:57:38 +010078 /*
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
81 *
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
85 */
86 part = &rq->rq_disk->part0;
Jens Axboe6c23a962011-01-07 08:43:37 +010087 hd_struct_get(part);
Jerome Marchand09e099d2011-01-05 16:57:38 +010088 }
Tejun Heo074a7ac2008-08-25 19:56:14 +090089 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +020090 part_inc_in_flight(part, rw);
Jerome Marchand09e099d2011-01-05 16:57:38 +010091 rq->part = part;
Jens Axboe26b82562008-01-29 13:54:41 +010092 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +020093
Tejun Heo074a7ac2008-08-25 19:56:14 +090094 part_stat_unlock();
Jens Axboe26b82562008-01-29 13:54:41 +010095}
96
Jens Axboe8324aa92008-01-29 14:51:59 +010097void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
99 int nr;
100
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
103 nr = q->nr_requests;
104 q->nr_congestion_on = nr;
105
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 if (nr < 1)
108 nr = 1;
109 q->nr_congestion_off = nr;
110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/**
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * @bdev: device
115 *
116 * Locates the passed device's request queue and returns the address of its
117 * backing_dev_info
118 *
119 * Will return NULL if the request queue cannot be located.
120 */
121struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122{
123 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200124 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126 if (q)
127 ret = &q->backing_dev_info;
128 return ret;
129}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130EXPORT_SYMBOL(blk_get_backing_dev_info);
131
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200132void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200134 memset(rq, 0, sizeof(*rq));
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700137 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200138 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100139 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900140 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200141 INIT_HLIST_NODE(&rq->hash);
142 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200143 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800144 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100145 rq->tag = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100146 rq->ref_count = 1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900147 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700148 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100149 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200151EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
NeilBrown5bb23a62007-09-27 12:46:13 +0200153static void req_bio_endio(struct request *rq, struct bio *bio,
154 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100155{
Tejun Heo143a87f2011-01-25 12:43:52 +0100156 if (error)
157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
158 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
159 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100160
Tejun Heo143a87f2011-01-25 12:43:52 +0100161 if (unlikely(nbytes > bio->bi_size)) {
162 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
163 __func__, nbytes, bio->bi_size);
164 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +0200165 }
Tejun Heo143a87f2011-01-25 12:43:52 +0100166
167 if (unlikely(rq->cmd_flags & REQ_QUIET))
168 set_bit(BIO_QUIET, &bio->bi_flags);
169
170 bio->bi_size -= nbytes;
171 bio->bi_sector += (nbytes >> 9);
172
173 if (bio_integrity(bio))
174 bio_integrity_advance(bio, nbytes);
175
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
178 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100179}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181void blk_dump_rq_flags(struct request *rq, char *msg)
182{
183 int bit;
184
Jens Axboe6728cb02008-01-31 13:03:55 +0100185 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200186 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
187 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Tejun Heo83096eb2009-05-07 22:24:39 +0900189 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq),
191 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900193 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200195 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100196 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200197 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 printk("%02x ", rq->cmd[bit]);
199 printk("\n");
200 }
201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_dump_rq_flags);
203
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500204static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200205{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500206 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200207
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500208 q = container_of(work, struct request_queue, delay_work.work);
209 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200210 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500211 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 *
219 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100222 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500223 */
224void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Bart Van Assche70460572012-11-28 13:45:56 +0100226 if (likely(!blk_queue_dead(q)))
227 queue_delayed_work(kblockd_workqueue, &q->delay_work,
228 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500230EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/**
233 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200234 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 *
236 * Description:
237 * blk_start_queue() will clear the stop flag on the queue, and call
238 * the request_fn for the queue if it was in a stopped state when
239 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200241void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200243 WARN_ON(!irqs_disabled());
244
Nick Piggin75ad23b2008-04-29 14:48:33 +0200245 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200246 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248EXPORT_SYMBOL(blk_start_queue);
249
250/**
251 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200252 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 *
254 * Description:
255 * The Linux block layer assumes that a block driver will consume all
256 * entries on the request queue when the request_fn strategy is called.
257 * Often this will not happen, because of hardware limitations (queue
258 * depth settings). If a device driver gets a 'queue full' response,
259 * or if it simply chooses not to queue more I/O at one point, it can
260 * call this function to prevent the request_fn from being called until
261 * the driver has signalled it's ready to go again. This happens by calling
262 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200264void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Tejun Heo136b5722012-08-21 13:18:24 -0700266 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200267 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269EXPORT_SYMBOL(blk_stop_queue);
270
271/**
272 * blk_sync_queue - cancel any pending callbacks on a queue
273 * @q: the queue
274 *
275 * Description:
276 * The block layer may perform asynchronous callback activity
277 * on a queue, such as calling the unplug function after a timeout.
278 * A block device may call blk_sync_queue to ensure that any
279 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200280 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 * that its ->make_request_fn will not re-add plugging prior to calling
282 * this function.
283 *
Vivek Goyalda527772011-03-02 19:05:33 -0500284 * This function does not cancel any asynchronous activity arising
285 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800286 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500287 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 */
289void blk_sync_queue(struct request_queue *q)
290{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100291 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500292 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293}
294EXPORT_SYMBOL(blk_sync_queue);
295
296/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100297 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
298 * @q: The queue to run
299 *
300 * Description:
301 * Invoke request handling on a queue if there are any pending requests.
302 * May be used to restart request handling after a request has completed.
303 * This variant runs the queue whether or not the queue has been
304 * stopped. Must be called with the queue lock held and interrupts
305 * disabled. See also @blk_run_queue.
306 */
307inline void __blk_run_queue_uncond(struct request_queue *q)
308{
309 if (unlikely(blk_queue_dead(q)))
310 return;
311
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100312 /*
313 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
314 * the queue lock internally. As a result multiple threads may be
315 * running such a request function concurrently. Keep track of the
316 * number of active request_fn invocations such that blk_drain_queue()
317 * can wait until all these request_fn calls have finished.
318 */
319 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100320 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100321 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100322}
323
324/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200325 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200327 *
328 * Description:
329 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200330 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200332void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Tejun Heoa538cd02009-04-23 11:05:17 +0900334 if (unlikely(blk_queue_stopped(q)))
335 return;
336
Bart Van Asschec246e802012-12-06 14:32:01 +0100337 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200338}
339EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200340
Nick Piggin75ad23b2008-04-29 14:48:33 +0200341/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200342 * blk_run_queue_async - run a single device queue in workqueue context
343 * @q: The queue to run
344 *
345 * Description:
346 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100347 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200348 */
349void blk_run_queue_async(struct request_queue *q)
350{
Bart Van Assche70460572012-11-28 13:45:56 +0100351 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700352 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200353}
Jens Axboec21e6be2011-04-19 13:32:46 +0200354EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200355
356/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200357 * blk_run_queue - run a single device queue
358 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200359 *
360 * Description:
361 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900362 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200363 */
364void blk_run_queue(struct request_queue *q)
365{
366 unsigned long flags;
367
368 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200369 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 spin_unlock_irqrestore(q->queue_lock, flags);
371}
372EXPORT_SYMBOL(blk_run_queue);
373
Jens Axboe165125e2007-07-24 09:28:11 +0200374void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500375{
376 kobject_put(&q->kobj);
377}
Jens Axboed86e0e82011-05-27 07:44:43 +0200378EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500379
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200380/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100381 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200382 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200383 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200384 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200385 * Drain requests from @q. If @drain_all is set, all requests are drained.
386 * If not, only ELVPRIV requests are drained. The caller is responsible
387 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200388 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100389static void __blk_drain_queue(struct request_queue *q, bool drain_all)
390 __releases(q->queue_lock)
391 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200392{
Asias He458f27a2012-06-15 08:45:25 +0200393 int i;
394
Bart Van Assche807592a2012-11-28 13:43:38 +0100395 lockdep_assert_held(q->queue_lock);
396
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200397 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100398 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200399
Tejun Heob855b042012-03-06 21:24:55 +0100400 /*
401 * The caller might be trying to drain @q before its
402 * elevator is initialized.
403 */
404 if (q->elevator)
405 elv_drain_elevator(q);
406
Tejun Heo5efd6112012-03-05 13:15:12 -0800407 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200408
Tejun Heo4eabc942011-12-15 20:03:04 +0100409 /*
410 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100411 * driver init after queue creation or is not yet fully
412 * active yet. Some drivers (e.g. fd and loop) get unhappy
413 * in such cases. Kick queue iff dispatch queue has
414 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100415 */
Tejun Heob855b042012-03-06 21:24:55 +0100416 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100417 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200418
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700419 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100420 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100421
422 /*
423 * Unfortunately, requests are queued at and tracked from
424 * multiple places and there's no single counter which can
425 * be drained. Check all the queues and counters.
426 */
427 if (drain_all) {
428 drain |= !list_empty(&q->queue_head);
429 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700430 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100431 drain |= q->in_flight[i];
432 drain |= !list_empty(&q->flush_queue[i]);
433 }
434 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200435
Tejun Heo481a7d62011-12-14 00:33:37 +0100436 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200437 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100438
439 spin_unlock_irq(q->queue_lock);
440
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200441 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100442
443 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200444 }
Asias He458f27a2012-06-15 08:45:25 +0200445
446 /*
447 * With queue marked dead, any woken up waiter will fail the
448 * allocation path, so the wakeup chaining is lost and we're
449 * left with hung waiters. We need to wake up those waiters.
450 */
451 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700452 struct request_list *rl;
453
Tejun Heoa0516612012-06-26 15:05:44 -0700454 blk_queue_for_each_rl(rl, q)
455 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
456 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200457 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200458}
459
Tejun Heoc9a929d2011-10-19 14:42:16 +0200460/**
Tejun Heod7325802012-03-05 13:14:58 -0800461 * blk_queue_bypass_start - enter queue bypass mode
462 * @q: queue of interest
463 *
464 * In bypass mode, only the dispatch FIFO queue of @q is used. This
465 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800466 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700467 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
468 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800469 */
470void blk_queue_bypass_start(struct request_queue *q)
471{
Tejun Heob82d4b12012-04-13 13:11:31 -0700472 bool drain;
473
Tejun Heod7325802012-03-05 13:14:58 -0800474 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700475 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800476 queue_flag_set(QUEUE_FLAG_BYPASS, q);
477 spin_unlock_irq(q->queue_lock);
478
Tejun Heob82d4b12012-04-13 13:11:31 -0700479 if (drain) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100480 spin_lock_irq(q->queue_lock);
481 __blk_drain_queue(q, false);
482 spin_unlock_irq(q->queue_lock);
483
Tejun Heob82d4b12012-04-13 13:11:31 -0700484 /* ensure blk_queue_bypass() is %true inside RCU read lock */
485 synchronize_rcu();
486 }
Tejun Heod7325802012-03-05 13:14:58 -0800487}
488EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
489
490/**
491 * blk_queue_bypass_end - leave queue bypass mode
492 * @q: queue of interest
493 *
494 * Leave bypass mode and restore the normal queueing behavior.
495 */
496void blk_queue_bypass_end(struct request_queue *q)
497{
498 spin_lock_irq(q->queue_lock);
499 if (!--q->bypass_depth)
500 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
501 WARN_ON_ONCE(q->bypass_depth < 0);
502 spin_unlock_irq(q->queue_lock);
503}
504EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
505
506/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200507 * blk_cleanup_queue - shutdown a request queue
508 * @q: request queue to shutdown
509 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100510 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
511 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500512 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100513void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500514{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200515 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de2008-09-18 09:22:54 -0700516
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100517 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500518 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100519 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200520 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800521
Tejun Heo80fd9972012-04-13 14:50:53 -0700522 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100523 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700524 * that, unlike blk_queue_bypass_start(), we aren't performing
525 * synchronize_rcu() after entering bypass mode to avoid the delay
526 * as some drivers create and destroy a lot of queues while
527 * probing. This is still safe because blk_release_queue() will be
528 * called only after the queue refcnt drops to zero and nothing,
529 * RCU or not, would be traversing the queue by then.
530 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800531 q->bypass_depth++;
532 queue_flag_set(QUEUE_FLAG_BYPASS, q);
533
Tejun Heoc9a929d2011-10-19 14:42:16 +0200534 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
535 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100536 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200537 spin_unlock_irq(lock);
538 mutex_unlock(&q->sysfs_lock);
539
Bart Van Asschec246e802012-12-06 14:32:01 +0100540 /*
541 * Drain all requests queued before DYING marking. Set DEAD flag to
542 * prevent that q->request_fn() gets invoked after draining finished.
543 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100544 spin_lock_irq(lock);
545 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100546 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100547 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200548
549 /* @q won't process any more request, flush async actions */
550 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
551 blk_sync_queue(q);
552
Asias He5e5cfac2012-05-24 23:28:52 +0800553 spin_lock_irq(lock);
554 if (q->queue_lock != &q->__queue_lock)
555 q->queue_lock = &q->__queue_lock;
556 spin_unlock_irq(lock);
557
Tejun Heoc9a929d2011-10-19 14:42:16 +0200558 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500559 blk_put_queue(q);
560}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561EXPORT_SYMBOL(blk_cleanup_queue);
562
Tejun Heo5b788ce2012-06-04 20:40:59 -0700563int blk_init_rl(struct request_list *rl, struct request_queue *q,
564 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400566 if (unlikely(rl->rq_pool))
567 return 0;
568
Tejun Heo5b788ce2012-06-04 20:40:59 -0700569 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200570 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
571 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200572 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
573 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Christoph Lameter19460892005-06-23 00:08:19 -0700575 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700576 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700577 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (!rl->rq_pool)
579 return -ENOMEM;
580
581 return 0;
582}
583
Tejun Heo5b788ce2012-06-04 20:40:59 -0700584void blk_exit_rl(struct request_list *rl)
585{
586 if (rl->rq_pool)
587 mempool_destroy(rl->rq_pool);
588}
589
Jens Axboe165125e2007-07-24 09:28:11 +0200590struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100592 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700593}
594EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Jens Axboe165125e2007-07-24 09:28:11 +0200596struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700597{
Jens Axboe165125e2007-07-24 09:28:11 +0200598 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700599 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700600
Jens Axboe8324aa92008-01-29 14:51:59 +0100601 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700602 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (!q)
604 return NULL;
605
Dan Carpenter00380a42012-03-23 09:58:54 +0100606 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100607 if (q->id < 0)
608 goto fail_q;
609
Jens Axboe0989a022009-06-12 14:42:56 +0200610 q->backing_dev_info.ra_pages =
611 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
612 q->backing_dev_info.state = 0;
613 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200614 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100615 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200616
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700617 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100618 if (err)
619 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700620
Matthew Garrett31373d02010-04-06 14:25:14 +0200621 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
622 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700623 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100624 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700625 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100626 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800627#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800628 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800629#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100630 INIT_LIST_HEAD(&q->flush_queue[0]);
631 INIT_LIST_HEAD(&q->flush_queue[1]);
632 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500633 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500634
Jens Axboe8324aa92008-01-29 14:51:59 +0100635 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Al Viro483f4af2006-03-18 18:34:37 -0500637 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700638 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500639
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500640 /*
641 * By default initialize queue_lock to internal lock and driver can
642 * override it later if need be.
643 */
644 q->queue_lock = &q->__queue_lock;
645
Tejun Heob82d4b12012-04-13 13:11:31 -0700646 /*
647 * A queue starts its life with bypass turned on to avoid
648 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700649 * init. The initial bypass will be finished when the queue is
650 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700651 */
652 q->bypass_depth = 1;
653 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
654
Tejun Heo5efd6112012-03-05 13:15:12 -0800655 if (blkcg_init_queue(q))
Tejun Heof51b8022012-03-05 13:15:05 -0800656 goto fail_id;
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100659
660fail_id:
661 ida_simple_remove(&blk_queue_ida, q->id);
662fail_q:
663 kmem_cache_free(blk_requestq_cachep, q);
664 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
Christoph Lameter19460892005-06-23 00:08:19 -0700666EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668/**
669 * blk_init_queue - prepare a request queue for use with a block device
670 * @rfn: The function to be called to process requests that have been
671 * placed on the queue.
672 * @lock: Request queue spin lock
673 *
674 * Description:
675 * If a block device wishes to use the standard request handling procedures,
676 * which sorts requests and coalesces adjacent requests, then it must
677 * call blk_init_queue(). The function @rfn will be called when there
678 * are requests on the queue that need to be processed. If the device
679 * supports plugging, then @rfn may not be called immediately when requests
680 * are available on the queue, but may be called at some time later instead.
681 * Plugged queues are generally unplugged when a buffer belonging to one
682 * of the requests on the queue is needed, or due to memory pressure.
683 *
684 * @rfn is not required, or even expected, to remove all requests off the
685 * queue, but only as many as it can handle at a time. If it does leave
686 * requests on the queue, it is responsible for arranging that the requests
687 * get dealt with eventually.
688 *
689 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200690 * request queue; this lock will be taken also from interrupt context, so irq
691 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200693 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 * it didn't succeed.
695 *
696 * Note:
697 * blk_init_queue() must be paired with a blk_cleanup_queue() call
698 * when the block device is deactivated (such as at module unload).
699 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700700
Jens Axboe165125e2007-07-24 09:28:11 +0200701struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100703 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700704}
705EXPORT_SYMBOL(blk_init_queue);
706
Jens Axboe165125e2007-07-24 09:28:11 +0200707struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700708blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
709{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600710 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600712 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
713 if (!uninit_q)
714 return NULL;
715
Mike Snitzer51514122011-11-23 10:59:13 +0100716 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600717 if (!q)
718 blk_cleanup_queue(uninit_q);
719
720 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200721}
722EXPORT_SYMBOL(blk_init_queue_node);
723
724struct request_queue *
725blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
726 spinlock_t *lock)
727{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 if (!q)
729 return NULL;
730
Tejun Heoa0516612012-06-26 15:05:44 -0700731 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500732 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900736 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700737 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500738
739 /* Override internal queue lock with supplied lock pointer */
740 if (lock)
741 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
Jens Axboef3b144a2009-03-06 08:48:33 +0100743 /*
744 * This also sets hw/phys segments, boundary and size
745 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200746 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Alan Stern44ec9542007-02-20 11:01:57 -0500748 q->sg_reserved_size = INT_MAX;
749
Tejun Heob82d4b12012-04-13 13:11:31 -0700750 /* init elevator */
751 if (elevator_init(q, NULL))
752 return NULL;
Tejun Heob82d4b12012-04-13 13:11:31 -0700753 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
Mike Snitzer51514122011-11-23 10:59:13 +0100755EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Tejun Heo09ac46c2011-12-14 00:33:38 +0100757bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100759 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100760 __blk_get_queue(q);
761 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763
Tejun Heo09ac46c2011-12-14 00:33:38 +0100764 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
Jens Axboed86e0e82011-05-27 07:44:43 +0200766EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Tejun Heo5b788ce2012-06-04 20:40:59 -0700768static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100770 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700771 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100772 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100773 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100774 }
775
Tejun Heo5b788ce2012-06-04 20:40:59 -0700776 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779/*
780 * ioc_batching returns true if the ioc is a valid batching request and
781 * should be given priority access to a request.
782 */
Jens Axboe165125e2007-07-24 09:28:11 +0200783static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
785 if (!ioc)
786 return 0;
787
788 /*
789 * Make sure the process is able to allocate at least 1 request
790 * even if the batch times out, otherwise we could theoretically
791 * lose wakeups.
792 */
793 return ioc->nr_batch_requests == q->nr_batching ||
794 (ioc->nr_batch_requests > 0
795 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
796}
797
798/*
799 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
800 * will cause the process to be a "batcher" on all queues in the system. This
801 * is the behaviour we want though - once it gets a wakeup it should be given
802 * a nice run.
803 */
Jens Axboe165125e2007-07-24 09:28:11 +0200804static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
806 if (!ioc || ioc_batching(q, ioc))
807 return;
808
809 ioc->nr_batch_requests = q->nr_batching;
810 ioc->last_waited = jiffies;
811}
812
Tejun Heo5b788ce2012-06-04 20:40:59 -0700813static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700815 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Tejun Heoa0516612012-06-26 15:05:44 -0700817 /*
818 * bdi isn't aware of blkcg yet. As all async IOs end up root
819 * blkcg anyway, just use root blkcg state.
820 */
821 if (rl == &q->root_rl &&
822 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200823 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Jens Axboe1faa16d2009-04-06 14:48:01 +0200825 if (rl->count[sync] + 1 <= q->nr_requests) {
826 if (waitqueue_active(&rl->wait[sync]))
827 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Tejun Heo5b788ce2012-06-04 20:40:59 -0700829 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
831}
832
833/*
834 * A request has just been released. Account for it, update the full and
835 * congestion status, wake up any waiters. Called under q->queue_lock.
836 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700837static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700839 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200840 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700842 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200843 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200844 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700845 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Tejun Heo5b788ce2012-06-04 20:40:59 -0700847 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Jens Axboe1faa16d2009-04-06 14:48:01 +0200849 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700850 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100854 * Determine if elevator data should be initialized when allocating the
855 * request associated with @bio.
856 */
857static bool blk_rq_should_init_elevator(struct bio *bio)
858{
859 if (!bio)
860 return true;
861
862 /*
863 * Flush requests do not use the elevator so skip initialization.
864 * This allows a request to share the flush and elevator data.
865 */
866 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
867 return false;
868
869 return true;
870}
871
Tejun Heoda8303c2011-10-19 14:33:05 +0200872/**
Tejun Heo852c7882012-03-05 13:15:27 -0800873 * rq_ioc - determine io_context for request allocation
874 * @bio: request being allocated is for this bio (can be %NULL)
875 *
876 * Determine io_context to use for request allocation for @bio. May return
877 * %NULL if %current->io_context doesn't exist.
878 */
879static struct io_context *rq_ioc(struct bio *bio)
880{
881#ifdef CONFIG_BLK_CGROUP
882 if (bio && bio->bi_ioc)
883 return bio->bi_ioc;
884#endif
885 return current->io_context;
886}
887
888/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700889 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700890 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200891 * @rw_flags: RW and SYNC flags
892 * @bio: bio to allocate request for (can be %NULL)
893 * @gfp_mask: allocation mask
894 *
895 * Get a free request from @q. This function may fail under memory
896 * pressure or if @q is dead.
897 *
898 * Must be callled with @q->queue_lock held and,
899 * Returns %NULL on failure, with @q->queue_lock held.
900 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700902static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700903 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700905 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800906 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700907 struct elevator_type *et = q->elevator->type;
908 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100909 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200910 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200911 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100913 if (unlikely(blk_queue_dying(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200914 return NULL;
915
Jens Axboe7749a8d2006-12-13 13:02:26 +0100916 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100917 if (may_queue == ELV_MQUEUE_NO)
918 goto rq_starved;
919
Jens Axboe1faa16d2009-04-06 14:48:01 +0200920 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
921 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100922 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100923 * The queue will fill after this allocation, so set
924 * it as full, and mark this process as "batching".
925 * This process will be allowed to complete a batch of
926 * requests, others will be blocked.
927 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700928 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100929 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700930 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100931 } else {
932 if (may_queue != ELV_MQUEUE_MUST
933 && !ioc_batching(q, ioc)) {
934 /*
935 * The queue is full and the allocating
936 * process is not a "batcher", and not
937 * exempted by the IO scheduler
938 */
Tejun Heob6792812012-03-05 13:15:23 -0800939 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100940 }
941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 }
Tejun Heoa0516612012-06-26 15:05:44 -0700943 /*
944 * bdi isn't aware of blkcg yet. As all async IOs end up
945 * root blkcg anyway, just use root blkcg state.
946 */
947 if (rl == &q->root_rl)
948 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 }
950
Jens Axboe082cf692005-06-28 16:35:11 +0200951 /*
952 * Only allow batching queuers to allocate up to 50% over the defined
953 * limit of requests, otherwise we could have thousands of requests
954 * allocated with any setting of ->nr_requests
955 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200956 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800957 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100958
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700959 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200960 rl->count[is_sync]++;
961 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200962
Tejun Heof1f8cc92011-12-14 00:33:42 +0100963 /*
964 * Decide whether the new request will be managed by elevator. If
965 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
966 * prevent the current elevator from being destroyed until the new
967 * request is freed. This guarantees icq's won't be destroyed and
968 * makes creating new ones safe.
969 *
970 * Also, lookup icq while holding queue_lock. If it doesn't exist,
971 * it will be created after releasing queue_lock.
972 */
Tejun Heod7325802012-03-05 13:14:58 -0800973 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200974 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700975 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100976 if (et->icq_cache && ioc)
977 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100978 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200979
Jens Axboef253b862010-10-24 22:06:02 +0200980 if (blk_queue_io_stat(q))
981 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 spin_unlock_irq(q->queue_lock);
983
Tejun Heo29e2b092012-04-19 16:29:21 -0700984 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700985 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700986 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800987 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Tejun Heo29e2b092012-04-19 16:29:21 -0700989 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700990 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700991 rq->cmd_flags = rw_flags | REQ_ALLOCED;
992
Tejun Heoaaf7c682012-04-19 16:29:22 -0700993 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700994 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700995 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700996 if (ioc)
997 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700998 if (!icq)
999 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001000 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001001
1002 rq->elv.icq = icq;
1003 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1004 goto fail_elvpriv;
1005
1006 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001007 if (icq)
1008 get_io_context(icq->ioc);
1009 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001010out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001011 /*
1012 * ioc may be NULL here, and ioc_batching will be false. That's
1013 * OK, if the queue is under the request limit then requests need
1014 * not count toward the nr_batch_requests limit. There will always
1015 * be some limit enforced by BLK_BATCH_TIME.
1016 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 if (ioc_batching(q, ioc))
1018 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001019
Jens Axboe1faa16d2009-04-06 14:48:01 +02001020 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001022
Tejun Heoaaf7c682012-04-19 16:29:22 -07001023fail_elvpriv:
1024 /*
1025 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1026 * and may fail indefinitely under memory pressure and thus
1027 * shouldn't stall IO. Treat this request as !elvpriv. This will
1028 * disturb iosched and blkcg but weird is bettern than dead.
1029 */
1030 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1031 dev_name(q->backing_dev_info.dev));
1032
1033 rq->cmd_flags &= ~REQ_ELVPRIV;
1034 rq->elv.icq = NULL;
1035
1036 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001037 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001038 spin_unlock_irq(q->queue_lock);
1039 goto out;
1040
Tejun Heob6792812012-03-05 13:15:23 -08001041fail_alloc:
1042 /*
1043 * Allocation failed presumably due to memory. Undo anything we
1044 * might have messed up.
1045 *
1046 * Allocating task should really be put onto the front of the wait
1047 * queue, but this is pretty rare.
1048 */
1049 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001050 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001051
1052 /*
1053 * in the very unlikely event that allocation failed and no
1054 * requests for this direction was pending, mark us starved so that
1055 * freeing of a request in the other direction will notice
1056 * us. another possible fix would be to split the rq mempool into
1057 * READ and WRITE
1058 */
1059rq_starved:
1060 if (unlikely(rl->count[is_sync] == 0))
1061 rl->starved[is_sync] = 1;
1062 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}
1064
Tejun Heoda8303c2011-10-19 14:33:05 +02001065/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001066 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001067 * @q: request_queue to allocate request from
1068 * @rw_flags: RW and SYNC flags
1069 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001070 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001071 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001072 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1073 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001074 *
1075 * Must be callled with @q->queue_lock held and,
1076 * Returns %NULL on failure, with @q->queue_lock held.
1077 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001079static struct request *get_request(struct request_queue *q, int rw_flags,
1080 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001082 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001083 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001084 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001086
1087 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001088retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001089 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001090 if (rq)
1091 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001093 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001094 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001095 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
Tejun Heoa06e05e2012-06-04 20:40:55 -07001098 /* wait on @rl and retry */
1099 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1100 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001101
Tejun Heoa06e05e2012-06-04 20:40:55 -07001102 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Tejun Heoa06e05e2012-06-04 20:40:55 -07001104 spin_unlock_irq(q->queue_lock);
1105 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
Tejun Heoa06e05e2012-06-04 20:40:55 -07001107 /*
1108 * After sleeping, we become a "batching" process and will be able
1109 * to allocate at least one request, and up to a big batch of them
1110 * for a small period time. See ioc_batching, ioc_set_batching
1111 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001112 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Tejun Heoa06e05e2012-06-04 20:40:55 -07001114 spin_lock_irq(q->queue_lock);
1115 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001116
Tejun Heoa06e05e2012-06-04 20:40:55 -07001117 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
Jens Axboe165125e2007-07-24 09:28:11 +02001120struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
1122 struct request *rq;
1123
1124 BUG_ON(rw != READ && rw != WRITE);
1125
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001126 /* create ioc upfront */
1127 create_io_context(gfp_mask, q->node);
1128
Nick Piggind6344532005-06-28 20:45:14 -07001129 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001130 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001131 if (!rq)
1132 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001133 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135 return rq;
1136}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137EXPORT_SYMBOL(blk_get_request);
1138
1139/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001140 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001141 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001142 * @bio: The bio describing the memory mappings that will be submitted for IO.
1143 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001144 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001145 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001146 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1147 * type commands. Where the struct request needs to be farther initialized by
1148 * the caller. It is passed a &struct bio, which describes the memory info of
1149 * the I/O transfer.
1150 *
1151 * The caller of blk_make_request must make sure that bi_io_vec
1152 * are set to describe the memory buffers. That bio_data_dir() will return
1153 * the needed direction of the request. (And all bio's in the passed bio-chain
1154 * are properly set accordingly)
1155 *
1156 * If called under none-sleepable conditions, mapped bio buffers must not
1157 * need bouncing, by calling the appropriate masked or flagged allocator,
1158 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1159 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001160 *
1161 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1162 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1163 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1164 * completion of a bio that hasn't been submitted yet, thus resulting in a
1165 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1166 * of bio_alloc(), as that avoids the mempool deadlock.
1167 * If possible a big IO should be split into smaller parts when allocation
1168 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001169 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001170struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1171 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001172{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001173 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1174
1175 if (unlikely(!rq))
1176 return ERR_PTR(-ENOMEM);
1177
1178 for_each_bio(bio) {
1179 struct bio *bounce_bio = bio;
1180 int ret;
1181
1182 blk_queue_bounce(q, &bounce_bio);
1183 ret = blk_rq_append_bio(q, rq, bounce_bio);
1184 if (unlikely(ret)) {
1185 blk_put_request(rq);
1186 return ERR_PTR(ret);
1187 }
1188 }
1189
1190 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001191}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001192EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001193
1194/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * blk_requeue_request - put a request back on queue
1196 * @q: request queue where request should be inserted
1197 * @rq: request to be inserted
1198 *
1199 * Description:
1200 * Drivers often keep queueing requests until the hardware cannot accept
1201 * more, when that condition happens we need to put the request back
1202 * on the queue. Must be called with queue lock held.
1203 */
Jens Axboe165125e2007-07-24 09:28:11 +02001204void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001206 blk_delete_timer(rq);
1207 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001208 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (blk_rq_tagged(rq))
1211 blk_queue_end_tag(q, rq);
1212
James Bottomleyba396a62009-05-27 14:17:08 +02001213 BUG_ON(blk_queued_rq(rq));
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 elv_requeue_request(q, rq);
1216}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217EXPORT_SYMBOL(blk_requeue_request);
1218
Jens Axboe73c10102011-03-08 13:19:51 +01001219static void add_acct_request(struct request_queue *q, struct request *rq,
1220 int where)
1221{
1222 drive_stat_acct(rq, 1);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001223 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001224}
1225
Tejun Heo074a7ac2008-08-25 19:56:14 +09001226static void part_round_stats_single(int cpu, struct hd_struct *part,
1227 unsigned long now)
1228{
1229 if (now == part->stamp)
1230 return;
1231
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001232 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001233 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001234 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001235 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1236 }
1237 part->stamp = now;
1238}
1239
1240/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001241 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1242 * @cpu: cpu number for stats access
1243 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 *
1245 * The average IO queue length and utilisation statistics are maintained
1246 * by observing the current state of the queue length and the amount of
1247 * time it has been in this state for.
1248 *
1249 * Normally, that accounting is done on IO completion, but that can result
1250 * in more than a second's worth of IO being accounted for within any one
1251 * second, leading to >100% utilisation. To deal with that, we call this
1252 * function to do a round-off before returning the results when reading
1253 * /proc/diskstats. This accounts immediately for all queue usage up to
1254 * the current jiffies and restarts the counters again.
1255 */
Tejun Heoc9959052008-08-25 19:47:21 +09001256void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001257{
1258 unsigned long now = jiffies;
1259
Tejun Heo074a7ac2008-08-25 19:56:14 +09001260 if (part->partno)
1261 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1262 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001263}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001264EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266/*
1267 * queue lock must be held
1268 */
Jens Axboe165125e2007-07-24 09:28:11 +02001269void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (unlikely(!q))
1272 return;
1273 if (unlikely(--req->ref_count))
1274 return;
1275
Tejun Heo8922e162005-10-20 16:23:44 +02001276 elv_completed_request(q, req);
1277
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001278 /* this is a bio leak */
1279 WARN_ON(req->bio != NULL);
1280
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 /*
1282 * Request may not have originated from ll_rw_blk. if not,
1283 * it didn't come out of our reserved rq pools
1284 */
Jens Axboe49171e52006-08-10 08:59:11 +02001285 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001286 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001287 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001290 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Tejun Heoa0516612012-06-26 15:05:44 -07001292 blk_free_request(rl, req);
1293 freed_request(rl, flags);
1294 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
1296}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001297EXPORT_SYMBOL_GPL(__blk_put_request);
1298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299void blk_put_request(struct request *req)
1300{
Tejun Heo8922e162005-10-20 16:23:44 +02001301 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02001302 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
FUJITA Tomonori52a93ba2008-07-15 21:21:45 +02001304 spin_lock_irqsave(q->queue_lock, flags);
1305 __blk_put_request(q, req);
1306 spin_unlock_irqrestore(q->queue_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308EXPORT_SYMBOL(blk_put_request);
1309
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001310/**
1311 * blk_add_request_payload - add a payload to a request
1312 * @rq: request to update
1313 * @page: page backing the payload
1314 * @len: length of the payload.
1315 *
1316 * This allows to later add a payload to an already submitted request by
1317 * a block driver. The driver needs to take care of freeing the payload
1318 * itself.
1319 *
1320 * Note that this is a quite horrible hack and nothing but handling of
1321 * discard requests should ever use it.
1322 */
1323void blk_add_request_payload(struct request *rq, struct page *page,
1324 unsigned int len)
1325{
1326 struct bio *bio = rq->bio;
1327
1328 bio->bi_io_vec->bv_page = page;
1329 bio->bi_io_vec->bv_offset = 0;
1330 bio->bi_io_vec->bv_len = len;
1331
1332 bio->bi_size = len;
1333 bio->bi_vcnt = 1;
1334 bio->bi_phys_segments = 1;
1335
1336 rq->__data_len = rq->resid_len = len;
1337 rq->nr_phys_segments = 1;
1338 rq->buffer = bio_data(bio);
1339}
1340EXPORT_SYMBOL_GPL(blk_add_request_payload);
1341
Jens Axboe73c10102011-03-08 13:19:51 +01001342static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1343 struct bio *bio)
1344{
1345 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1346
Jens Axboe73c10102011-03-08 13:19:51 +01001347 if (!ll_back_merge_fn(q, req, bio))
1348 return false;
1349
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001350 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001351
1352 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1353 blk_rq_set_mixed_merge(req);
1354
1355 req->biotail->bi_next = bio;
1356 req->biotail = bio;
1357 req->__data_len += bio->bi_size;
1358 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1359
1360 drive_stat_acct(req, 0);
1361 return true;
1362}
1363
1364static bool bio_attempt_front_merge(struct request_queue *q,
1365 struct request *req, struct bio *bio)
1366{
1367 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001368
Jens Axboe73c10102011-03-08 13:19:51 +01001369 if (!ll_front_merge_fn(q, req, bio))
1370 return false;
1371
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001372 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001373
1374 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1375 blk_rq_set_mixed_merge(req);
1376
Jens Axboe73c10102011-03-08 13:19:51 +01001377 bio->bi_next = req->bio;
1378 req->bio = bio;
1379
1380 /*
1381 * may not be valid. if the low level driver said
1382 * it didn't need a bounce buffer then it better
1383 * not touch req->buffer either...
1384 */
1385 req->buffer = bio_data(bio);
1386 req->__sector = bio->bi_sector;
1387 req->__data_len += bio->bi_size;
1388 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1389
1390 drive_stat_acct(req, 0);
1391 return true;
1392}
1393
Tejun Heobd87b582011-10-19 14:33:08 +02001394/**
1395 * attempt_plug_merge - try to merge with %current's plugged list
1396 * @q: request_queue new bio is being queued at
1397 * @bio: new bio being queued
1398 * @request_count: out parameter for number of traversed plugged requests
1399 *
1400 * Determine whether @bio being queued on @q can be merged with a request
1401 * on %current's plugged list. Returns %true if merge was successful,
1402 * otherwise %false.
1403 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001404 * Plugging coalesces IOs from the same issuer for the same purpose without
1405 * going through @q->queue_lock. As such it's more of an issuing mechanism
1406 * than scheduling, and the request, while may have elvpriv data, is not
1407 * added on the elevator at this point. In addition, we don't have
1408 * reliable access to the elevator outside queue lock. Only check basic
1409 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001410 */
Tejun Heobd87b582011-10-19 14:33:08 +02001411static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1412 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001413{
1414 struct blk_plug *plug;
1415 struct request *rq;
1416 bool ret = false;
1417
Tejun Heobd87b582011-10-19 14:33:08 +02001418 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001419 if (!plug)
1420 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001421 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001422
1423 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1424 int el_ret;
1425
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001426 if (rq->q == q)
1427 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001428
Tejun Heo07c2bd32012-02-08 09:19:42 +01001429 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001430 continue;
1431
Tejun Heo050c8ea2012-02-08 09:19:38 +01001432 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001433 if (el_ret == ELEVATOR_BACK_MERGE) {
1434 ret = bio_attempt_back_merge(q, rq, bio);
1435 if (ret)
1436 break;
1437 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1438 ret = bio_attempt_front_merge(q, rq, bio);
1439 if (ret)
1440 break;
1441 }
1442 }
1443out:
1444 return ret;
1445}
1446
Jens Axboe86db1e22008-01-29 14:53:40 +01001447void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001448{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001449 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001450
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001451 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1452 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001453 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001454
Tejun Heo52d9e672006-01-06 09:49:58 +01001455 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001456 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001457 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001458 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001459}
1460
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001461void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001463 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001464 struct blk_plug *plug;
1465 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1466 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001467 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 /*
1470 * low level driver can indicate that it wants pages above a
1471 * certain limit bounced to low memory (ie for highmem, or even
1472 * ISA dma in theory)
1473 */
1474 blk_queue_bounce(q, &bio);
1475
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001476 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1477 bio_endio(bio, -EIO);
1478 return;
1479 }
1480
Tejun Heo4fed9472010-09-03 11:56:17 +02001481 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001482 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001483 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001484 goto get_rq;
1485 }
1486
Jens Axboe73c10102011-03-08 13:19:51 +01001487 /*
1488 * Check if we can merge with the plugged list before grabbing
1489 * any locks.
1490 */
Tejun Heobd87b582011-10-19 14:33:08 +02001491 if (attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001492 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001493
1494 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
1496 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001497 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001498 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001499 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001500 if (!attempt_back_merge(q, req))
1501 elv_merged_request(q, req, el_ret);
1502 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001503 }
Jens Axboe73c10102011-03-08 13:19:51 +01001504 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001505 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001506 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001507 if (!attempt_front_merge(q, req))
1508 elv_merged_request(q, req, el_ret);
1509 goto out_unlock;
1510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 }
1512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001514 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001515 * This sync check and mask will be re-done in init_request_from_bio(),
1516 * but we need to set it earlier to expose the sync flag to the
1517 * rq allocator and io schedulers.
1518 */
1519 rw_flags = bio_data_dir(bio);
1520 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001521 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001522
1523 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001524 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001525 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001526 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001527 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001528 if (unlikely(!req)) {
1529 bio_endio(bio, -ENODEV); /* @q is dead */
1530 goto out_unlock;
1531 }
Nick Piggind6344532005-06-28 20:45:14 -07001532
Nick Piggin450991b2005-06-28 20:45:13 -07001533 /*
1534 * After dropping the lock and possibly sleeping here, our request
1535 * may now be mergeable after it had proven unmergeable (above).
1536 * We don't worry about that case for efficiency. It won't happen
1537 * often, and the elevators are able to handle it.
1538 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001539 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Tao Ma9562ad92011-10-24 16:11:30 +02001541 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001542 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001543
Jens Axboe73c10102011-03-08 13:19:51 +01001544 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001545 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001546 /*
1547 * If this is the first request added after a plug, fire
1548 * of a plug trace. If others have been added before, check
1549 * if we have multiple devices in this plug. If so, make a
1550 * note to sort the list before dispatch.
1551 */
1552 if (list_empty(&plug->list))
1553 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001554 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001555 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001556 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001557 trace_block_plug(q);
1558 }
Jens Axboe73c10102011-03-08 13:19:51 +01001559 }
Shaohua Lia6327162011-08-24 16:04:32 +02001560 list_add_tail(&req->queuelist, &plug->list);
1561 drive_stat_acct(req, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001562 } else {
1563 spin_lock_irq(q->queue_lock);
1564 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001565 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001566out_unlock:
1567 spin_unlock_irq(q->queue_lock);
1568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
Jens Axboec20e8de2011-09-12 12:03:37 +02001570EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572/*
1573 * If bio->bi_dev is a partition, remap the location
1574 */
1575static inline void blk_partition_remap(struct bio *bio)
1576{
1577 struct block_device *bdev = bio->bi_bdev;
1578
Jens Axboebf2de6f2007-09-27 13:01:25 +02001579 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 bio->bi_sector += p->start_sect;
1583 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001584
Mike Snitzerd07335e2010-11-16 12:52:38 +01001585 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1586 bdev->bd_dev,
1587 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 }
1589}
1590
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591static void handle_bad_sector(struct bio *bio)
1592{
1593 char b[BDEVNAME_SIZE];
1594
1595 printk(KERN_INFO "attempt to access beyond end of device\n");
1596 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1597 bdevname(bio->bi_bdev, b),
1598 bio->bi_rw,
1599 (unsigned long long)bio->bi_sector + bio_sectors(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001600 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 set_bit(BIO_EOF, &bio->bi_flags);
1603}
1604
Akinobu Mitac17bb492006-12-08 02:39:46 -08001605#ifdef CONFIG_FAIL_MAKE_REQUEST
1606
1607static DECLARE_FAULT_ATTR(fail_make_request);
1608
1609static int __init setup_fail_make_request(char *str)
1610{
1611 return setup_fault_attr(&fail_make_request, str);
1612}
1613__setup("fail_make_request=", setup_fail_make_request);
1614
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001615static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001616{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001617 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001618}
1619
1620static int __init fail_make_request_debugfs(void)
1621{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001622 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1623 NULL, &fail_make_request);
1624
1625 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001626}
1627
1628late_initcall(fail_make_request_debugfs);
1629
1630#else /* CONFIG_FAIL_MAKE_REQUEST */
1631
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001632static inline bool should_fail_request(struct hd_struct *part,
1633 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001634{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001635 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001636}
1637
1638#endif /* CONFIG_FAIL_MAKE_REQUEST */
1639
Jens Axboec07e2b42007-07-18 13:27:58 +02001640/*
1641 * Check whether this bio extends beyond the end of the device.
1642 */
1643static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1644{
1645 sector_t maxsector;
1646
1647 if (!nr_sectors)
1648 return 0;
1649
1650 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001651 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001652 if (maxsector) {
1653 sector_t sector = bio->bi_sector;
1654
1655 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1656 /*
1657 * This may well happen - the kernel calls bread()
1658 * without checking the size of the device, e.g., when
1659 * mounting a device.
1660 */
1661 handle_bad_sector(bio);
1662 return 1;
1663 }
1664 }
1665
1666 return 0;
1667}
1668
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001669static noinline_for_stack bool
1670generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Jens Axboe165125e2007-07-24 09:28:11 +02001672 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001673 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001674 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001675 char b[BDEVNAME_SIZE];
1676 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
1678 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Jens Axboec07e2b42007-07-18 13:27:58 +02001680 if (bio_check_eod(bio, nr_sectors))
1681 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001683 q = bdev_get_queue(bio->bi_bdev);
1684 if (unlikely(!q)) {
1685 printk(KERN_ERR
1686 "generic_make_request: Trying to access "
1687 "nonexistent block-device %s (%Lu)\n",
1688 bdevname(bio->bi_bdev, b),
1689 (long long) bio->bi_sector);
1690 goto end_io;
1691 }
1692
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001693 if (likely(bio_is_rw(bio) &&
1694 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001695 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1696 bdevname(bio->bi_bdev, b),
1697 bio_sectors(bio),
1698 queue_max_hw_sectors(q));
1699 goto end_io;
1700 }
1701
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001702 part = bio->bi_bdev->bd_part;
1703 if (should_fail_request(part, bio->bi_size) ||
1704 should_fail_request(&part_to_disk(part)->part0,
1705 bio->bi_size))
1706 goto end_io;
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001709 * If this device has partitions, remap block n
1710 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001712 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001714 if (bio_check_eod(bio, nr_sectors))
1715 goto end_io;
1716
1717 /*
1718 * Filter flush bio's early so that make_request based
1719 * drivers without flush support don't have to worry
1720 * about them.
1721 */
1722 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1723 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1724 if (!nr_sectors) {
1725 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001726 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001730 if ((bio->bi_rw & REQ_DISCARD) &&
1731 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001732 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001733 err = -EOPNOTSUPP;
1734 goto end_io;
1735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001737 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 err = -EOPNOTSUPP;
1739 goto end_io;
1740 }
1741
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001742 /*
1743 * Various block parts want %current->io_context and lazy ioc
1744 * allocation ends up trading a lot of pain for a small amount of
1745 * memory. Just allocate it upfront. This may fail and block
1746 * layer knows how to live with it.
1747 */
1748 create_io_context(GFP_ATOMIC, q->node);
1749
Tejun Heobc16a4f2011-10-19 14:33:01 +02001750 if (blk_throtl_bio(q, bio))
1751 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001752
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001753 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001754 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001755
1756end_io:
1757 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001758 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759}
1760
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001761/**
1762 * generic_make_request - hand a buffer to its device driver for I/O
1763 * @bio: The bio describing the location in memory and on the device.
1764 *
1765 * generic_make_request() is used to make I/O requests of block
1766 * devices. It is passed a &struct bio, which describes the I/O that needs
1767 * to be done.
1768 *
1769 * generic_make_request() does not return any status. The
1770 * success/failure status of the request, along with notification of
1771 * completion, is delivered asynchronously through the bio->bi_end_io
1772 * function described (one day) else where.
1773 *
1774 * The caller of generic_make_request must make sure that bi_io_vec
1775 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1776 * set to describe the device address, and the
1777 * bi_end_io and optionally bi_private are set to describe how
1778 * completion notification should be signaled.
1779 *
1780 * generic_make_request and the drivers it calls may use bi_next if this
1781 * bio happens to be merged with someone else, and may resubmit the bio to
1782 * a lower device by calling into generic_make_request recursively, which
1783 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001784 */
1785void generic_make_request(struct bio *bio)
1786{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001787 struct bio_list bio_list_on_stack;
1788
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001789 if (!generic_make_request_checks(bio))
1790 return;
1791
1792 /*
1793 * We only want one ->make_request_fn to be active at a time, else
1794 * stack usage with stacked devices could be a problem. So use
1795 * current->bio_list to keep a list of requests submited by a
1796 * make_request_fn function. current->bio_list is also used as a
1797 * flag to say if generic_make_request is currently active in this
1798 * task or not. If it is NULL, then no make_request is active. If
1799 * it is non-NULL, then a make_request is active, and new requests
1800 * should be added at the tail
1801 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001802 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001803 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001804 return;
1805 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001806
Neil Brownd89d8792007-05-01 09:53:42 +02001807 /* following loop may be a bit non-obvious, and so deserves some
1808 * explanation.
1809 * Before entering the loop, bio->bi_next is NULL (as all callers
1810 * ensure that) so we have a list with a single bio.
1811 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001812 * we assign bio_list to a pointer to the bio_list_on_stack,
1813 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001814 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001815 * through a recursive call to generic_make_request. If it
1816 * did, we find a non-NULL value in bio_list and re-enter the loop
1817 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001818 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001819 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001820 */
1821 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001822 bio_list_init(&bio_list_on_stack);
1823 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001824 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001825 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1826
1827 q->make_request_fn(q, bio);
1828
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001829 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001830 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001831 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001832}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833EXPORT_SYMBOL(generic_make_request);
1834
1835/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001836 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1838 * @bio: The &struct bio which describes the I/O
1839 *
1840 * submit_bio() is very similar in purpose to generic_make_request(), and
1841 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001842 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 *
1844 */
1845void submit_bio(int rw, struct bio *bio)
1846{
Jens Axboe22e2c502005-06-27 10:55:12 +02001847 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
Jens Axboebf2de6f2007-09-27 13:01:25 +02001849 /*
1850 * If it's a regular read/write or a barrier with data attached,
1851 * go through the normal accounting stuff before submission.
1852 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001853 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001854 unsigned int count;
1855
1856 if (unlikely(rw & REQ_WRITE_SAME))
1857 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1858 else
1859 count = bio_sectors(bio);
1860
Jens Axboebf2de6f2007-09-27 13:01:25 +02001861 if (rw & WRITE) {
1862 count_vm_events(PGPGOUT, count);
1863 } else {
1864 task_io_account_read(bio->bi_size);
1865 count_vm_events(PGPGIN, count);
1866 }
1867
1868 if (unlikely(block_dump)) {
1869 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001870 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001871 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001872 (rw & WRITE) ? "WRITE" : "READ",
1873 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001874 bdevname(bio->bi_bdev, b),
1875 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 }
1878
1879 generic_make_request(bio);
1880}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881EXPORT_SYMBOL(submit_bio);
1882
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001883/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001884 * blk_rq_check_limits - Helper function to check a request for the queue limit
1885 * @q: the queue
1886 * @rq: the request being checked
1887 *
1888 * Description:
1889 * @rq may have been made based on weaker limitations of upper-level queues
1890 * in request stacking drivers, and it may violate the limitation of @q.
1891 * Since the block layer and the underlying device driver trust @rq
1892 * after it is inserted to @q, it should be checked against @q before
1893 * the insertion using this generic function.
1894 *
1895 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001896 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001897 * Request stacking drivers like request-based dm may change the queue
1898 * limits while requests are in the queue (e.g. dm's table swapping).
1899 * Such request stacking drivers should check those requests agaist
1900 * the new queue limits again when they dispatch those requests,
1901 * although such checkings are also done against the old queue limits
1902 * when submitting requests.
1903 */
1904int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1905{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001906 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001907 return 0;
1908
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001909 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001910 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1911 return -EIO;
1912 }
1913
1914 /*
1915 * queue's settings related to segment counting like q->bounce_pfn
1916 * may differ from that of other stacking queues.
1917 * Recalculate it to check the request correctly on this queue's
1918 * limitation.
1919 */
1920 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001921 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001922 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1923 return -EIO;
1924 }
1925
1926 return 0;
1927}
1928EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1929
1930/**
1931 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1932 * @q: the queue to submit the request
1933 * @rq: the request being queued
1934 */
1935int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1936{
1937 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001938 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001939
1940 if (blk_rq_check_limits(q, rq))
1941 return -EIO;
1942
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001943 if (rq->rq_disk &&
1944 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001945 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001946
1947 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001948 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01001949 spin_unlock_irqrestore(q->queue_lock, flags);
1950 return -ENODEV;
1951 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001952
1953 /*
1954 * Submitting request must be dequeued before calling this function
1955 * because it will be linked to another request_queue
1956 */
1957 BUG_ON(blk_queued_rq(rq));
1958
Jeff Moyer4853aba2011-08-15 21:37:25 +02001959 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1960 where = ELEVATOR_INSERT_FLUSH;
1961
1962 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001963 if (where == ELEVATOR_INSERT_FLUSH)
1964 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001965 spin_unlock_irqrestore(q->queue_lock, flags);
1966
1967 return 0;
1968}
1969EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1970
Tejun Heo80a761f2009-07-03 17:48:17 +09001971/**
1972 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1973 * @rq: request to examine
1974 *
1975 * Description:
1976 * A request could be merge of IOs which require different failure
1977 * handling. This function determines the number of bytes which
1978 * can be failed from the beginning of the request without
1979 * crossing into area which need to be retried further.
1980 *
1981 * Return:
1982 * The number of bytes to fail.
1983 *
1984 * Context:
1985 * queue_lock must be held.
1986 */
1987unsigned int blk_rq_err_bytes(const struct request *rq)
1988{
1989 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1990 unsigned int bytes = 0;
1991 struct bio *bio;
1992
1993 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1994 return blk_rq_bytes(rq);
1995
1996 /*
1997 * Currently the only 'mixing' which can happen is between
1998 * different fastfail types. We can safely fail portions
1999 * which have all the failfast bits that the first one has -
2000 * the ones which are at least as eager to fail as the first
2001 * one.
2002 */
2003 for (bio = rq->bio; bio; bio = bio->bi_next) {
2004 if ((bio->bi_rw & ff) != ff)
2005 break;
2006 bytes += bio->bi_size;
2007 }
2008
2009 /* this could lead to infinite loop */
2010 BUG_ON(blk_rq_bytes(rq) && !bytes);
2011 return bytes;
2012}
2013EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2014
Jens Axboebc58ba92009-01-23 10:54:44 +01002015static void blk_account_io_completion(struct request *req, unsigned int bytes)
2016{
Jens Axboec2553b52009-04-24 08:10:11 +02002017 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002018 const int rw = rq_data_dir(req);
2019 struct hd_struct *part;
2020 int cpu;
2021
2022 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002023 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002024 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2025 part_stat_unlock();
2026 }
2027}
2028
2029static void blk_account_io_done(struct request *req)
2030{
Jens Axboebc58ba92009-01-23 10:54:44 +01002031 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002032 * Account IO completion. flush_rq isn't accounted as a
2033 * normal IO on queueing nor completion. Accounting the
2034 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002035 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002036 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002037 unsigned long duration = jiffies - req->start_time;
2038 const int rw = rq_data_dir(req);
2039 struct hd_struct *part;
2040 int cpu;
2041
2042 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002043 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002044
2045 part_stat_inc(cpu, part, ios[rw]);
2046 part_stat_add(cpu, part, ticks[rw], duration);
2047 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002048 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002049
Jens Axboe6c23a962011-01-07 08:43:37 +01002050 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002051 part_stat_unlock();
2052 }
2053}
2054
Tejun Heo53a08802008-12-03 12:41:26 +01002055/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002056 * blk_peek_request - peek at the top of a request queue
2057 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002058 *
2059 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002060 * Return the request at the top of @q. The returned request
2061 * should be started using blk_start_request() before LLD starts
2062 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002063 *
2064 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002065 * Pointer to the request at the top of @q if available. Null
2066 * otherwise.
2067 *
2068 * Context:
2069 * queue_lock must be held.
2070 */
2071struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002072{
2073 struct request *rq;
2074 int ret;
2075
2076 while ((rq = __elv_next_request(q)) != NULL) {
2077 if (!(rq->cmd_flags & REQ_STARTED)) {
2078 /*
2079 * This is the first time the device driver
2080 * sees this request (possibly after
2081 * requeueing). Notify IO scheduler.
2082 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002083 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002084 elv_activate_rq(q, rq);
2085
2086 /*
2087 * just mark as started even if we don't start
2088 * it, a request that has been delayed should
2089 * not be passed by new incoming requests
2090 */
2091 rq->cmd_flags |= REQ_STARTED;
2092 trace_block_rq_issue(q, rq);
2093 }
2094
2095 if (!q->boundary_rq || q->boundary_rq == rq) {
2096 q->end_sector = rq_end_sector(rq);
2097 q->boundary_rq = NULL;
2098 }
2099
2100 if (rq->cmd_flags & REQ_DONTPREP)
2101 break;
2102
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002103 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002104 /*
2105 * make sure space for the drain appears we
2106 * know we can do this because max_hw_segments
2107 * has been adjusted to be one fewer than the
2108 * device can handle
2109 */
2110 rq->nr_phys_segments++;
2111 }
2112
2113 if (!q->prep_rq_fn)
2114 break;
2115
2116 ret = q->prep_rq_fn(q, rq);
2117 if (ret == BLKPREP_OK) {
2118 break;
2119 } else if (ret == BLKPREP_DEFER) {
2120 /*
2121 * the request may have been (partially) prepped.
2122 * we need to keep this request in the front to
2123 * avoid resource deadlock. REQ_STARTED will
2124 * prevent other fs requests from passing this one.
2125 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002126 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002127 !(rq->cmd_flags & REQ_DONTPREP)) {
2128 /*
2129 * remove the space for the drain we added
2130 * so that we don't add it again
2131 */
2132 --rq->nr_phys_segments;
2133 }
2134
2135 rq = NULL;
2136 break;
2137 } else if (ret == BLKPREP_KILL) {
2138 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002139 /*
2140 * Mark this request as started so we don't trigger
2141 * any debug logic in the end I/O path.
2142 */
2143 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002144 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002145 } else {
2146 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2147 break;
2148 }
2149 }
2150
2151 return rq;
2152}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002153EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002154
Tejun Heo9934c8c2009-05-08 11:54:16 +09002155void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002156{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002157 struct request_queue *q = rq->q;
2158
Tejun Heo158dbda2009-04-23 11:05:18 +09002159 BUG_ON(list_empty(&rq->queuelist));
2160 BUG_ON(ELV_ON_HASH(rq));
2161
2162 list_del_init(&rq->queuelist);
2163
2164 /*
2165 * the time frame between a request being removed from the lists
2166 * and to it is freed is accounted as io that is in progress at
2167 * the driver side.
2168 */
Divyesh Shah91952912010-04-01 15:01:41 -07002169 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002170 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002171 set_io_start_time_ns(rq);
2172 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002173}
2174
Tejun Heo5efccd12009-04-23 11:05:18 +09002175/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002176 * blk_start_request - start request processing on the driver
2177 * @req: request to dequeue
2178 *
2179 * Description:
2180 * Dequeue @req and start timeout timer on it. This hands off the
2181 * request to the driver.
2182 *
2183 * Block internal functions which don't want to start timer should
2184 * call blk_dequeue_request().
2185 *
2186 * Context:
2187 * queue_lock must be held.
2188 */
2189void blk_start_request(struct request *req)
2190{
2191 blk_dequeue_request(req);
2192
2193 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002194 * We are now handing the request to the hardware, initialize
2195 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002196 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002197 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002198 if (unlikely(blk_bidi_rq(req)))
2199 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2200
Tejun Heo9934c8c2009-05-08 11:54:16 +09002201 blk_add_timer(req);
2202}
2203EXPORT_SYMBOL(blk_start_request);
2204
2205/**
2206 * blk_fetch_request - fetch a request from a request queue
2207 * @q: request queue to fetch a request from
2208 *
2209 * Description:
2210 * Return the request at the top of @q. The request is started on
2211 * return and LLD can start processing it immediately.
2212 *
2213 * Return:
2214 * Pointer to the request at the top of @q if available. Null
2215 * otherwise.
2216 *
2217 * Context:
2218 * queue_lock must be held.
2219 */
2220struct request *blk_fetch_request(struct request_queue *q)
2221{
2222 struct request *rq;
2223
2224 rq = blk_peek_request(q);
2225 if (rq)
2226 blk_start_request(rq);
2227 return rq;
2228}
2229EXPORT_SYMBOL(blk_fetch_request);
2230
2231/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002232 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002233 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002234 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002235 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002236 *
2237 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002238 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2239 * the request structure even if @req doesn't have leftover.
2240 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002241 *
2242 * This special helper function is only for request stacking drivers
2243 * (e.g. request-based dm) so that they can handle partial completion.
2244 * Actual device drivers should use blk_end_request instead.
2245 *
2246 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2247 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002248 *
2249 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002250 * %false - this request doesn't have any more data
2251 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002252 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002253bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254{
Kiyoshi Ueda5450d3e2007-12-11 17:53:03 -05002255 int total_bytes, bio_nbytes, next_idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 struct bio *bio;
2257
Tejun Heo2e60e022009-04-23 11:05:18 +09002258 if (!req->bio)
2259 return false;
2260
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002261 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002264 * For fs requests, rq is just carrier of independent bio's
2265 * and each partial completion should be handled separately.
2266 * Reset per-request error on each partial completion.
2267 *
2268 * TODO: tj: This is too subtle. It would be better to let
2269 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002271 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 req->errors = 0;
2273
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002274 if (error && req->cmd_type == REQ_TYPE_FS &&
2275 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002276 char *error_type;
2277
2278 switch (error) {
2279 case -ENOLINK:
2280 error_type = "recoverable transport";
2281 break;
2282 case -EREMOTEIO:
2283 error_type = "critical target";
2284 break;
2285 case -EBADE:
2286 error_type = "critical nexus";
2287 break;
2288 case -EIO:
2289 default:
2290 error_type = "I/O";
2291 break;
2292 }
Yi Zou37d7b342012-08-30 16:26:25 -07002293 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2294 error_type, req->rq_disk ?
2295 req->rq_disk->disk_name : "?",
2296 (unsigned long long)blk_rq_pos(req));
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 }
2299
Jens Axboebc58ba92009-01-23 10:54:44 +01002300 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 total_bytes = bio_nbytes = 0;
2303 while ((bio = req->bio) != NULL) {
2304 int nbytes;
2305
2306 if (nr_bytes >= bio->bi_size) {
2307 req->bio = bio->bi_next;
2308 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02002309 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 next_idx = 0;
2311 bio_nbytes = 0;
2312 } else {
2313 int idx = bio->bi_idx + next_idx;
2314
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002315 if (unlikely(idx >= bio->bi_vcnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 blk_dump_rq_flags(req, "__end_that");
Jens Axboe6728cb02008-01-31 13:03:55 +01002317 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
Kazuhisa Ichikawaaf498d72009-05-12 13:27:45 +02002318 __func__, idx, bio->bi_vcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 break;
2320 }
2321
2322 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2323 BIO_BUG_ON(nbytes > bio->bi_size);
2324
2325 /*
2326 * not a complete bvec done
2327 */
2328 if (unlikely(nbytes > nr_bytes)) {
2329 bio_nbytes += nr_bytes;
2330 total_bytes += nr_bytes;
2331 break;
2332 }
2333
2334 /*
2335 * advance to the next vector
2336 */
2337 next_idx++;
2338 bio_nbytes += nbytes;
2339 }
2340
2341 total_bytes += nbytes;
2342 nr_bytes -= nbytes;
2343
Jens Axboe6728cb02008-01-31 13:03:55 +01002344 bio = req->bio;
2345 if (bio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 /*
2347 * end more in this run, or just return 'not-done'
2348 */
2349 if (unlikely(nr_bytes <= 0))
2350 break;
2351 }
2352 }
2353
2354 /*
2355 * completely done
2356 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002357 if (!req->bio) {
2358 /*
2359 * Reset counters so that the request stacking driver
2360 * can find how many bytes remain in the request
2361 * later.
2362 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002363 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002364 return false;
2365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
2367 /*
2368 * if the request wasn't completed, update state
2369 */
2370 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02002371 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 bio->bi_idx += next_idx;
2373 bio_iovec(bio)->bv_offset += nr_bytes;
2374 bio_iovec(bio)->bv_len -= nr_bytes;
2375 }
2376
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002377 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002378 req->buffer = bio_data(req->bio);
2379
2380 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002381 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002382 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002383
Tejun Heo80a761f2009-07-03 17:48:17 +09002384 /* mixed attributes always follow the first bio */
2385 if (req->cmd_flags & REQ_MIXED_MERGE) {
2386 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2387 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2388 }
2389
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002390 /*
2391 * If total number of sectors is less than the first segment
2392 * size, something has gone terribly wrong.
2393 */
2394 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002395 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002396 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002397 }
2398
2399 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002401
Tejun Heo2e60e022009-04-23 11:05:18 +09002402 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403}
Tejun Heo2e60e022009-04-23 11:05:18 +09002404EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Tejun Heo2e60e022009-04-23 11:05:18 +09002406static bool blk_update_bidi_request(struct request *rq, int error,
2407 unsigned int nr_bytes,
2408 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002409{
Tejun Heo2e60e022009-04-23 11:05:18 +09002410 if (blk_update_request(rq, error, nr_bytes))
2411 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002412
Tejun Heo2e60e022009-04-23 11:05:18 +09002413 /* Bidi request must be completed as a whole */
2414 if (unlikely(blk_bidi_rq(rq)) &&
2415 blk_update_request(rq->next_rq, error, bidi_bytes))
2416 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002417
Jens Axboee2e1a142010-06-09 10:42:09 +02002418 if (blk_queue_add_random(rq->q))
2419 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002420
2421 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422}
2423
James Bottomley28018c22010-07-01 19:49:17 +09002424/**
2425 * blk_unprep_request - unprepare a request
2426 * @req: the request
2427 *
2428 * This function makes a request ready for complete resubmission (or
2429 * completion). It happens only after all error handling is complete,
2430 * so represents the appropriate moment to deallocate any resources
2431 * that were allocated to the request in the prep_rq_fn. The queue
2432 * lock is held when calling this.
2433 */
2434void blk_unprep_request(struct request *req)
2435{
2436 struct request_queue *q = req->q;
2437
2438 req->cmd_flags &= ~REQ_DONTPREP;
2439 if (q->unprep_rq_fn)
2440 q->unprep_rq_fn(q, req);
2441}
2442EXPORT_SYMBOL_GPL(blk_unprep_request);
2443
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444/*
2445 * queue lock must be held
2446 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002447static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002449 if (blk_rq_tagged(req))
2450 blk_queue_end_tag(req->q, req);
2451
James Bottomleyba396a62009-05-27 14:17:08 +02002452 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002454 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002455 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456
Mike Andersone78042e2008-10-30 02:16:20 -07002457 blk_delete_timer(req);
2458
James Bottomley28018c22010-07-01 19:49:17 +09002459 if (req->cmd_flags & REQ_DONTPREP)
2460 blk_unprep_request(req);
2461
2462
Jens Axboebc58ba92009-01-23 10:54:44 +01002463 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002464
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002466 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002467 else {
2468 if (blk_bidi_rq(req))
2469 __blk_put_request(req->next_rq->q, req->next_rq);
2470
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 }
2473}
2474
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002475/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002476 * blk_end_bidi_request - Complete a bidi request
2477 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002478 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002479 * @nr_bytes: number of bytes to complete @rq
2480 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002481 *
2482 * Description:
2483 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002484 * Drivers that supports bidi can safely call this member for any
2485 * type of request, bidi or uni. In the later case @bidi_bytes is
2486 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002487 *
2488 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002489 * %false - we are done with this request
2490 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002491 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002492static bool blk_end_bidi_request(struct request *rq, int error,
2493 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002494{
2495 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002496 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002497
Tejun Heo2e60e022009-04-23 11:05:18 +09002498 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2499 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002500
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002501 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002502 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002503 spin_unlock_irqrestore(q->queue_lock, flags);
2504
Tejun Heo2e60e022009-04-23 11:05:18 +09002505 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002506}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002507
2508/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002509 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2510 * @rq: the request to complete
2511 * @error: %0 for success, < %0 for error
2512 * @nr_bytes: number of bytes to complete @rq
2513 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002514 *
2515 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002516 * Identical to blk_end_bidi_request() except that queue lock is
2517 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002518 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002519 * Return:
2520 * %false - we are done with this request
2521 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002522 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002523bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002524 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002525{
Tejun Heo2e60e022009-04-23 11:05:18 +09002526 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2527 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002528
Tejun Heo2e60e022009-04-23 11:05:18 +09002529 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002530
Tejun Heo2e60e022009-04-23 11:05:18 +09002531 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002532}
2533
2534/**
2535 * blk_end_request - Helper function for drivers to complete the request.
2536 * @rq: the request being processed
2537 * @error: %0 for success, < %0 for error
2538 * @nr_bytes: number of bytes to complete
2539 *
2540 * Description:
2541 * Ends I/O on a number of bytes attached to @rq.
2542 * If @rq has leftover, sets it up for the next range of segments.
2543 *
2544 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002545 * %false - we are done with this request
2546 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002547 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002548bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002549{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002550 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002551}
Jens Axboe56ad1742009-07-28 22:11:24 +02002552EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002553
2554/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002555 * blk_end_request_all - Helper function for drives to finish the request.
2556 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002557 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002558 *
2559 * Description:
2560 * Completely finish @rq.
2561 */
2562void blk_end_request_all(struct request *rq, int error)
2563{
2564 bool pending;
2565 unsigned int bidi_bytes = 0;
2566
2567 if (unlikely(blk_bidi_rq(rq)))
2568 bidi_bytes = blk_rq_bytes(rq->next_rq);
2569
2570 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2571 BUG_ON(pending);
2572}
Jens Axboe56ad1742009-07-28 22:11:24 +02002573EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002574
2575/**
2576 * blk_end_request_cur - Helper function to finish the current request chunk.
2577 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002578 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002579 *
2580 * Description:
2581 * Complete the current consecutively mapped chunk from @rq.
2582 *
2583 * Return:
2584 * %false - we are done with this request
2585 * %true - still buffers pending for this request
2586 */
2587bool blk_end_request_cur(struct request *rq, int error)
2588{
2589 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2590}
Jens Axboe56ad1742009-07-28 22:11:24 +02002591EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002592
2593/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002594 * blk_end_request_err - Finish a request till the next failure boundary.
2595 * @rq: the request to finish till the next failure boundary for
2596 * @error: must be negative errno
2597 *
2598 * Description:
2599 * Complete @rq till the next failure boundary.
2600 *
2601 * Return:
2602 * %false - we are done with this request
2603 * %true - still buffers pending for this request
2604 */
2605bool blk_end_request_err(struct request *rq, int error)
2606{
2607 WARN_ON(error >= 0);
2608 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2609}
2610EXPORT_SYMBOL_GPL(blk_end_request_err);
2611
2612/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002613 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002614 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002615 * @error: %0 for success, < %0 for error
2616 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002617 *
2618 * Description:
2619 * Must be called with queue lock held unlike blk_end_request().
2620 *
2621 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002622 * %false - we are done with this request
2623 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002624 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002625bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002626{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002627 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002628}
Jens Axboe56ad1742009-07-28 22:11:24 +02002629EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002630
2631/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002632 * __blk_end_request_all - Helper function for drives to finish the request.
2633 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002634 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002635 *
2636 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002637 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002638 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002639void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002640{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002641 bool pending;
2642 unsigned int bidi_bytes = 0;
2643
2644 if (unlikely(blk_bidi_rq(rq)))
2645 bidi_bytes = blk_rq_bytes(rq->next_rq);
2646
2647 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2648 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002649}
Jens Axboe56ad1742009-07-28 22:11:24 +02002650EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002651
2652/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002653 * __blk_end_request_cur - Helper function to finish the current request chunk.
2654 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002655 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002656 *
2657 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002658 * Complete the current consecutively mapped chunk from @rq. Must
2659 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002660 *
2661 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002662 * %false - we are done with this request
2663 * %true - still buffers pending for this request
2664 */
2665bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002666{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002667 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002668}
Jens Axboe56ad1742009-07-28 22:11:24 +02002669EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002670
Tejun Heo80a761f2009-07-03 17:48:17 +09002671/**
2672 * __blk_end_request_err - Finish a request till the next failure boundary.
2673 * @rq: the request to finish till the next failure boundary for
2674 * @error: must be negative errno
2675 *
2676 * Description:
2677 * Complete @rq till the next failure boundary. Must be called
2678 * with queue lock held.
2679 *
2680 * Return:
2681 * %false - we are done with this request
2682 * %true - still buffers pending for this request
2683 */
2684bool __blk_end_request_err(struct request *rq, int error)
2685{
2686 WARN_ON(error >= 0);
2687 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2688}
2689EXPORT_SYMBOL_GPL(__blk_end_request_err);
2690
Jens Axboe86db1e22008-01-29 14:53:40 +01002691void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2692 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002694 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002695 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696
David Woodhousefb2dce82008-08-05 18:01:53 +01002697 if (bio_has_data(bio)) {
2698 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002699 rq->buffer = bio_data(bio);
2700 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002701 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
NeilBrown66846572007-08-16 13:31:28 +02002704 if (bio->bi_bdev)
2705 rq->rq_disk = bio->bi_bdev->bd_disk;
2706}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002708#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2709/**
2710 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2711 * @rq: the request to be flushed
2712 *
2713 * Description:
2714 * Flush all pages in @rq.
2715 */
2716void rq_flush_dcache_pages(struct request *rq)
2717{
2718 struct req_iterator iter;
2719 struct bio_vec *bvec;
2720
2721 rq_for_each_segment(bvec, rq, iter)
2722 flush_dcache_page(bvec->bv_page);
2723}
2724EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2725#endif
2726
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002727/**
2728 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2729 * @q : the queue of the device being checked
2730 *
2731 * Description:
2732 * Check if underlying low-level drivers of a device are busy.
2733 * If the drivers want to export their busy state, they must set own
2734 * exporting function using blk_queue_lld_busy() first.
2735 *
2736 * Basically, this function is used only by request stacking drivers
2737 * to stop dispatching requests to underlying devices when underlying
2738 * devices are busy. This behavior helps more I/O merging on the queue
2739 * of the request stacking driver and prevents I/O throughput regression
2740 * on burst I/O load.
2741 *
2742 * Return:
2743 * 0 - Not busy (The request stacking driver should dispatch request)
2744 * 1 - Busy (The request stacking driver should stop dispatching request)
2745 */
2746int blk_lld_busy(struct request_queue *q)
2747{
2748 if (q->lld_busy_fn)
2749 return q->lld_busy_fn(q);
2750
2751 return 0;
2752}
2753EXPORT_SYMBOL_GPL(blk_lld_busy);
2754
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002755/**
2756 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2757 * @rq: the clone request to be cleaned up
2758 *
2759 * Description:
2760 * Free all bios in @rq for a cloned request.
2761 */
2762void blk_rq_unprep_clone(struct request *rq)
2763{
2764 struct bio *bio;
2765
2766 while ((bio = rq->bio) != NULL) {
2767 rq->bio = bio->bi_next;
2768
2769 bio_put(bio);
2770 }
2771}
2772EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2773
2774/*
2775 * Copy attributes of the original request to the clone request.
2776 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2777 */
2778static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2779{
2780 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002781 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002782 dst->cmd_type = src->cmd_type;
2783 dst->__sector = blk_rq_pos(src);
2784 dst->__data_len = blk_rq_bytes(src);
2785 dst->nr_phys_segments = src->nr_phys_segments;
2786 dst->ioprio = src->ioprio;
2787 dst->extra_len = src->extra_len;
2788}
2789
2790/**
2791 * blk_rq_prep_clone - Helper function to setup clone request
2792 * @rq: the request to be setup
2793 * @rq_src: original request to be cloned
2794 * @bs: bio_set that bios for clone are allocated from
2795 * @gfp_mask: memory allocation mask for bio
2796 * @bio_ctr: setup function to be called for each clone bio.
2797 * Returns %0 for success, non %0 for failure.
2798 * @data: private data to be passed to @bio_ctr
2799 *
2800 * Description:
2801 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2802 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2803 * are not copied, and copying such parts is the caller's responsibility.
2804 * Also, pages which the original bios are pointing to are not copied
2805 * and the cloned bios just point same pages.
2806 * So cloned bios must be completed before original bios, which means
2807 * the caller must complete @rq before @rq_src.
2808 */
2809int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2810 struct bio_set *bs, gfp_t gfp_mask,
2811 int (*bio_ctr)(struct bio *, struct bio *, void *),
2812 void *data)
2813{
2814 struct bio *bio, *bio_src;
2815
2816 if (!bs)
2817 bs = fs_bio_set;
2818
2819 blk_rq_init(NULL, rq);
2820
2821 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002822 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002823 if (!bio)
2824 goto free_and_out;
2825
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002826 if (bio_ctr && bio_ctr(bio, bio_src, data))
2827 goto free_and_out;
2828
2829 if (rq->bio) {
2830 rq->biotail->bi_next = bio;
2831 rq->biotail = bio;
2832 } else
2833 rq->bio = rq->biotail = bio;
2834 }
2835
2836 __blk_rq_prep_clone(rq, rq_src);
2837
2838 return 0;
2839
2840free_and_out:
2841 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002842 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002843 blk_rq_unprep_clone(rq);
2844
2845 return -ENOMEM;
2846}
2847EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2848
Jens Axboe18887ad2008-07-28 13:08:45 +02002849int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850{
2851 return queue_work(kblockd_workqueue, work);
2852}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853EXPORT_SYMBOL(kblockd_schedule_work);
2854
Vivek Goyale43473b2010-09-15 17:06:35 -04002855int kblockd_schedule_delayed_work(struct request_queue *q,
2856 struct delayed_work *dwork, unsigned long delay)
2857{
2858 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2859}
2860EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2861
Jens Axboe73c10102011-03-08 13:19:51 +01002862#define PLUG_MAGIC 0x91827364
2863
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002864/**
2865 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2866 * @plug: The &struct blk_plug that needs to be initialized
2867 *
2868 * Description:
2869 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2870 * pending I/O should the task end up blocking between blk_start_plug() and
2871 * blk_finish_plug(). This is important from a performance perspective, but
2872 * also ensures that we don't deadlock. For instance, if the task is blocking
2873 * for a memory allocation, memory reclaim could end up wanting to free a
2874 * page belonging to that request that is currently residing in our private
2875 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2876 * this kind of deadlock.
2877 */
Jens Axboe73c10102011-03-08 13:19:51 +01002878void blk_start_plug(struct blk_plug *plug)
2879{
2880 struct task_struct *tsk = current;
2881
2882 plug->magic = PLUG_MAGIC;
2883 INIT_LIST_HEAD(&plug->list);
NeilBrown048c9372011-04-18 09:52:22 +02002884 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002885
2886 /*
2887 * If this is a nested plug, don't actually assign it. It will be
2888 * flushed on its own.
2889 */
2890 if (!tsk->plug) {
2891 /*
2892 * Store ordering should not be needed here, since a potential
2893 * preempt will imply a full memory barrier
2894 */
2895 tsk->plug = plug;
2896 }
2897}
2898EXPORT_SYMBOL(blk_start_plug);
2899
2900static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2901{
2902 struct request *rqa = container_of(a, struct request, queuelist);
2903 struct request *rqb = container_of(b, struct request, queuelist);
2904
Jianpeng Ma975927b2012-10-25 21:58:17 +02002905 return !(rqa->q < rqb->q ||
2906 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01002907}
2908
Jens Axboe49cac012011-04-16 13:51:05 +02002909/*
2910 * If 'from_schedule' is true, then postpone the dispatch of requests
2911 * until a safe kblockd context. We due this to avoid accidental big
2912 * additional stack usage in driver dispatch, in places where the originally
2913 * plugger did not intend it.
2914 */
Jens Axboef6603782011-04-15 15:49:07 +02002915static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002916 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002917 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002918{
Jens Axboe49cac012011-04-16 13:51:05 +02002919 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002920
Bart Van Assche70460572012-11-28 13:45:56 +01002921 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002922 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002923 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002924 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002925 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002926}
2927
NeilBrown74018dc2012-07-31 09:08:15 +02002928static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002929{
2930 LIST_HEAD(callbacks);
2931
Shaohua Li2a7d5552012-07-31 09:08:15 +02002932 while (!list_empty(&plug->cb_list)) {
2933 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002934
Shaohua Li2a7d5552012-07-31 09:08:15 +02002935 while (!list_empty(&callbacks)) {
2936 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002937 struct blk_plug_cb,
2938 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002939 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02002940 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002941 }
NeilBrown048c9372011-04-18 09:52:22 +02002942 }
2943}
2944
NeilBrown9cbb1752012-07-31 09:08:14 +02002945struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2946 int size)
2947{
2948 struct blk_plug *plug = current->plug;
2949 struct blk_plug_cb *cb;
2950
2951 if (!plug)
2952 return NULL;
2953
2954 list_for_each_entry(cb, &plug->cb_list, list)
2955 if (cb->callback == unplug && cb->data == data)
2956 return cb;
2957
2958 /* Not currently on the callback list */
2959 BUG_ON(size < sizeof(*cb));
2960 cb = kzalloc(size, GFP_ATOMIC);
2961 if (cb) {
2962 cb->data = data;
2963 cb->callback = unplug;
2964 list_add(&cb->list, &plug->cb_list);
2965 }
2966 return cb;
2967}
2968EXPORT_SYMBOL(blk_check_plugged);
2969
Jens Axboe49cac012011-04-16 13:51:05 +02002970void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01002971{
2972 struct request_queue *q;
2973 unsigned long flags;
2974 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02002975 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002976 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01002977
2978 BUG_ON(plug->magic != PLUG_MAGIC);
2979
NeilBrown74018dc2012-07-31 09:08:15 +02002980 flush_plug_callbacks(plug, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01002981 if (list_empty(&plug->list))
2982 return;
2983
NeilBrown109b8122011-04-11 14:13:10 +02002984 list_splice_init(&plug->list, &list);
2985
Jianpeng Ma422765c2013-01-11 14:46:09 +01002986 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01002987
2988 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02002989 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02002990
2991 /*
2992 * Save and disable interrupts here, to avoid doing it for every
2993 * queue lock we have to take.
2994 */
Jens Axboe73c10102011-03-08 13:19:51 +01002995 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02002996 while (!list_empty(&list)) {
2997 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01002998 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01002999 BUG_ON(!rq->q);
3000 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003001 /*
3002 * This drops the queue lock
3003 */
3004 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003005 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003006 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003007 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003008 spin_lock(q->queue_lock);
3009 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003010
3011 /*
3012 * Short-circuit if @q is dead
3013 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003014 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003015 __blk_end_request_all(rq, -ENODEV);
3016 continue;
3017 }
3018
Jens Axboe73c10102011-03-08 13:19:51 +01003019 /*
3020 * rq is already accounted, so use raw insert
3021 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003022 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3023 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3024 else
3025 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003026
3027 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003028 }
3029
Jens Axboe99e22592011-04-18 09:59:55 +02003030 /*
3031 * This drops the queue lock
3032 */
3033 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003034 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003035
Jens Axboe73c10102011-03-08 13:19:51 +01003036 local_irq_restore(flags);
3037}
Jens Axboe73c10102011-03-08 13:19:51 +01003038
3039void blk_finish_plug(struct blk_plug *plug)
3040{
Jens Axboef6603782011-04-15 15:49:07 +02003041 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003042
3043 if (plug == current->plug)
3044 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003045}
3046EXPORT_SYMBOL(blk_finish_plug);
3047
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048int __init blk_dev_init(void)
3049{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003050 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3051 sizeof(((struct request *)0)->cmd_flags));
3052
Tejun Heo89b90be2011-01-03 15:01:47 +01003053 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3054 kblockd_workqueue = alloc_workqueue("kblockd",
3055 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 if (!kblockd_workqueue)
3057 panic("Failed to create kblockd\n");
3058
3059 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003060 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061
Jens Axboe8324aa92008-01-29 14:51:59 +01003062 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003063 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 return 0;
3066}