blob: 8bdd0121212a51a1dba3c568c5a6a3e070447318 [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>
Jens Axboe320ae512013-10-24 09:20:05 +010019#include <linux/blk-mq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/highmem.h>
21#include <linux/mm.h>
22#include <linux/kernel_stat.h>
23#include <linux/string.h>
24#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080029#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080030#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010031#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020032#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070033#include <linux/ratelimit.h>
Lin Ming6c954662013-03-23 11:42:26 +080034#include <linux/pm_runtime.h>
Li Zefan55782132009-06-09 13:43:05 +080035
36#define CREATE_TRACE_POINTS
37#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Jens Axboe8324aa92008-01-29 14:51:59 +010039#include "blk.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080040#include "blk-cgroup.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010041
Mike Snitzerd07335e2010-11-16 12:52:38 +010042EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020043EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070044EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
NeilBrowncbae8d42012-12-14 20:49:27 +010045EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010046
Tejun Heoa73f7302011-12-14 00:33:37 +010047DEFINE_IDA(blk_queue_ida);
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049/*
50 * For the allocated request tables
51 */
Jens Axboe320ae512013-10-24 09:20:05 +010052struct kmem_cache *request_cachep = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * For queue allocation
56 */
Jens Axboe6728cb02008-01-31 13:03:55 +010057struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * Controlling structure to kblockd
61 */
Jens Axboeff856ba2006-01-09 16:02:34 +010062static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Jens Axboe8324aa92008-01-29 14:51:59 +010064void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065{
66 int nr;
67
68 nr = q->nr_requests - (q->nr_requests / 8) + 1;
69 if (nr > q->nr_requests)
70 nr = q->nr_requests;
71 q->nr_congestion_on = nr;
72
73 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
74 if (nr < 1)
75 nr = 1;
76 q->nr_congestion_off = nr;
77}
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/**
80 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
81 * @bdev: device
82 *
83 * Locates the passed device's request queue and returns the address of its
84 * backing_dev_info
85 *
86 * Will return NULL if the request queue cannot be located.
87 */
88struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
89{
90 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +020091 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93 if (q)
94 ret = &q->backing_dev_info;
95 return ret;
96}
Linus Torvalds1da177e2005-04-16 15:20:36 -070097EXPORT_SYMBOL(blk_get_backing_dev_info);
98
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +020099void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200101 memset(rq, 0, sizeof(*rq));
102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700104 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200105 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100106 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900107 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200108 INIT_HLIST_NODE(&rq->hash);
109 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200110 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800111 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100112 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900113 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700114 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100115 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200117EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
NeilBrown5bb23a62007-09-27 12:46:13 +0200119static void req_bio_endio(struct request *rq, struct bio *bio,
120 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100121{
Tejun Heo143a87f2011-01-25 12:43:52 +0100122 if (error)
123 clear_bit(BIO_UPTODATE, &bio->bi_flags);
124 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
125 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100126
Tejun Heo143a87f2011-01-25 12:43:52 +0100127 if (unlikely(rq->cmd_flags & REQ_QUIET))
128 set_bit(BIO_QUIET, &bio->bi_flags);
129
Kent Overstreetf79ea412012-09-20 16:38:30 -0700130 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100131
132 /* don't actually finish bio if it's part of flush sequence */
133 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
134 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100135}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137void blk_dump_rq_flags(struct request *rq, char *msg)
138{
139 int bit;
140
Jens Axboe59533162013-05-23 12:25:08 +0200141 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200142 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200143 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Tejun Heo83096eb2009-05-07 22:24:39 +0900145 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
146 (unsigned long long)blk_rq_pos(rq),
147 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900148 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900149 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200151 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100152 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200153 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 printk("%02x ", rq->cmd[bit]);
155 printk("\n");
156 }
157}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158EXPORT_SYMBOL(blk_dump_rq_flags);
159
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500160static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200161{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500162 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200163
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500164 q = container_of(work, struct request_queue, delay_work.work);
165 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200166 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500167 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500171 * blk_delay_queue - restart queueing after defined interval
172 * @q: The &struct request_queue in question
173 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 *
175 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500176 * Sometimes queueing needs to be postponed for a little while, to allow
177 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100178 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500179 */
180void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
Bart Van Assche70460572012-11-28 13:45:56 +0100182 if (likely(!blk_queue_dead(q)))
183 queue_delayed_work(kblockd_workqueue, &q->delay_work,
184 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500186EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/**
189 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200190 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 *
192 * Description:
193 * blk_start_queue() will clear the stop flag on the queue, and call
194 * the request_fn for the queue if it was in a stopped state when
195 * entered. Also see blk_stop_queue(). Queue lock must be held.
196 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200197void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200199 WARN_ON(!irqs_disabled());
200
Nick Piggin75ad23b2008-04-29 14:48:33 +0200201 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200202 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204EXPORT_SYMBOL(blk_start_queue);
205
206/**
207 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200208 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 *
210 * Description:
211 * The Linux block layer assumes that a block driver will consume all
212 * entries on the request queue when the request_fn strategy is called.
213 * Often this will not happen, because of hardware limitations (queue
214 * depth settings). If a device driver gets a 'queue full' response,
215 * or if it simply chooses not to queue more I/O at one point, it can
216 * call this function to prevent the request_fn from being called until
217 * the driver has signalled it's ready to go again. This happens by calling
218 * blk_start_queue() to restart queue operations. Queue lock must be held.
219 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200220void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Tejun Heo136b5722012-08-21 13:18:24 -0700222 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200223 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225EXPORT_SYMBOL(blk_stop_queue);
226
227/**
228 * blk_sync_queue - cancel any pending callbacks on a queue
229 * @q: the queue
230 *
231 * Description:
232 * The block layer may perform asynchronous callback activity
233 * on a queue, such as calling the unplug function after a timeout.
234 * A block device may call blk_sync_queue to ensure that any
235 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200236 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 * that its ->make_request_fn will not re-add plugging prior to calling
238 * this function.
239 *
Vivek Goyalda527772011-03-02 19:05:33 -0500240 * This function does not cancel any asynchronous activity arising
241 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800242 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500243 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 */
245void blk_sync_queue(struct request_queue *q)
246{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100247 del_timer_sync(&q->timeout);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500248 cancel_delayed_work_sync(&q->delay_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250EXPORT_SYMBOL(blk_sync_queue);
251
252/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100253 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
254 * @q: The queue to run
255 *
256 * Description:
257 * Invoke request handling on a queue if there are any pending requests.
258 * May be used to restart request handling after a request has completed.
259 * This variant runs the queue whether or not the queue has been
260 * stopped. Must be called with the queue lock held and interrupts
261 * disabled. See also @blk_run_queue.
262 */
263inline void __blk_run_queue_uncond(struct request_queue *q)
264{
265 if (unlikely(blk_queue_dead(q)))
266 return;
267
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100268 /*
269 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
270 * the queue lock internally. As a result multiple threads may be
271 * running such a request function concurrently. Keep track of the
272 * number of active request_fn invocations such that blk_drain_queue()
273 * can wait until all these request_fn calls have finished.
274 */
275 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100276 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100277 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100278}
279
280/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200281 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200283 *
284 * Description:
285 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200286 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200288void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
Tejun Heoa538cd02009-04-23 11:05:17 +0900290 if (unlikely(blk_queue_stopped(q)))
291 return;
292
Bart Van Asschec246e802012-12-06 14:32:01 +0100293 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200294}
295EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200296
Nick Piggin75ad23b2008-04-29 14:48:33 +0200297/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200298 * blk_run_queue_async - run a single device queue in workqueue context
299 * @q: The queue to run
300 *
301 * Description:
302 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100303 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200304 */
305void blk_run_queue_async(struct request_queue *q)
306{
Bart Van Assche70460572012-11-28 13:45:56 +0100307 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700308 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200309}
Jens Axboec21e6be2011-04-19 13:32:46 +0200310EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200311
312/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200313 * blk_run_queue - run a single device queue
314 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200315 *
316 * Description:
317 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900318 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200319 */
320void blk_run_queue(struct request_queue *q)
321{
322 unsigned long flags;
323
324 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200325 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 spin_unlock_irqrestore(q->queue_lock, flags);
327}
328EXPORT_SYMBOL(blk_run_queue);
329
Jens Axboe165125e2007-07-24 09:28:11 +0200330void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500331{
332 kobject_put(&q->kobj);
333}
Jens Axboed86e0e82011-05-27 07:44:43 +0200334EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500335
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200336/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100337 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200338 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200339 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200340 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200341 * Drain requests from @q. If @drain_all is set, all requests are drained.
342 * If not, only ELVPRIV requests are drained. The caller is responsible
343 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200344 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100345static void __blk_drain_queue(struct request_queue *q, bool drain_all)
346 __releases(q->queue_lock)
347 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200348{
Asias He458f27a2012-06-15 08:45:25 +0200349 int i;
350
Bart Van Assche807592a2012-11-28 13:43:38 +0100351 lockdep_assert_held(q->queue_lock);
352
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100354 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200355
Tejun Heob855b042012-03-06 21:24:55 +0100356 /*
357 * The caller might be trying to drain @q before its
358 * elevator is initialized.
359 */
360 if (q->elevator)
361 elv_drain_elevator(q);
362
Tejun Heo5efd6112012-03-05 13:15:12 -0800363 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200364
Tejun Heo4eabc942011-12-15 20:03:04 +0100365 /*
366 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100367 * driver init after queue creation or is not yet fully
368 * active yet. Some drivers (e.g. fd and loop) get unhappy
369 * in such cases. Kick queue iff dispatch queue has
370 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100371 */
Tejun Heob855b042012-03-06 21:24:55 +0100372 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100373 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200374
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700375 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100376 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100377
378 /*
379 * Unfortunately, requests are queued at and tracked from
380 * multiple places and there's no single counter which can
381 * be drained. Check all the queues and counters.
382 */
383 if (drain_all) {
384 drain |= !list_empty(&q->queue_head);
385 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700386 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100387 drain |= q->in_flight[i];
388 drain |= !list_empty(&q->flush_queue[i]);
389 }
390 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200391
Tejun Heo481a7d62011-12-14 00:33:37 +0100392 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200393 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100394
395 spin_unlock_irq(q->queue_lock);
396
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200397 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100398
399 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200400 }
Asias He458f27a2012-06-15 08:45:25 +0200401
402 /*
403 * With queue marked dead, any woken up waiter will fail the
404 * allocation path, so the wakeup chaining is lost and we're
405 * left with hung waiters. We need to wake up those waiters.
406 */
407 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700408 struct request_list *rl;
409
Tejun Heoa0516612012-06-26 15:05:44 -0700410 blk_queue_for_each_rl(rl, q)
411 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
412 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200413 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200414}
415
Tejun Heoc9a929d2011-10-19 14:42:16 +0200416/**
Tejun Heod7325802012-03-05 13:14:58 -0800417 * blk_queue_bypass_start - enter queue bypass mode
418 * @q: queue of interest
419 *
420 * In bypass mode, only the dispatch FIFO queue of @q is used. This
421 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800422 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700423 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
424 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800425 */
426void blk_queue_bypass_start(struct request_queue *q)
427{
Tejun Heob82d4b12012-04-13 13:11:31 -0700428 bool drain;
429
Tejun Heod7325802012-03-05 13:14:58 -0800430 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700431 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800432 queue_flag_set(QUEUE_FLAG_BYPASS, q);
433 spin_unlock_irq(q->queue_lock);
434
Tejun Heob82d4b12012-04-13 13:11:31 -0700435 if (drain) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100436 spin_lock_irq(q->queue_lock);
437 __blk_drain_queue(q, false);
438 spin_unlock_irq(q->queue_lock);
439
Tejun Heob82d4b12012-04-13 13:11:31 -0700440 /* ensure blk_queue_bypass() is %true inside RCU read lock */
441 synchronize_rcu();
442 }
Tejun Heod7325802012-03-05 13:14:58 -0800443}
444EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
445
446/**
447 * blk_queue_bypass_end - leave queue bypass mode
448 * @q: queue of interest
449 *
450 * Leave bypass mode and restore the normal queueing behavior.
451 */
452void blk_queue_bypass_end(struct request_queue *q)
453{
454 spin_lock_irq(q->queue_lock);
455 if (!--q->bypass_depth)
456 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
457 WARN_ON_ONCE(q->bypass_depth < 0);
458 spin_unlock_irq(q->queue_lock);
459}
460EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
461
462/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200463 * blk_cleanup_queue - shutdown a request queue
464 * @q: request queue to shutdown
465 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100466 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
467 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500468 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100469void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500470{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200471 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700472
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100473 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500474 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100475 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200476 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800477
Tejun Heo80fd9972012-04-13 14:50:53 -0700478 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100479 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700480 * that, unlike blk_queue_bypass_start(), we aren't performing
481 * synchronize_rcu() after entering bypass mode to avoid the delay
482 * as some drivers create and destroy a lot of queues while
483 * probing. This is still safe because blk_release_queue() will be
484 * called only after the queue refcnt drops to zero and nothing,
485 * RCU or not, would be traversing the queue by then.
486 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800487 q->bypass_depth++;
488 queue_flag_set(QUEUE_FLAG_BYPASS, q);
489
Tejun Heoc9a929d2011-10-19 14:42:16 +0200490 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
491 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100492 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200493 spin_unlock_irq(lock);
494 mutex_unlock(&q->sysfs_lock);
495
Bart Van Asschec246e802012-12-06 14:32:01 +0100496 /*
497 * Drain all requests queued before DYING marking. Set DEAD flag to
498 * prevent that q->request_fn() gets invoked after draining finished.
499 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100500 spin_lock_irq(lock);
501 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100502 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100503 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200504
505 /* @q won't process any more request, flush async actions */
506 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
507 blk_sync_queue(q);
508
Asias He5e5cfac2012-05-24 23:28:52 +0800509 spin_lock_irq(lock);
510 if (q->queue_lock != &q->__queue_lock)
511 q->queue_lock = &q->__queue_lock;
512 spin_unlock_irq(lock);
513
Tejun Heoc9a929d2011-10-19 14:42:16 +0200514 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500515 blk_put_queue(q);
516}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517EXPORT_SYMBOL(blk_cleanup_queue);
518
Tejun Heo5b788ce2012-06-04 20:40:59 -0700519int blk_init_rl(struct request_list *rl, struct request_queue *q,
520 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400522 if (unlikely(rl->rq_pool))
523 return 0;
524
Tejun Heo5b788ce2012-06-04 20:40:59 -0700525 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200526 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
527 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200528 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
529 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Christoph Lameter19460892005-06-23 00:08:19 -0700531 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700532 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700533 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (!rl->rq_pool)
535 return -ENOMEM;
536
537 return 0;
538}
539
Tejun Heo5b788ce2012-06-04 20:40:59 -0700540void blk_exit_rl(struct request_list *rl)
541{
542 if (rl->rq_pool)
543 mempool_destroy(rl->rq_pool);
544}
545
Jens Axboe165125e2007-07-24 09:28:11 +0200546struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100548 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700549}
550EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Jens Axboe165125e2007-07-24 09:28:11 +0200552struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700553{
Jens Axboe165125e2007-07-24 09:28:11 +0200554 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700555 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700556
Jens Axboe8324aa92008-01-29 14:51:59 +0100557 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700558 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (!q)
560 return NULL;
561
Jens Axboe320ae512013-10-24 09:20:05 +0100562 if (percpu_counter_init(&q->mq_usage_counter, 0))
563 goto fail_q;
564
Dan Carpenter00380a42012-03-23 09:58:54 +0100565 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100566 if (q->id < 0)
Jens Axboe320ae512013-10-24 09:20:05 +0100567 goto fail_c;
Tejun Heoa73f7302011-12-14 00:33:37 +0100568
Jens Axboe0989a022009-06-12 14:42:56 +0200569 q->backing_dev_info.ra_pages =
570 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
571 q->backing_dev_info.state = 0;
572 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200573 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100574 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200575
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700576 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100577 if (err)
578 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700579
Matthew Garrett31373d02010-04-06 14:25:14 +0200580 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
581 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700582 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100583 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700584 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100585 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800586#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800587 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800588#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100589 INIT_LIST_HEAD(&q->flush_queue[0]);
590 INIT_LIST_HEAD(&q->flush_queue[1]);
591 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500592 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500593
Jens Axboe8324aa92008-01-29 14:51:59 +0100594 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Al Viro483f4af2006-03-18 18:34:37 -0500596 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700597 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500598
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500599 /*
600 * By default initialize queue_lock to internal lock and driver can
601 * override it later if need be.
602 */
603 q->queue_lock = &q->__queue_lock;
604
Tejun Heob82d4b12012-04-13 13:11:31 -0700605 /*
606 * A queue starts its life with bypass turned on to avoid
607 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700608 * init. The initial bypass will be finished when the queue is
609 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700610 */
611 q->bypass_depth = 1;
612 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
613
Jens Axboe320ae512013-10-24 09:20:05 +0100614 init_waitqueue_head(&q->mq_freeze_wq);
615
Tejun Heo5efd6112012-03-05 13:15:12 -0800616 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400617 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800618
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100620
Mikulas Patockafff49962013-10-14 12:11:36 -0400621fail_bdi:
622 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100623fail_id:
624 ida_simple_remove(&blk_queue_ida, q->id);
Jens Axboe320ae512013-10-24 09:20:05 +0100625fail_c:
626 percpu_counter_destroy(&q->mq_usage_counter);
Tejun Heoa73f7302011-12-14 00:33:37 +0100627fail_q:
628 kmem_cache_free(blk_requestq_cachep, q);
629 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630}
Christoph Lameter19460892005-06-23 00:08:19 -0700631EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633/**
634 * blk_init_queue - prepare a request queue for use with a block device
635 * @rfn: The function to be called to process requests that have been
636 * placed on the queue.
637 * @lock: Request queue spin lock
638 *
639 * Description:
640 * If a block device wishes to use the standard request handling procedures,
641 * which sorts requests and coalesces adjacent requests, then it must
642 * call blk_init_queue(). The function @rfn will be called when there
643 * are requests on the queue that need to be processed. If the device
644 * supports plugging, then @rfn may not be called immediately when requests
645 * are available on the queue, but may be called at some time later instead.
646 * Plugged queues are generally unplugged when a buffer belonging to one
647 * of the requests on the queue is needed, or due to memory pressure.
648 *
649 * @rfn is not required, or even expected, to remove all requests off the
650 * queue, but only as many as it can handle at a time. If it does leave
651 * requests on the queue, it is responsible for arranging that the requests
652 * get dealt with eventually.
653 *
654 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200655 * request queue; this lock will be taken also from interrupt context, so irq
656 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200658 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 * it didn't succeed.
660 *
661 * Note:
662 * blk_init_queue() must be paired with a blk_cleanup_queue() call
663 * when the block device is deactivated (such as at module unload).
664 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700665
Jens Axboe165125e2007-07-24 09:28:11 +0200666struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100668 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700669}
670EXPORT_SYMBOL(blk_init_queue);
671
Jens Axboe165125e2007-07-24 09:28:11 +0200672struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700673blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
674{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600675 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600677 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
678 if (!uninit_q)
679 return NULL;
680
Mike Snitzer51514122011-11-23 10:59:13 +0100681 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600682 if (!q)
683 blk_cleanup_queue(uninit_q);
684
685 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200686}
687EXPORT_SYMBOL(blk_init_queue_node);
688
689struct request_queue *
690blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
691 spinlock_t *lock)
692{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 if (!q)
694 return NULL;
695
Tejun Heoa0516612012-06-26 15:05:44 -0700696 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500697 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900701 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700702 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500703
704 /* Override internal queue lock with supplied lock pointer */
705 if (lock)
706 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Jens Axboef3b144a2009-03-06 08:48:33 +0100708 /*
709 * This also sets hw/phys segments, boundary and size
710 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200711 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Alan Stern44ec9542007-02-20 11:01:57 -0500713 q->sg_reserved_size = INT_MAX;
714
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600715 /* Protect q->elevator from elevator_change */
716 mutex_lock(&q->sysfs_lock);
717
Tejun Heob82d4b12012-04-13 13:11:31 -0700718 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600719 if (elevator_init(q, NULL)) {
720 mutex_unlock(&q->sysfs_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700721 return NULL;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600722 }
723
724 mutex_unlock(&q->sysfs_lock);
725
Tejun Heob82d4b12012-04-13 13:11:31 -0700726 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
Mike Snitzer51514122011-11-23 10:59:13 +0100728EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Tejun Heo09ac46c2011-12-14 00:33:38 +0100730bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100732 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100733 __blk_get_queue(q);
734 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736
Tejun Heo09ac46c2011-12-14 00:33:38 +0100737 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
Jens Axboed86e0e82011-05-27 07:44:43 +0200739EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Tejun Heo5b788ce2012-06-04 20:40:59 -0700741static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100743 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700744 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100745 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100746 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100747 }
748
Tejun Heo5b788ce2012-06-04 20:40:59 -0700749 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/*
753 * ioc_batching returns true if the ioc is a valid batching request and
754 * should be given priority access to a request.
755 */
Jens Axboe165125e2007-07-24 09:28:11 +0200756static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
758 if (!ioc)
759 return 0;
760
761 /*
762 * Make sure the process is able to allocate at least 1 request
763 * even if the batch times out, otherwise we could theoretically
764 * lose wakeups.
765 */
766 return ioc->nr_batch_requests == q->nr_batching ||
767 (ioc->nr_batch_requests > 0
768 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
769}
770
771/*
772 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
773 * will cause the process to be a "batcher" on all queues in the system. This
774 * is the behaviour we want though - once it gets a wakeup it should be given
775 * a nice run.
776 */
Jens Axboe165125e2007-07-24 09:28:11 +0200777static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
779 if (!ioc || ioc_batching(q, ioc))
780 return;
781
782 ioc->nr_batch_requests = q->nr_batching;
783 ioc->last_waited = jiffies;
784}
785
Tejun Heo5b788ce2012-06-04 20:40:59 -0700786static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700788 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Tejun Heoa0516612012-06-26 15:05:44 -0700790 /*
791 * bdi isn't aware of blkcg yet. As all async IOs end up root
792 * blkcg anyway, just use root blkcg state.
793 */
794 if (rl == &q->root_rl &&
795 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200796 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Jens Axboe1faa16d2009-04-06 14:48:01 +0200798 if (rl->count[sync] + 1 <= q->nr_requests) {
799 if (waitqueue_active(&rl->wait[sync]))
800 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Tejun Heo5b788ce2012-06-04 20:40:59 -0700802 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804}
805
806/*
807 * A request has just been released. Account for it, update the full and
808 * congestion status, wake up any waiters. Called under q->queue_lock.
809 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700810static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700812 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200813 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700815 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200816 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200817 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700818 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Tejun Heo5b788ce2012-06-04 20:40:59 -0700820 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Jens Axboe1faa16d2009-04-06 14:48:01 +0200822 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700823 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824}
825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100827 * Determine if elevator data should be initialized when allocating the
828 * request associated with @bio.
829 */
830static bool blk_rq_should_init_elevator(struct bio *bio)
831{
832 if (!bio)
833 return true;
834
835 /*
836 * Flush requests do not use the elevator so skip initialization.
837 * This allows a request to share the flush and elevator data.
838 */
839 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
840 return false;
841
842 return true;
843}
844
Tejun Heoda8303c2011-10-19 14:33:05 +0200845/**
Tejun Heo852c7882012-03-05 13:15:27 -0800846 * rq_ioc - determine io_context for request allocation
847 * @bio: request being allocated is for this bio (can be %NULL)
848 *
849 * Determine io_context to use for request allocation for @bio. May return
850 * %NULL if %current->io_context doesn't exist.
851 */
852static struct io_context *rq_ioc(struct bio *bio)
853{
854#ifdef CONFIG_BLK_CGROUP
855 if (bio && bio->bi_ioc)
856 return bio->bi_ioc;
857#endif
858 return current->io_context;
859}
860
861/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700862 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700863 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200864 * @rw_flags: RW and SYNC flags
865 * @bio: bio to allocate request for (can be %NULL)
866 * @gfp_mask: allocation mask
867 *
868 * Get a free request from @q. This function may fail under memory
869 * pressure or if @q is dead.
870 *
871 * Must be callled with @q->queue_lock held and,
872 * Returns %NULL on failure, with @q->queue_lock held.
873 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700875static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700876 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700878 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800879 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700880 struct elevator_type *et = q->elevator->type;
881 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100882 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200883 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200884 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100886 if (unlikely(blk_queue_dying(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200887 return NULL;
888
Jens Axboe7749a8d2006-12-13 13:02:26 +0100889 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100890 if (may_queue == ELV_MQUEUE_NO)
891 goto rq_starved;
892
Jens Axboe1faa16d2009-04-06 14:48:01 +0200893 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
894 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100895 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100896 * The queue will fill after this allocation, so set
897 * it as full, and mark this process as "batching".
898 * This process will be allowed to complete a batch of
899 * requests, others will be blocked.
900 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700901 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100902 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700903 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100904 } else {
905 if (may_queue != ELV_MQUEUE_MUST
906 && !ioc_batching(q, ioc)) {
907 /*
908 * The queue is full and the allocating
909 * process is not a "batcher", and not
910 * exempted by the IO scheduler
911 */
Tejun Heob6792812012-03-05 13:15:23 -0800912 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100913 }
914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 }
Tejun Heoa0516612012-06-26 15:05:44 -0700916 /*
917 * bdi isn't aware of blkcg yet. As all async IOs end up
918 * root blkcg anyway, just use root blkcg state.
919 */
920 if (rl == &q->root_rl)
921 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
923
Jens Axboe082cf692005-06-28 16:35:11 +0200924 /*
925 * Only allow batching queuers to allocate up to 50% over the defined
926 * limit of requests, otherwise we could have thousands of requests
927 * allocated with any setting of ->nr_requests
928 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200929 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800930 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100931
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700932 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200933 rl->count[is_sync]++;
934 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200935
Tejun Heof1f8cc92011-12-14 00:33:42 +0100936 /*
937 * Decide whether the new request will be managed by elevator. If
938 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
939 * prevent the current elevator from being destroyed until the new
940 * request is freed. This guarantees icq's won't be destroyed and
941 * makes creating new ones safe.
942 *
943 * Also, lookup icq while holding queue_lock. If it doesn't exist,
944 * it will be created after releasing queue_lock.
945 */
Tejun Heod7325802012-03-05 13:14:58 -0800946 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200947 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700948 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100949 if (et->icq_cache && ioc)
950 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100951 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200952
Jens Axboef253b862010-10-24 22:06:02 +0200953 if (blk_queue_io_stat(q))
954 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 spin_unlock_irq(q->queue_lock);
956
Tejun Heo29e2b092012-04-19 16:29:21 -0700957 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700958 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700959 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800960 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Tejun Heo29e2b092012-04-19 16:29:21 -0700962 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700963 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700964 rq->cmd_flags = rw_flags | REQ_ALLOCED;
965
Tejun Heoaaf7c682012-04-19 16:29:22 -0700966 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700967 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700968 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700969 if (ioc)
970 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700971 if (!icq)
972 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700973 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700974
975 rq->elv.icq = icq;
976 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
977 goto fail_elvpriv;
978
979 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -0700980 if (icq)
981 get_io_context(icq->ioc);
982 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700983out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100984 /*
985 * ioc may be NULL here, and ioc_batching will be false. That's
986 * OK, if the queue is under the request limit then requests need
987 * not count toward the nr_batch_requests limit. There will always
988 * be some limit enforced by BLK_BATCH_TIME.
989 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (ioc_batching(q, ioc))
991 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +0100992
Jens Axboe1faa16d2009-04-06 14:48:01 +0200993 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 return rq;
Tejun Heob6792812012-03-05 13:15:23 -0800995
Tejun Heoaaf7c682012-04-19 16:29:22 -0700996fail_elvpriv:
997 /*
998 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
999 * and may fail indefinitely under memory pressure and thus
1000 * shouldn't stall IO. Treat this request as !elvpriv. This will
1001 * disturb iosched and blkcg but weird is bettern than dead.
1002 */
1003 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1004 dev_name(q->backing_dev_info.dev));
1005
1006 rq->cmd_flags &= ~REQ_ELVPRIV;
1007 rq->elv.icq = NULL;
1008
1009 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001010 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001011 spin_unlock_irq(q->queue_lock);
1012 goto out;
1013
Tejun Heob6792812012-03-05 13:15:23 -08001014fail_alloc:
1015 /*
1016 * Allocation failed presumably due to memory. Undo anything we
1017 * might have messed up.
1018 *
1019 * Allocating task should really be put onto the front of the wait
1020 * queue, but this is pretty rare.
1021 */
1022 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001023 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001024
1025 /*
1026 * in the very unlikely event that allocation failed and no
1027 * requests for this direction was pending, mark us starved so that
1028 * freeing of a request in the other direction will notice
1029 * us. another possible fix would be to split the rq mempool into
1030 * READ and WRITE
1031 */
1032rq_starved:
1033 if (unlikely(rl->count[is_sync] == 0))
1034 rl->starved[is_sync] = 1;
1035 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
Tejun Heoda8303c2011-10-19 14:33:05 +02001038/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001039 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001040 * @q: request_queue to allocate request from
1041 * @rw_flags: RW and SYNC flags
1042 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001043 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001044 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001045 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1046 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001047 *
1048 * Must be callled with @q->queue_lock held and,
1049 * Returns %NULL on failure, with @q->queue_lock held.
1050 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001052static struct request *get_request(struct request_queue *q, int rw_flags,
1053 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001055 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001056 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001057 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001059
1060 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001061retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001062 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001063 if (rq)
1064 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001066 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001067 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001068 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Tejun Heoa06e05e2012-06-04 20:40:55 -07001071 /* wait on @rl and retry */
1072 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1073 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001074
Tejun Heoa06e05e2012-06-04 20:40:55 -07001075 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
Tejun Heoa06e05e2012-06-04 20:40:55 -07001077 spin_unlock_irq(q->queue_lock);
1078 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Tejun Heoa06e05e2012-06-04 20:40:55 -07001080 /*
1081 * After sleeping, we become a "batching" process and will be able
1082 * to allocate at least one request, and up to a big batch of them
1083 * for a small period time. See ioc_batching, ioc_set_batching
1084 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001085 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Tejun Heoa06e05e2012-06-04 20:40:55 -07001087 spin_lock_irq(q->queue_lock);
1088 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001089
Tejun Heoa06e05e2012-06-04 20:40:55 -07001090 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091}
1092
Jens Axboe320ae512013-10-24 09:20:05 +01001093static struct request *blk_old_get_request(struct request_queue *q, int rw,
1094 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 struct request *rq;
1097
1098 BUG_ON(rw != READ && rw != WRITE);
1099
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001100 /* create ioc upfront */
1101 create_io_context(gfp_mask, q->node);
1102
Nick Piggind6344532005-06-28 20:45:14 -07001103 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001104 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001105 if (!rq)
1106 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001107 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109 return rq;
1110}
Jens Axboe320ae512013-10-24 09:20:05 +01001111
1112struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1113{
1114 if (q->mq_ops)
Christoph Hellwig3228f482013-10-28 13:33:58 -06001115 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001116 else
1117 return blk_old_get_request(q, rw, gfp_mask);
1118}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119EXPORT_SYMBOL(blk_get_request);
1120
1121/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001122 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001123 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001124 * @bio: The bio describing the memory mappings that will be submitted for IO.
1125 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001126 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001127 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001128 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1129 * type commands. Where the struct request needs to be farther initialized by
1130 * the caller. It is passed a &struct bio, which describes the memory info of
1131 * the I/O transfer.
1132 *
1133 * The caller of blk_make_request must make sure that bi_io_vec
1134 * are set to describe the memory buffers. That bio_data_dir() will return
1135 * the needed direction of the request. (And all bio's in the passed bio-chain
1136 * are properly set accordingly)
1137 *
1138 * If called under none-sleepable conditions, mapped bio buffers must not
1139 * need bouncing, by calling the appropriate masked or flagged allocator,
1140 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1141 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001142 *
1143 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1144 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1145 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1146 * completion of a bio that hasn't been submitted yet, thus resulting in a
1147 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1148 * of bio_alloc(), as that avoids the mempool deadlock.
1149 * If possible a big IO should be split into smaller parts when allocation
1150 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001151 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001152struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1153 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001154{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001155 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1156
1157 if (unlikely(!rq))
1158 return ERR_PTR(-ENOMEM);
1159
1160 for_each_bio(bio) {
1161 struct bio *bounce_bio = bio;
1162 int ret;
1163
1164 blk_queue_bounce(q, &bounce_bio);
1165 ret = blk_rq_append_bio(q, rq, bounce_bio);
1166 if (unlikely(ret)) {
1167 blk_put_request(rq);
1168 return ERR_PTR(ret);
1169 }
1170 }
1171
1172 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001173}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001174EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001175
1176/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 * blk_requeue_request - put a request back on queue
1178 * @q: request queue where request should be inserted
1179 * @rq: request to be inserted
1180 *
1181 * Description:
1182 * Drivers often keep queueing requests until the hardware cannot accept
1183 * more, when that condition happens we need to put the request back
1184 * on the queue. Must be called with queue lock held.
1185 */
Jens Axboe165125e2007-07-24 09:28:11 +02001186void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001188 blk_delete_timer(rq);
1189 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001190 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 if (blk_rq_tagged(rq))
1193 blk_queue_end_tag(q, rq);
1194
James Bottomleyba396a62009-05-27 14:17:08 +02001195 BUG_ON(blk_queued_rq(rq));
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 elv_requeue_request(q, rq);
1198}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199EXPORT_SYMBOL(blk_requeue_request);
1200
Jens Axboe73c10102011-03-08 13:19:51 +01001201static void add_acct_request(struct request_queue *q, struct request *rq,
1202 int where)
1203{
Jens Axboe320ae512013-10-24 09:20:05 +01001204 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001205 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001206}
1207
Tejun Heo074a7ac2008-08-25 19:56:14 +09001208static void part_round_stats_single(int cpu, struct hd_struct *part,
1209 unsigned long now)
1210{
1211 if (now == part->stamp)
1212 return;
1213
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001214 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001215 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001216 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001217 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1218 }
1219 part->stamp = now;
1220}
1221
1222/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001223 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1224 * @cpu: cpu number for stats access
1225 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 *
1227 * The average IO queue length and utilisation statistics are maintained
1228 * by observing the current state of the queue length and the amount of
1229 * time it has been in this state for.
1230 *
1231 * Normally, that accounting is done on IO completion, but that can result
1232 * in more than a second's worth of IO being accounted for within any one
1233 * second, leading to >100% utilisation. To deal with that, we call this
1234 * function to do a round-off before returning the results when reading
1235 * /proc/diskstats. This accounts immediately for all queue usage up to
1236 * the current jiffies and restarts the counters again.
1237 */
Tejun Heoc9959052008-08-25 19:47:21 +09001238void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001239{
1240 unsigned long now = jiffies;
1241
Tejun Heo074a7ac2008-08-25 19:56:14 +09001242 if (part->partno)
1243 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1244 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001245}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001246EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001247
Lin Mingc8158812013-03-23 11:42:27 +08001248#ifdef CONFIG_PM_RUNTIME
1249static void blk_pm_put_request(struct request *rq)
1250{
1251 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1252 pm_runtime_mark_last_busy(rq->q->dev);
1253}
1254#else
1255static inline void blk_pm_put_request(struct request *rq) {}
1256#endif
1257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258/*
1259 * queue lock must be held
1260 */
Jens Axboe165125e2007-07-24 09:28:11 +02001261void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (unlikely(!q))
1264 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Lin Mingc8158812013-03-23 11:42:27 +08001266 blk_pm_put_request(req);
1267
Tejun Heo8922e162005-10-20 16:23:44 +02001268 elv_completed_request(q, req);
1269
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001270 /* this is a bio leak */
1271 WARN_ON(req->bio != NULL);
1272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 /*
1274 * Request may not have originated from ll_rw_blk. if not,
1275 * it didn't come out of our reserved rq pools
1276 */
Jens Axboe49171e52006-08-10 08:59:11 +02001277 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001278 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001279 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001282 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Tejun Heoa0516612012-06-26 15:05:44 -07001284 blk_free_request(rl, req);
1285 freed_request(rl, flags);
1286 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001289EXPORT_SYMBOL_GPL(__blk_put_request);
1290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291void blk_put_request(struct request *req)
1292{
Jens Axboe165125e2007-07-24 09:28:11 +02001293 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Jens Axboe320ae512013-10-24 09:20:05 +01001295 if (q->mq_ops)
1296 blk_mq_free_request(req);
1297 else {
1298 unsigned long flags;
1299
1300 spin_lock_irqsave(q->queue_lock, flags);
1301 __blk_put_request(q, req);
1302 spin_unlock_irqrestore(q->queue_lock, flags);
1303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305EXPORT_SYMBOL(blk_put_request);
1306
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001307/**
1308 * blk_add_request_payload - add a payload to a request
1309 * @rq: request to update
1310 * @page: page backing the payload
1311 * @len: length of the payload.
1312 *
1313 * This allows to later add a payload to an already submitted request by
1314 * a block driver. The driver needs to take care of freeing the payload
1315 * itself.
1316 *
1317 * Note that this is a quite horrible hack and nothing but handling of
1318 * discard requests should ever use it.
1319 */
1320void blk_add_request_payload(struct request *rq, struct page *page,
1321 unsigned int len)
1322{
1323 struct bio *bio = rq->bio;
1324
1325 bio->bi_io_vec->bv_page = page;
1326 bio->bi_io_vec->bv_offset = 0;
1327 bio->bi_io_vec->bv_len = len;
1328
1329 bio->bi_size = len;
1330 bio->bi_vcnt = 1;
1331 bio->bi_phys_segments = 1;
1332
1333 rq->__data_len = rq->resid_len = len;
1334 rq->nr_phys_segments = 1;
1335 rq->buffer = bio_data(bio);
1336}
1337EXPORT_SYMBOL_GPL(blk_add_request_payload);
1338
Jens Axboe320ae512013-10-24 09:20:05 +01001339bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1340 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001341{
1342 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1343
Jens Axboe73c10102011-03-08 13:19:51 +01001344 if (!ll_back_merge_fn(q, req, bio))
1345 return false;
1346
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001347 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001348
1349 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1350 blk_rq_set_mixed_merge(req);
1351
1352 req->biotail->bi_next = bio;
1353 req->biotail = bio;
1354 req->__data_len += bio->bi_size;
1355 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1356
Jens Axboe320ae512013-10-24 09:20:05 +01001357 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001358 return true;
1359}
1360
Jens Axboe320ae512013-10-24 09:20:05 +01001361bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1362 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001363{
1364 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001365
Jens Axboe73c10102011-03-08 13:19:51 +01001366 if (!ll_front_merge_fn(q, req, bio))
1367 return false;
1368
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001369 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001370
1371 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1372 blk_rq_set_mixed_merge(req);
1373
Jens Axboe73c10102011-03-08 13:19:51 +01001374 bio->bi_next = req->bio;
1375 req->bio = bio;
1376
1377 /*
1378 * may not be valid. if the low level driver said
1379 * it didn't need a bounce buffer then it better
1380 * not touch req->buffer either...
1381 */
1382 req->buffer = bio_data(bio);
1383 req->__sector = bio->bi_sector;
1384 req->__data_len += bio->bi_size;
1385 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1386
Jens Axboe320ae512013-10-24 09:20:05 +01001387 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001388 return true;
1389}
1390
Tejun Heobd87b582011-10-19 14:33:08 +02001391/**
Jens Axboe320ae512013-10-24 09:20:05 +01001392 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001393 * @q: request_queue new bio is being queued at
1394 * @bio: new bio being queued
1395 * @request_count: out parameter for number of traversed plugged requests
1396 *
1397 * Determine whether @bio being queued on @q can be merged with a request
1398 * on %current's plugged list. Returns %true if merge was successful,
1399 * otherwise %false.
1400 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001401 * Plugging coalesces IOs from the same issuer for the same purpose without
1402 * going through @q->queue_lock. As such it's more of an issuing mechanism
1403 * than scheduling, and the request, while may have elvpriv data, is not
1404 * added on the elevator at this point. In addition, we don't have
1405 * reliable access to the elevator outside queue lock. Only check basic
1406 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001407 */
Jens Axboe320ae512013-10-24 09:20:05 +01001408bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1409 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001410{
1411 struct blk_plug *plug;
1412 struct request *rq;
1413 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001414 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001415
Alireza Haghdoost23779fb2013-10-23 17:08:16 +01001416 if (blk_queue_nomerges(q))
1417 goto out;
1418
Tejun Heobd87b582011-10-19 14:33:08 +02001419 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001420 if (!plug)
1421 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001422 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001423
Shaohua Li92f399c2013-10-29 12:01:03 -06001424 if (q->mq_ops)
1425 plug_list = &plug->mq_list;
1426 else
1427 plug_list = &plug->list;
1428
1429 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001430 int el_ret;
1431
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001432 if (rq->q == q)
1433 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001434
Tejun Heo07c2bd32012-02-08 09:19:42 +01001435 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001436 continue;
1437
Tejun Heo050c8ea2012-02-08 09:19:38 +01001438 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001439 if (el_ret == ELEVATOR_BACK_MERGE) {
1440 ret = bio_attempt_back_merge(q, rq, bio);
1441 if (ret)
1442 break;
1443 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1444 ret = bio_attempt_front_merge(q, rq, bio);
1445 if (ret)
1446 break;
1447 }
1448 }
1449out:
1450 return ret;
1451}
1452
Jens Axboe86db1e22008-01-29 14:53:40 +01001453void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001454{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001455 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001456
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001457 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1458 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001459 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001460
Tejun Heo52d9e672006-01-06 09:49:58 +01001461 req->errors = 0;
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001462 req->__sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001463 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001464 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001465}
1466
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001467void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001469 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001470 struct blk_plug *plug;
1471 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1472 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001473 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 /*
1476 * low level driver can indicate that it wants pages above a
1477 * certain limit bounced to low memory (ie for highmem, or even
1478 * ISA dma in theory)
1479 */
1480 blk_queue_bounce(q, &bio);
1481
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001482 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1483 bio_endio(bio, -EIO);
1484 return;
1485 }
1486
Tejun Heo4fed9472010-09-03 11:56:17 +02001487 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001488 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001489 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001490 goto get_rq;
1491 }
1492
Jens Axboe73c10102011-03-08 13:19:51 +01001493 /*
1494 * Check if we can merge with the plugged list before grabbing
1495 * any locks.
1496 */
Jens Axboe320ae512013-10-24 09:20:05 +01001497 if (blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001498 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001499
1500 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001503 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001504 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001505 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001506 if (!attempt_back_merge(q, req))
1507 elv_merged_request(q, req, el_ret);
1508 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001509 }
Jens Axboe73c10102011-03-08 13:19:51 +01001510 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001511 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001512 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001513 if (!attempt_front_merge(q, req))
1514 elv_merged_request(q, req, el_ret);
1515 goto out_unlock;
1516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 }
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001520 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001521 * This sync check and mask will be re-done in init_request_from_bio(),
1522 * but we need to set it earlier to expose the sync flag to the
1523 * rq allocator and io schedulers.
1524 */
1525 rw_flags = bio_data_dir(bio);
1526 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001527 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001528
1529 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001530 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001531 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001532 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001533 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001534 if (unlikely(!req)) {
1535 bio_endio(bio, -ENODEV); /* @q is dead */
1536 goto out_unlock;
1537 }
Nick Piggind6344532005-06-28 20:45:14 -07001538
Nick Piggin450991b2005-06-28 20:45:13 -07001539 /*
1540 * After dropping the lock and possibly sleeping here, our request
1541 * may now be mergeable after it had proven unmergeable (above).
1542 * We don't worry about that case for efficiency. It won't happen
1543 * often, and the elevators are able to handle it.
1544 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001545 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Tao Ma9562ad92011-10-24 16:11:30 +02001547 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001548 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001549
Jens Axboe73c10102011-03-08 13:19:51 +01001550 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001551 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001552 /*
1553 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001554 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001555 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001556 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001557 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001558 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001559 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001560 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001561 trace_block_plug(q);
1562 }
Jens Axboe73c10102011-03-08 13:19:51 +01001563 }
Shaohua Lia6327162011-08-24 16:04:32 +02001564 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001565 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001566 } else {
1567 spin_lock_irq(q->queue_lock);
1568 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001569 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001570out_unlock:
1571 spin_unlock_irq(q->queue_lock);
1572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573}
Jens Axboec20e8de2011-09-12 12:03:37 +02001574EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576/*
1577 * If bio->bi_dev is a partition, remap the location
1578 */
1579static inline void blk_partition_remap(struct bio *bio)
1580{
1581 struct block_device *bdev = bio->bi_bdev;
1582
Jens Axboebf2de6f2007-09-27 13:01:25 +02001583 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 bio->bi_sector += p->start_sect;
1587 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001588
Mike Snitzerd07335e2010-11-16 12:52:38 +01001589 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1590 bdev->bd_dev,
1591 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
1593}
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595static void handle_bad_sector(struct bio *bio)
1596{
1597 char b[BDEVNAME_SIZE];
1598
1599 printk(KERN_INFO "attempt to access beyond end of device\n");
1600 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1601 bdevname(bio->bi_bdev, b),
1602 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001603 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001604 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 set_bit(BIO_EOF, &bio->bi_flags);
1607}
1608
Akinobu Mitac17bb492006-12-08 02:39:46 -08001609#ifdef CONFIG_FAIL_MAKE_REQUEST
1610
1611static DECLARE_FAULT_ATTR(fail_make_request);
1612
1613static int __init setup_fail_make_request(char *str)
1614{
1615 return setup_fault_attr(&fail_make_request, str);
1616}
1617__setup("fail_make_request=", setup_fail_make_request);
1618
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001619static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001620{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001621 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001622}
1623
1624static int __init fail_make_request_debugfs(void)
1625{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001626 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1627 NULL, &fail_make_request);
1628
1629 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001630}
1631
1632late_initcall(fail_make_request_debugfs);
1633
1634#else /* CONFIG_FAIL_MAKE_REQUEST */
1635
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001636static inline bool should_fail_request(struct hd_struct *part,
1637 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001638{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001639 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001640}
1641
1642#endif /* CONFIG_FAIL_MAKE_REQUEST */
1643
Jens Axboec07e2b42007-07-18 13:27:58 +02001644/*
1645 * Check whether this bio extends beyond the end of the device.
1646 */
1647static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1648{
1649 sector_t maxsector;
1650
1651 if (!nr_sectors)
1652 return 0;
1653
1654 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001655 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001656 if (maxsector) {
1657 sector_t sector = bio->bi_sector;
1658
1659 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1660 /*
1661 * This may well happen - the kernel calls bread()
1662 * without checking the size of the device, e.g., when
1663 * mounting a device.
1664 */
1665 handle_bad_sector(bio);
1666 return 1;
1667 }
1668 }
1669
1670 return 0;
1671}
1672
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001673static noinline_for_stack bool
1674generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Jens Axboe165125e2007-07-24 09:28:11 +02001676 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001677 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001678 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001679 char b[BDEVNAME_SIZE];
1680 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
1682 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Jens Axboec07e2b42007-07-18 13:27:58 +02001684 if (bio_check_eod(bio, nr_sectors))
1685 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001687 q = bdev_get_queue(bio->bi_bdev);
1688 if (unlikely(!q)) {
1689 printk(KERN_ERR
1690 "generic_make_request: Trying to access "
1691 "nonexistent block-device %s (%Lu)\n",
1692 bdevname(bio->bi_bdev, b),
1693 (long long) bio->bi_sector);
1694 goto end_io;
1695 }
1696
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001697 if (likely(bio_is_rw(bio) &&
1698 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001699 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1700 bdevname(bio->bi_bdev, b),
1701 bio_sectors(bio),
1702 queue_max_hw_sectors(q));
1703 goto end_io;
1704 }
1705
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001706 part = bio->bi_bdev->bd_part;
1707 if (should_fail_request(part, bio->bi_size) ||
1708 should_fail_request(&part_to_disk(part)->part0,
1709 bio->bi_size))
1710 goto end_io;
1711
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001713 * If this device has partitions, remap block n
1714 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001716 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001718 if (bio_check_eod(bio, nr_sectors))
1719 goto end_io;
1720
1721 /*
1722 * Filter flush bio's early so that make_request based
1723 * drivers without flush support don't have to worry
1724 * about them.
1725 */
1726 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1727 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1728 if (!nr_sectors) {
1729 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001730 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001734 if ((bio->bi_rw & REQ_DISCARD) &&
1735 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001736 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001737 err = -EOPNOTSUPP;
1738 goto end_io;
1739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001741 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 err = -EOPNOTSUPP;
1743 goto end_io;
1744 }
1745
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001746 /*
1747 * Various block parts want %current->io_context and lazy ioc
1748 * allocation ends up trading a lot of pain for a small amount of
1749 * memory. Just allocate it upfront. This may fail and block
1750 * layer knows how to live with it.
1751 */
1752 create_io_context(GFP_ATOMIC, q->node);
1753
Tejun Heobc16a4f2011-10-19 14:33:01 +02001754 if (blk_throtl_bio(q, bio))
1755 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001756
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001757 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001758 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001759
1760end_io:
1761 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001762 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
1764
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001765/**
1766 * generic_make_request - hand a buffer to its device driver for I/O
1767 * @bio: The bio describing the location in memory and on the device.
1768 *
1769 * generic_make_request() is used to make I/O requests of block
1770 * devices. It is passed a &struct bio, which describes the I/O that needs
1771 * to be done.
1772 *
1773 * generic_make_request() does not return any status. The
1774 * success/failure status of the request, along with notification of
1775 * completion, is delivered asynchronously through the bio->bi_end_io
1776 * function described (one day) else where.
1777 *
1778 * The caller of generic_make_request must make sure that bi_io_vec
1779 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1780 * set to describe the device address, and the
1781 * bi_end_io and optionally bi_private are set to describe how
1782 * completion notification should be signaled.
1783 *
1784 * generic_make_request and the drivers it calls may use bi_next if this
1785 * bio happens to be merged with someone else, and may resubmit the bio to
1786 * a lower device by calling into generic_make_request recursively, which
1787 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001788 */
1789void generic_make_request(struct bio *bio)
1790{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001791 struct bio_list bio_list_on_stack;
1792
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001793 if (!generic_make_request_checks(bio))
1794 return;
1795
1796 /*
1797 * We only want one ->make_request_fn to be active at a time, else
1798 * stack usage with stacked devices could be a problem. So use
1799 * current->bio_list to keep a list of requests submited by a
1800 * make_request_fn function. current->bio_list is also used as a
1801 * flag to say if generic_make_request is currently active in this
1802 * task or not. If it is NULL, then no make_request is active. If
1803 * it is non-NULL, then a make_request is active, and new requests
1804 * should be added at the tail
1805 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001806 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001807 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001808 return;
1809 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001810
Neil Brownd89d8792007-05-01 09:53:42 +02001811 /* following loop may be a bit non-obvious, and so deserves some
1812 * explanation.
1813 * Before entering the loop, bio->bi_next is NULL (as all callers
1814 * ensure that) so we have a list with a single bio.
1815 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001816 * we assign bio_list to a pointer to the bio_list_on_stack,
1817 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001818 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001819 * through a recursive call to generic_make_request. If it
1820 * did, we find a non-NULL value in bio_list and re-enter the loop
1821 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001822 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001823 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001824 */
1825 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001826 bio_list_init(&bio_list_on_stack);
1827 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001828 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001829 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1830
1831 q->make_request_fn(q, bio);
1832
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001833 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001834 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001835 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001836}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837EXPORT_SYMBOL(generic_make_request);
1838
1839/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001840 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1842 * @bio: The &struct bio which describes the I/O
1843 *
1844 * submit_bio() is very similar in purpose to generic_make_request(), and
1845 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001846 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 *
1848 */
1849void submit_bio(int rw, struct bio *bio)
1850{
Jens Axboe22e2c502005-06-27 10:55:12 +02001851 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Jens Axboebf2de6f2007-09-27 13:01:25 +02001853 /*
1854 * If it's a regular read/write or a barrier with data attached,
1855 * go through the normal accounting stuff before submission.
1856 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001857 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001858 unsigned int count;
1859
1860 if (unlikely(rw & REQ_WRITE_SAME))
1861 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1862 else
1863 count = bio_sectors(bio);
1864
Jens Axboebf2de6f2007-09-27 13:01:25 +02001865 if (rw & WRITE) {
1866 count_vm_events(PGPGOUT, count);
1867 } else {
1868 task_io_account_read(bio->bi_size);
1869 count_vm_events(PGPGIN, count);
1870 }
1871
1872 if (unlikely(block_dump)) {
1873 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001874 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001875 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001876 (rw & WRITE) ? "WRITE" : "READ",
1877 (unsigned long long)bio->bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001878 bdevname(bio->bi_bdev, b),
1879 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 }
1882
1883 generic_make_request(bio);
1884}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885EXPORT_SYMBOL(submit_bio);
1886
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001887/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001888 * blk_rq_check_limits - Helper function to check a request for the queue limit
1889 * @q: the queue
1890 * @rq: the request being checked
1891 *
1892 * Description:
1893 * @rq may have been made based on weaker limitations of upper-level queues
1894 * in request stacking drivers, and it may violate the limitation of @q.
1895 * Since the block layer and the underlying device driver trust @rq
1896 * after it is inserted to @q, it should be checked against @q before
1897 * the insertion using this generic function.
1898 *
1899 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001900 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001901 * Request stacking drivers like request-based dm may change the queue
1902 * limits while requests are in the queue (e.g. dm's table swapping).
1903 * Such request stacking drivers should check those requests agaist
1904 * the new queue limits again when they dispatch those requests,
1905 * although such checkings are also done against the old queue limits
1906 * when submitting requests.
1907 */
1908int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1909{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001910 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001911 return 0;
1912
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001913 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001914 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1915 return -EIO;
1916 }
1917
1918 /*
1919 * queue's settings related to segment counting like q->bounce_pfn
1920 * may differ from that of other stacking queues.
1921 * Recalculate it to check the request correctly on this queue's
1922 * limitation.
1923 */
1924 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001925 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001926 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1927 return -EIO;
1928 }
1929
1930 return 0;
1931}
1932EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1933
1934/**
1935 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1936 * @q: the queue to submit the request
1937 * @rq: the request being queued
1938 */
1939int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1940{
1941 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001942 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001943
1944 if (blk_rq_check_limits(q, rq))
1945 return -EIO;
1946
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001947 if (rq->rq_disk &&
1948 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001949 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001950
1951 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001952 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01001953 spin_unlock_irqrestore(q->queue_lock, flags);
1954 return -ENODEV;
1955 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001956
1957 /*
1958 * Submitting request must be dequeued before calling this function
1959 * because it will be linked to another request_queue
1960 */
1961 BUG_ON(blk_queued_rq(rq));
1962
Jeff Moyer4853aba2011-08-15 21:37:25 +02001963 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1964 where = ELEVATOR_INSERT_FLUSH;
1965
1966 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001967 if (where == ELEVATOR_INSERT_FLUSH)
1968 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001969 spin_unlock_irqrestore(q->queue_lock, flags);
1970
1971 return 0;
1972}
1973EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1974
Tejun Heo80a761f2009-07-03 17:48:17 +09001975/**
1976 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1977 * @rq: request to examine
1978 *
1979 * Description:
1980 * A request could be merge of IOs which require different failure
1981 * handling. This function determines the number of bytes which
1982 * can be failed from the beginning of the request without
1983 * crossing into area which need to be retried further.
1984 *
1985 * Return:
1986 * The number of bytes to fail.
1987 *
1988 * Context:
1989 * queue_lock must be held.
1990 */
1991unsigned int blk_rq_err_bytes(const struct request *rq)
1992{
1993 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1994 unsigned int bytes = 0;
1995 struct bio *bio;
1996
1997 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1998 return blk_rq_bytes(rq);
1999
2000 /*
2001 * Currently the only 'mixing' which can happen is between
2002 * different fastfail types. We can safely fail portions
2003 * which have all the failfast bits that the first one has -
2004 * the ones which are at least as eager to fail as the first
2005 * one.
2006 */
2007 for (bio = rq->bio; bio; bio = bio->bi_next) {
2008 if ((bio->bi_rw & ff) != ff)
2009 break;
2010 bytes += bio->bi_size;
2011 }
2012
2013 /* this could lead to infinite loop */
2014 BUG_ON(blk_rq_bytes(rq) && !bytes);
2015 return bytes;
2016}
2017EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2018
Jens Axboe320ae512013-10-24 09:20:05 +01002019void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002020{
Jens Axboec2553b52009-04-24 08:10:11 +02002021 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002022 const int rw = rq_data_dir(req);
2023 struct hd_struct *part;
2024 int cpu;
2025
2026 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002027 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002028 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2029 part_stat_unlock();
2030 }
2031}
2032
Jens Axboe320ae512013-10-24 09:20:05 +01002033void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002034{
Jens Axboebc58ba92009-01-23 10:54:44 +01002035 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002036 * Account IO completion. flush_rq isn't accounted as a
2037 * normal IO on queueing nor completion. Accounting the
2038 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002039 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002040 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002041 unsigned long duration = jiffies - req->start_time;
2042 const int rw = rq_data_dir(req);
2043 struct hd_struct *part;
2044 int cpu;
2045
2046 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002047 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002048
2049 part_stat_inc(cpu, part, ios[rw]);
2050 part_stat_add(cpu, part, ticks[rw], duration);
2051 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002052 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002053
Jens Axboe6c23a962011-01-07 08:43:37 +01002054 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002055 part_stat_unlock();
2056 }
2057}
2058
Lin Mingc8158812013-03-23 11:42:27 +08002059#ifdef CONFIG_PM_RUNTIME
2060/*
2061 * Don't process normal requests when queue is suspended
2062 * or in the process of suspending/resuming
2063 */
2064static struct request *blk_pm_peek_request(struct request_queue *q,
2065 struct request *rq)
2066{
2067 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2068 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2069 return NULL;
2070 else
2071 return rq;
2072}
2073#else
2074static inline struct request *blk_pm_peek_request(struct request_queue *q,
2075 struct request *rq)
2076{
2077 return rq;
2078}
2079#endif
2080
Jens Axboe320ae512013-10-24 09:20:05 +01002081void blk_account_io_start(struct request *rq, bool new_io)
2082{
2083 struct hd_struct *part;
2084 int rw = rq_data_dir(rq);
2085 int cpu;
2086
2087 if (!blk_do_io_stat(rq))
2088 return;
2089
2090 cpu = part_stat_lock();
2091
2092 if (!new_io) {
2093 part = rq->part;
2094 part_stat_inc(cpu, part, merges[rw]);
2095 } else {
2096 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2097 if (!hd_struct_try_get(part)) {
2098 /*
2099 * The partition is already being removed,
2100 * the request will be accounted on the disk only
2101 *
2102 * We take a reference on disk->part0 although that
2103 * partition will never be deleted, so we can treat
2104 * it as any other partition.
2105 */
2106 part = &rq->rq_disk->part0;
2107 hd_struct_get(part);
2108 }
2109 part_round_stats(cpu, part);
2110 part_inc_in_flight(part, rw);
2111 rq->part = part;
2112 }
2113
2114 part_stat_unlock();
2115}
2116
Tejun Heo53a08802008-12-03 12:41:26 +01002117/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002118 * blk_peek_request - peek at the top of a request queue
2119 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002120 *
2121 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002122 * Return the request at the top of @q. The returned request
2123 * should be started using blk_start_request() before LLD starts
2124 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002125 *
2126 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002127 * Pointer to the request at the top of @q if available. Null
2128 * otherwise.
2129 *
2130 * Context:
2131 * queue_lock must be held.
2132 */
2133struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002134{
2135 struct request *rq;
2136 int ret;
2137
2138 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002139
2140 rq = blk_pm_peek_request(q, rq);
2141 if (!rq)
2142 break;
2143
Tejun Heo158dbda2009-04-23 11:05:18 +09002144 if (!(rq->cmd_flags & REQ_STARTED)) {
2145 /*
2146 * This is the first time the device driver
2147 * sees this request (possibly after
2148 * requeueing). Notify IO scheduler.
2149 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002150 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002151 elv_activate_rq(q, rq);
2152
2153 /*
2154 * just mark as started even if we don't start
2155 * it, a request that has been delayed should
2156 * not be passed by new incoming requests
2157 */
2158 rq->cmd_flags |= REQ_STARTED;
2159 trace_block_rq_issue(q, rq);
2160 }
2161
2162 if (!q->boundary_rq || q->boundary_rq == rq) {
2163 q->end_sector = rq_end_sector(rq);
2164 q->boundary_rq = NULL;
2165 }
2166
2167 if (rq->cmd_flags & REQ_DONTPREP)
2168 break;
2169
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002170 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002171 /*
2172 * make sure space for the drain appears we
2173 * know we can do this because max_hw_segments
2174 * has been adjusted to be one fewer than the
2175 * device can handle
2176 */
2177 rq->nr_phys_segments++;
2178 }
2179
2180 if (!q->prep_rq_fn)
2181 break;
2182
2183 ret = q->prep_rq_fn(q, rq);
2184 if (ret == BLKPREP_OK) {
2185 break;
2186 } else if (ret == BLKPREP_DEFER) {
2187 /*
2188 * the request may have been (partially) prepped.
2189 * we need to keep this request in the front to
2190 * avoid resource deadlock. REQ_STARTED will
2191 * prevent other fs requests from passing this one.
2192 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002193 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002194 !(rq->cmd_flags & REQ_DONTPREP)) {
2195 /*
2196 * remove the space for the drain we added
2197 * so that we don't add it again
2198 */
2199 --rq->nr_phys_segments;
2200 }
2201
2202 rq = NULL;
2203 break;
2204 } else if (ret == BLKPREP_KILL) {
2205 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002206 /*
2207 * Mark this request as started so we don't trigger
2208 * any debug logic in the end I/O path.
2209 */
2210 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002211 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002212 } else {
2213 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2214 break;
2215 }
2216 }
2217
2218 return rq;
2219}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002220EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002221
Tejun Heo9934c8c2009-05-08 11:54:16 +09002222void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002223{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002224 struct request_queue *q = rq->q;
2225
Tejun Heo158dbda2009-04-23 11:05:18 +09002226 BUG_ON(list_empty(&rq->queuelist));
2227 BUG_ON(ELV_ON_HASH(rq));
2228
2229 list_del_init(&rq->queuelist);
2230
2231 /*
2232 * the time frame between a request being removed from the lists
2233 * and to it is freed is accounted as io that is in progress at
2234 * the driver side.
2235 */
Divyesh Shah91952912010-04-01 15:01:41 -07002236 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002237 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002238 set_io_start_time_ns(rq);
2239 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002240}
2241
Tejun Heo5efccd12009-04-23 11:05:18 +09002242/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002243 * blk_start_request - start request processing on the driver
2244 * @req: request to dequeue
2245 *
2246 * Description:
2247 * Dequeue @req and start timeout timer on it. This hands off the
2248 * request to the driver.
2249 *
2250 * Block internal functions which don't want to start timer should
2251 * call blk_dequeue_request().
2252 *
2253 * Context:
2254 * queue_lock must be held.
2255 */
2256void blk_start_request(struct request *req)
2257{
2258 blk_dequeue_request(req);
2259
2260 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002261 * We are now handing the request to the hardware, initialize
2262 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002263 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002264 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002265 if (unlikely(blk_bidi_rq(req)))
2266 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2267
Jeff Moyer4912aa62013-10-08 14:36:41 -04002268 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002269 blk_add_timer(req);
2270}
2271EXPORT_SYMBOL(blk_start_request);
2272
2273/**
2274 * blk_fetch_request - fetch a request from a request queue
2275 * @q: request queue to fetch a request from
2276 *
2277 * Description:
2278 * Return the request at the top of @q. The request is started on
2279 * return and LLD can start processing it immediately.
2280 *
2281 * Return:
2282 * Pointer to the request at the top of @q if available. Null
2283 * otherwise.
2284 *
2285 * Context:
2286 * queue_lock must be held.
2287 */
2288struct request *blk_fetch_request(struct request_queue *q)
2289{
2290 struct request *rq;
2291
2292 rq = blk_peek_request(q);
2293 if (rq)
2294 blk_start_request(rq);
2295 return rq;
2296}
2297EXPORT_SYMBOL(blk_fetch_request);
2298
2299/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002300 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002301 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002302 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002303 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002304 *
2305 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002306 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2307 * the request structure even if @req doesn't have leftover.
2308 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002309 *
2310 * This special helper function is only for request stacking drivers
2311 * (e.g. request-based dm) so that they can handle partial completion.
2312 * Actual device drivers should use blk_end_request instead.
2313 *
2314 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2315 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002316 *
2317 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002318 * %false - this request doesn't have any more data
2319 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002320 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002321bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002323 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
Tejun Heo2e60e022009-04-23 11:05:18 +09002325 if (!req->bio)
2326 return false;
2327
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002328 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002331 * For fs requests, rq is just carrier of independent bio's
2332 * and each partial completion should be handled separately.
2333 * Reset per-request error on each partial completion.
2334 *
2335 * TODO: tj: This is too subtle. It would be better to let
2336 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002338 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 req->errors = 0;
2340
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002341 if (error && req->cmd_type == REQ_TYPE_FS &&
2342 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002343 char *error_type;
2344
2345 switch (error) {
2346 case -ENOLINK:
2347 error_type = "recoverable transport";
2348 break;
2349 case -EREMOTEIO:
2350 error_type = "critical target";
2351 break;
2352 case -EBADE:
2353 error_type = "critical nexus";
2354 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002355 case -ETIMEDOUT:
2356 error_type = "timeout";
2357 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002358 case -ENOSPC:
2359 error_type = "critical space allocation";
2360 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002361 case -ENODATA:
2362 error_type = "critical medium";
2363 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002364 case -EIO:
2365 default:
2366 error_type = "I/O";
2367 break;
2368 }
Yi Zou37d7b342012-08-30 16:26:25 -07002369 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2370 error_type, req->rq_disk ?
2371 req->rq_disk->disk_name : "?",
2372 (unsigned long long)blk_rq_pos(req));
2373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 }
2375
Jens Axboebc58ba92009-01-23 10:54:44 +01002376 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002377
Kent Overstreetf79ea412012-09-20 16:38:30 -07002378 total_bytes = 0;
2379 while (req->bio) {
2380 struct bio *bio = req->bio;
2381 unsigned bio_bytes = min(bio->bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Kent Overstreetf79ea412012-09-20 16:38:30 -07002383 if (bio_bytes == bio->bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Kent Overstreetf79ea412012-09-20 16:38:30 -07002386 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
Kent Overstreetf79ea412012-09-20 16:38:30 -07002388 total_bytes += bio_bytes;
2389 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
Kent Overstreetf79ea412012-09-20 16:38:30 -07002391 if (!nr_bytes)
2392 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 }
2394
2395 /*
2396 * completely done
2397 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002398 if (!req->bio) {
2399 /*
2400 * Reset counters so that the request stacking driver
2401 * can find how many bytes remain in the request
2402 * later.
2403 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002404 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002405 return false;
2406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002408 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002409 req->buffer = bio_data(req->bio);
2410
2411 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002412 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002413 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002414
Tejun Heo80a761f2009-07-03 17:48:17 +09002415 /* mixed attributes always follow the first bio */
2416 if (req->cmd_flags & REQ_MIXED_MERGE) {
2417 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2418 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2419 }
2420
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002421 /*
2422 * If total number of sectors is less than the first segment
2423 * size, something has gone terribly wrong.
2424 */
2425 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002426 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002427 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002428 }
2429
2430 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002432
Tejun Heo2e60e022009-04-23 11:05:18 +09002433 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434}
Tejun Heo2e60e022009-04-23 11:05:18 +09002435EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Tejun Heo2e60e022009-04-23 11:05:18 +09002437static bool blk_update_bidi_request(struct request *rq, int error,
2438 unsigned int nr_bytes,
2439 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002440{
Tejun Heo2e60e022009-04-23 11:05:18 +09002441 if (blk_update_request(rq, error, nr_bytes))
2442 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002443
Tejun Heo2e60e022009-04-23 11:05:18 +09002444 /* Bidi request must be completed as a whole */
2445 if (unlikely(blk_bidi_rq(rq)) &&
2446 blk_update_request(rq->next_rq, error, bidi_bytes))
2447 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002448
Jens Axboee2e1a142010-06-09 10:42:09 +02002449 if (blk_queue_add_random(rq->q))
2450 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002451
2452 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453}
2454
James Bottomley28018c22010-07-01 19:49:17 +09002455/**
2456 * blk_unprep_request - unprepare a request
2457 * @req: the request
2458 *
2459 * This function makes a request ready for complete resubmission (or
2460 * completion). It happens only after all error handling is complete,
2461 * so represents the appropriate moment to deallocate any resources
2462 * that were allocated to the request in the prep_rq_fn. The queue
2463 * lock is held when calling this.
2464 */
2465void blk_unprep_request(struct request *req)
2466{
2467 struct request_queue *q = req->q;
2468
2469 req->cmd_flags &= ~REQ_DONTPREP;
2470 if (q->unprep_rq_fn)
2471 q->unprep_rq_fn(q, req);
2472}
2473EXPORT_SYMBOL_GPL(blk_unprep_request);
2474
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475/*
2476 * queue lock must be held
2477 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002478static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002480 if (blk_rq_tagged(req))
2481 blk_queue_end_tag(req->q, req);
2482
James Bottomleyba396a62009-05-27 14:17:08 +02002483 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002485 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002486 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Mike Andersone78042e2008-10-30 02:16:20 -07002488 blk_delete_timer(req);
2489
James Bottomley28018c22010-07-01 19:49:17 +09002490 if (req->cmd_flags & REQ_DONTPREP)
2491 blk_unprep_request(req);
2492
Jens Axboebc58ba92009-01-23 10:54:44 +01002493 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002494
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002496 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002497 else {
2498 if (blk_bidi_rq(req))
2499 __blk_put_request(req->next_rq->q, req->next_rq);
2500
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 }
2503}
2504
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002505/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002506 * blk_end_bidi_request - Complete a bidi request
2507 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002508 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002509 * @nr_bytes: number of bytes to complete @rq
2510 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002511 *
2512 * Description:
2513 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002514 * Drivers that supports bidi can safely call this member for any
2515 * type of request, bidi or uni. In the later case @bidi_bytes is
2516 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002517 *
2518 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002519 * %false - we are done with this request
2520 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002521 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002522static bool blk_end_bidi_request(struct request *rq, int error,
2523 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002524{
2525 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002526 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002527
Tejun Heo2e60e022009-04-23 11:05:18 +09002528 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2529 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002530
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002531 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002532 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002533 spin_unlock_irqrestore(q->queue_lock, flags);
2534
Tejun Heo2e60e022009-04-23 11:05:18 +09002535 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002536}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002537
2538/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002539 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2540 * @rq: the request to complete
2541 * @error: %0 for success, < %0 for error
2542 * @nr_bytes: number of bytes to complete @rq
2543 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002544 *
2545 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002546 * Identical to blk_end_bidi_request() except that queue lock is
2547 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002548 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002549 * Return:
2550 * %false - we are done with this request
2551 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002552 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002553bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002554 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002555{
Tejun Heo2e60e022009-04-23 11:05:18 +09002556 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2557 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002558
Tejun Heo2e60e022009-04-23 11:05:18 +09002559 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002560
Tejun Heo2e60e022009-04-23 11:05:18 +09002561 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002562}
2563
2564/**
2565 * blk_end_request - Helper function for drivers to complete the request.
2566 * @rq: the request being processed
2567 * @error: %0 for success, < %0 for error
2568 * @nr_bytes: number of bytes to complete
2569 *
2570 * Description:
2571 * Ends I/O on a number of bytes attached to @rq.
2572 * If @rq has leftover, sets it up for the next range of segments.
2573 *
2574 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002575 * %false - we are done with this request
2576 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002577 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002578bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002579{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002580 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002581}
Jens Axboe56ad1742009-07-28 22:11:24 +02002582EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002583
2584/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002585 * blk_end_request_all - Helper function for drives to finish the request.
2586 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002587 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002588 *
2589 * Description:
2590 * Completely finish @rq.
2591 */
2592void blk_end_request_all(struct request *rq, int error)
2593{
2594 bool pending;
2595 unsigned int bidi_bytes = 0;
2596
2597 if (unlikely(blk_bidi_rq(rq)))
2598 bidi_bytes = blk_rq_bytes(rq->next_rq);
2599
2600 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2601 BUG_ON(pending);
2602}
Jens Axboe56ad1742009-07-28 22:11:24 +02002603EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002604
2605/**
2606 * blk_end_request_cur - Helper function to finish the current request chunk.
2607 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002608 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002609 *
2610 * Description:
2611 * Complete the current consecutively mapped chunk from @rq.
2612 *
2613 * Return:
2614 * %false - we are done with this request
2615 * %true - still buffers pending for this request
2616 */
2617bool blk_end_request_cur(struct request *rq, int error)
2618{
2619 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2620}
Jens Axboe56ad1742009-07-28 22:11:24 +02002621EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002622
2623/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002624 * blk_end_request_err - Finish a request till the next failure boundary.
2625 * @rq: the request to finish till the next failure boundary for
2626 * @error: must be negative errno
2627 *
2628 * Description:
2629 * Complete @rq till the next failure boundary.
2630 *
2631 * Return:
2632 * %false - we are done with this request
2633 * %true - still buffers pending for this request
2634 */
2635bool blk_end_request_err(struct request *rq, int error)
2636{
2637 WARN_ON(error >= 0);
2638 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2639}
2640EXPORT_SYMBOL_GPL(blk_end_request_err);
2641
2642/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002643 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002644 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002645 * @error: %0 for success, < %0 for error
2646 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002647 *
2648 * Description:
2649 * Must be called with queue lock held unlike blk_end_request().
2650 *
2651 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002652 * %false - we are done with this request
2653 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002654 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002655bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002656{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002657 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002658}
Jens Axboe56ad1742009-07-28 22:11:24 +02002659EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002660
2661/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002662 * __blk_end_request_all - Helper function for drives to finish the request.
2663 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002664 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002665 *
2666 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002667 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002668 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002669void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002670{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002671 bool pending;
2672 unsigned int bidi_bytes = 0;
2673
2674 if (unlikely(blk_bidi_rq(rq)))
2675 bidi_bytes = blk_rq_bytes(rq->next_rq);
2676
2677 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2678 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002679}
Jens Axboe56ad1742009-07-28 22:11:24 +02002680EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002681
2682/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002683 * __blk_end_request_cur - Helper function to finish the current request chunk.
2684 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002685 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002686 *
2687 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002688 * Complete the current consecutively mapped chunk from @rq. Must
2689 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002690 *
2691 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002692 * %false - we are done with this request
2693 * %true - still buffers pending for this request
2694 */
2695bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002696{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002697 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002698}
Jens Axboe56ad1742009-07-28 22:11:24 +02002699EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002700
Tejun Heo80a761f2009-07-03 17:48:17 +09002701/**
2702 * __blk_end_request_err - Finish a request till the next failure boundary.
2703 * @rq: the request to finish till the next failure boundary for
2704 * @error: must be negative errno
2705 *
2706 * Description:
2707 * Complete @rq till the next failure boundary. Must be called
2708 * with queue lock held.
2709 *
2710 * Return:
2711 * %false - we are done with this request
2712 * %true - still buffers pending for this request
2713 */
2714bool __blk_end_request_err(struct request *rq, int error)
2715{
2716 WARN_ON(error >= 0);
2717 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2718}
2719EXPORT_SYMBOL_GPL(__blk_end_request_err);
2720
Jens Axboe86db1e22008-01-29 14:53:40 +01002721void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2722 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002724 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002725 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726
David Woodhousefb2dce82008-08-05 18:01:53 +01002727 if (bio_has_data(bio)) {
2728 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002729 rq->buffer = bio_data(bio);
2730 }
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002731 rq->__data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
NeilBrown66846572007-08-16 13:31:28 +02002734 if (bio->bi_bdev)
2735 rq->rq_disk = bio->bi_bdev->bd_disk;
2736}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002738#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2739/**
2740 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2741 * @rq: the request to be flushed
2742 *
2743 * Description:
2744 * Flush all pages in @rq.
2745 */
2746void rq_flush_dcache_pages(struct request *rq)
2747{
2748 struct req_iterator iter;
2749 struct bio_vec *bvec;
2750
2751 rq_for_each_segment(bvec, rq, iter)
2752 flush_dcache_page(bvec->bv_page);
2753}
2754EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2755#endif
2756
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002757/**
2758 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2759 * @q : the queue of the device being checked
2760 *
2761 * Description:
2762 * Check if underlying low-level drivers of a device are busy.
2763 * If the drivers want to export their busy state, they must set own
2764 * exporting function using blk_queue_lld_busy() first.
2765 *
2766 * Basically, this function is used only by request stacking drivers
2767 * to stop dispatching requests to underlying devices when underlying
2768 * devices are busy. This behavior helps more I/O merging on the queue
2769 * of the request stacking driver and prevents I/O throughput regression
2770 * on burst I/O load.
2771 *
2772 * Return:
2773 * 0 - Not busy (The request stacking driver should dispatch request)
2774 * 1 - Busy (The request stacking driver should stop dispatching request)
2775 */
2776int blk_lld_busy(struct request_queue *q)
2777{
2778 if (q->lld_busy_fn)
2779 return q->lld_busy_fn(q);
2780
2781 return 0;
2782}
2783EXPORT_SYMBOL_GPL(blk_lld_busy);
2784
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002785/**
2786 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2787 * @rq: the clone request to be cleaned up
2788 *
2789 * Description:
2790 * Free all bios in @rq for a cloned request.
2791 */
2792void blk_rq_unprep_clone(struct request *rq)
2793{
2794 struct bio *bio;
2795
2796 while ((bio = rq->bio) != NULL) {
2797 rq->bio = bio->bi_next;
2798
2799 bio_put(bio);
2800 }
2801}
2802EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2803
2804/*
2805 * Copy attributes of the original request to the clone request.
2806 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2807 */
2808static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2809{
2810 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002811 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002812 dst->cmd_type = src->cmd_type;
2813 dst->__sector = blk_rq_pos(src);
2814 dst->__data_len = blk_rq_bytes(src);
2815 dst->nr_phys_segments = src->nr_phys_segments;
2816 dst->ioprio = src->ioprio;
2817 dst->extra_len = src->extra_len;
2818}
2819
2820/**
2821 * blk_rq_prep_clone - Helper function to setup clone request
2822 * @rq: the request to be setup
2823 * @rq_src: original request to be cloned
2824 * @bs: bio_set that bios for clone are allocated from
2825 * @gfp_mask: memory allocation mask for bio
2826 * @bio_ctr: setup function to be called for each clone bio.
2827 * Returns %0 for success, non %0 for failure.
2828 * @data: private data to be passed to @bio_ctr
2829 *
2830 * Description:
2831 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2832 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2833 * are not copied, and copying such parts is the caller's responsibility.
2834 * Also, pages which the original bios are pointing to are not copied
2835 * and the cloned bios just point same pages.
2836 * So cloned bios must be completed before original bios, which means
2837 * the caller must complete @rq before @rq_src.
2838 */
2839int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2840 struct bio_set *bs, gfp_t gfp_mask,
2841 int (*bio_ctr)(struct bio *, struct bio *, void *),
2842 void *data)
2843{
2844 struct bio *bio, *bio_src;
2845
2846 if (!bs)
2847 bs = fs_bio_set;
2848
2849 blk_rq_init(NULL, rq);
2850
2851 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002852 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002853 if (!bio)
2854 goto free_and_out;
2855
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002856 if (bio_ctr && bio_ctr(bio, bio_src, data))
2857 goto free_and_out;
2858
2859 if (rq->bio) {
2860 rq->biotail->bi_next = bio;
2861 rq->biotail = bio;
2862 } else
2863 rq->bio = rq->biotail = bio;
2864 }
2865
2866 __blk_rq_prep_clone(rq, rq_src);
2867
2868 return 0;
2869
2870free_and_out:
2871 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002872 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002873 blk_rq_unprep_clone(rq);
2874
2875 return -ENOMEM;
2876}
2877EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2878
Jens Axboe18887ad2008-07-28 13:08:45 +02002879int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880{
2881 return queue_work(kblockd_workqueue, work);
2882}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883EXPORT_SYMBOL(kblockd_schedule_work);
2884
Vivek Goyale43473b2010-09-15 17:06:35 -04002885int kblockd_schedule_delayed_work(struct request_queue *q,
2886 struct delayed_work *dwork, unsigned long delay)
2887{
2888 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2889}
2890EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2891
Jens Axboe73c10102011-03-08 13:19:51 +01002892#define PLUG_MAGIC 0x91827364
2893
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002894/**
2895 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2896 * @plug: The &struct blk_plug that needs to be initialized
2897 *
2898 * Description:
2899 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2900 * pending I/O should the task end up blocking between blk_start_plug() and
2901 * blk_finish_plug(). This is important from a performance perspective, but
2902 * also ensures that we don't deadlock. For instance, if the task is blocking
2903 * for a memory allocation, memory reclaim could end up wanting to free a
2904 * page belonging to that request that is currently residing in our private
2905 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2906 * this kind of deadlock.
2907 */
Jens Axboe73c10102011-03-08 13:19:51 +01002908void blk_start_plug(struct blk_plug *plug)
2909{
2910 struct task_struct *tsk = current;
2911
2912 plug->magic = PLUG_MAGIC;
2913 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01002914 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02002915 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002916
2917 /*
2918 * If this is a nested plug, don't actually assign it. It will be
2919 * flushed on its own.
2920 */
2921 if (!tsk->plug) {
2922 /*
2923 * Store ordering should not be needed here, since a potential
2924 * preempt will imply a full memory barrier
2925 */
2926 tsk->plug = plug;
2927 }
2928}
2929EXPORT_SYMBOL(blk_start_plug);
2930
2931static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2932{
2933 struct request *rqa = container_of(a, struct request, queuelist);
2934 struct request *rqb = container_of(b, struct request, queuelist);
2935
Jianpeng Ma975927b2012-10-25 21:58:17 +02002936 return !(rqa->q < rqb->q ||
2937 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01002938}
2939
Jens Axboe49cac012011-04-16 13:51:05 +02002940/*
2941 * If 'from_schedule' is true, then postpone the dispatch of requests
2942 * until a safe kblockd context. We due this to avoid accidental big
2943 * additional stack usage in driver dispatch, in places where the originally
2944 * plugger did not intend it.
2945 */
Jens Axboef6603782011-04-15 15:49:07 +02002946static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002947 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002948 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002949{
Jens Axboe49cac012011-04-16 13:51:05 +02002950 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002951
Bart Van Assche70460572012-11-28 13:45:56 +01002952 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002953 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002954 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002955 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002956 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002957}
2958
NeilBrown74018dc2012-07-31 09:08:15 +02002959static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002960{
2961 LIST_HEAD(callbacks);
2962
Shaohua Li2a7d5552012-07-31 09:08:15 +02002963 while (!list_empty(&plug->cb_list)) {
2964 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002965
Shaohua Li2a7d5552012-07-31 09:08:15 +02002966 while (!list_empty(&callbacks)) {
2967 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002968 struct blk_plug_cb,
2969 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002970 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02002971 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002972 }
NeilBrown048c9372011-04-18 09:52:22 +02002973 }
2974}
2975
NeilBrown9cbb1752012-07-31 09:08:14 +02002976struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2977 int size)
2978{
2979 struct blk_plug *plug = current->plug;
2980 struct blk_plug_cb *cb;
2981
2982 if (!plug)
2983 return NULL;
2984
2985 list_for_each_entry(cb, &plug->cb_list, list)
2986 if (cb->callback == unplug && cb->data == data)
2987 return cb;
2988
2989 /* Not currently on the callback list */
2990 BUG_ON(size < sizeof(*cb));
2991 cb = kzalloc(size, GFP_ATOMIC);
2992 if (cb) {
2993 cb->data = data;
2994 cb->callback = unplug;
2995 list_add(&cb->list, &plug->cb_list);
2996 }
2997 return cb;
2998}
2999EXPORT_SYMBOL(blk_check_plugged);
3000
Jens Axboe49cac012011-04-16 13:51:05 +02003001void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003002{
3003 struct request_queue *q;
3004 unsigned long flags;
3005 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003006 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003007 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003008
3009 BUG_ON(plug->magic != PLUG_MAGIC);
3010
NeilBrown74018dc2012-07-31 09:08:15 +02003011 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003012
3013 if (!list_empty(&plug->mq_list))
3014 blk_mq_flush_plug_list(plug, from_schedule);
3015
Jens Axboe73c10102011-03-08 13:19:51 +01003016 if (list_empty(&plug->list))
3017 return;
3018
NeilBrown109b8122011-04-11 14:13:10 +02003019 list_splice_init(&plug->list, &list);
3020
Jianpeng Ma422765c2013-01-11 14:46:09 +01003021 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003022
3023 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003024 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003025
3026 /*
3027 * Save and disable interrupts here, to avoid doing it for every
3028 * queue lock we have to take.
3029 */
Jens Axboe73c10102011-03-08 13:19:51 +01003030 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003031 while (!list_empty(&list)) {
3032 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003033 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003034 BUG_ON(!rq->q);
3035 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003036 /*
3037 * This drops the queue lock
3038 */
3039 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003040 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003041 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003042 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003043 spin_lock(q->queue_lock);
3044 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003045
3046 /*
3047 * Short-circuit if @q is dead
3048 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003049 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003050 __blk_end_request_all(rq, -ENODEV);
3051 continue;
3052 }
3053
Jens Axboe73c10102011-03-08 13:19:51 +01003054 /*
3055 * rq is already accounted, so use raw insert
3056 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003057 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3058 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3059 else
3060 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003061
3062 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003063 }
3064
Jens Axboe99e22592011-04-18 09:59:55 +02003065 /*
3066 * This drops the queue lock
3067 */
3068 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003069 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003070
Jens Axboe73c10102011-03-08 13:19:51 +01003071 local_irq_restore(flags);
3072}
Jens Axboe73c10102011-03-08 13:19:51 +01003073
3074void blk_finish_plug(struct blk_plug *plug)
3075{
Jens Axboef6603782011-04-15 15:49:07 +02003076 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003077
3078 if (plug == current->plug)
3079 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003080}
3081EXPORT_SYMBOL(blk_finish_plug);
3082
Lin Ming6c954662013-03-23 11:42:26 +08003083#ifdef CONFIG_PM_RUNTIME
3084/**
3085 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3086 * @q: the queue of the device
3087 * @dev: the device the queue belongs to
3088 *
3089 * Description:
3090 * Initialize runtime-PM-related fields for @q and start auto suspend for
3091 * @dev. Drivers that want to take advantage of request-based runtime PM
3092 * should call this function after @dev has been initialized, and its
3093 * request queue @q has been allocated, and runtime PM for it can not happen
3094 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3095 * cases, driver should call this function before any I/O has taken place.
3096 *
3097 * This function takes care of setting up using auto suspend for the device,
3098 * the autosuspend delay is set to -1 to make runtime suspend impossible
3099 * until an updated value is either set by user or by driver. Drivers do
3100 * not need to touch other autosuspend settings.
3101 *
3102 * The block layer runtime PM is request based, so only works for drivers
3103 * that use request as their IO unit instead of those directly use bio's.
3104 */
3105void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3106{
3107 q->dev = dev;
3108 q->rpm_status = RPM_ACTIVE;
3109 pm_runtime_set_autosuspend_delay(q->dev, -1);
3110 pm_runtime_use_autosuspend(q->dev);
3111}
3112EXPORT_SYMBOL(blk_pm_runtime_init);
3113
3114/**
3115 * blk_pre_runtime_suspend - Pre runtime suspend check
3116 * @q: the queue of the device
3117 *
3118 * Description:
3119 * This function will check if runtime suspend is allowed for the device
3120 * by examining if there are any requests pending in the queue. If there
3121 * are requests pending, the device can not be runtime suspended; otherwise,
3122 * the queue's status will be updated to SUSPENDING and the driver can
3123 * proceed to suspend the device.
3124 *
3125 * For the not allowed case, we mark last busy for the device so that
3126 * runtime PM core will try to autosuspend it some time later.
3127 *
3128 * This function should be called near the start of the device's
3129 * runtime_suspend callback.
3130 *
3131 * Return:
3132 * 0 - OK to runtime suspend the device
3133 * -EBUSY - Device should not be runtime suspended
3134 */
3135int blk_pre_runtime_suspend(struct request_queue *q)
3136{
3137 int ret = 0;
3138
3139 spin_lock_irq(q->queue_lock);
3140 if (q->nr_pending) {
3141 ret = -EBUSY;
3142 pm_runtime_mark_last_busy(q->dev);
3143 } else {
3144 q->rpm_status = RPM_SUSPENDING;
3145 }
3146 spin_unlock_irq(q->queue_lock);
3147 return ret;
3148}
3149EXPORT_SYMBOL(blk_pre_runtime_suspend);
3150
3151/**
3152 * blk_post_runtime_suspend - Post runtime suspend processing
3153 * @q: the queue of the device
3154 * @err: return value of the device's runtime_suspend function
3155 *
3156 * Description:
3157 * Update the queue's runtime status according to the return value of the
3158 * device's runtime suspend function and mark last busy for the device so
3159 * that PM core will try to auto suspend the device at a later time.
3160 *
3161 * This function should be called near the end of the device's
3162 * runtime_suspend callback.
3163 */
3164void blk_post_runtime_suspend(struct request_queue *q, int err)
3165{
3166 spin_lock_irq(q->queue_lock);
3167 if (!err) {
3168 q->rpm_status = RPM_SUSPENDED;
3169 } else {
3170 q->rpm_status = RPM_ACTIVE;
3171 pm_runtime_mark_last_busy(q->dev);
3172 }
3173 spin_unlock_irq(q->queue_lock);
3174}
3175EXPORT_SYMBOL(blk_post_runtime_suspend);
3176
3177/**
3178 * blk_pre_runtime_resume - Pre runtime resume processing
3179 * @q: the queue of the device
3180 *
3181 * Description:
3182 * Update the queue's runtime status to RESUMING in preparation for the
3183 * runtime resume of the device.
3184 *
3185 * This function should be called near the start of the device's
3186 * runtime_resume callback.
3187 */
3188void blk_pre_runtime_resume(struct request_queue *q)
3189{
3190 spin_lock_irq(q->queue_lock);
3191 q->rpm_status = RPM_RESUMING;
3192 spin_unlock_irq(q->queue_lock);
3193}
3194EXPORT_SYMBOL(blk_pre_runtime_resume);
3195
3196/**
3197 * blk_post_runtime_resume - Post runtime resume processing
3198 * @q: the queue of the device
3199 * @err: return value of the device's runtime_resume function
3200 *
3201 * Description:
3202 * Update the queue's runtime status according to the return value of the
3203 * device's runtime_resume function. If it is successfully resumed, process
3204 * the requests that are queued into the device's queue when it is resuming
3205 * and then mark last busy and initiate autosuspend for it.
3206 *
3207 * This function should be called near the end of the device's
3208 * runtime_resume callback.
3209 */
3210void blk_post_runtime_resume(struct request_queue *q, int err)
3211{
3212 spin_lock_irq(q->queue_lock);
3213 if (!err) {
3214 q->rpm_status = RPM_ACTIVE;
3215 __blk_run_queue(q);
3216 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003217 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003218 } else {
3219 q->rpm_status = RPM_SUSPENDED;
3220 }
3221 spin_unlock_irq(q->queue_lock);
3222}
3223EXPORT_SYMBOL(blk_post_runtime_resume);
3224#endif
3225
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226int __init blk_dev_init(void)
3227{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003228 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3229 sizeof(((struct request *)0)->cmd_flags));
3230
Tejun Heo89b90be2011-01-03 15:01:47 +01003231 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3232 kblockd_workqueue = alloc_workqueue("kblockd",
Viresh Kumar695588f2013-04-24 17:12:56 +05303233 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3234 WQ_POWER_EFFICIENT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 if (!kblockd_workqueue)
3236 panic("Failed to create kblockd\n");
3237
3238 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003239 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240
Jens Axboe8324aa92008-01-29 14:51:59 +01003241 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003242 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 return 0;
3245}