blob: 30f6153a40c27c7154dd453301807c7d39d1451c [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"
Ming Lei43a5e4e2013-12-26 21:31:35 +080041#include "blk-mq.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010042
Mike Snitzerd07335e2010-11-16 12:52:38 +010043EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020044EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070045EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060046EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010047EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010048
Tejun Heoa73f7302011-12-14 00:33:37 +010049DEFINE_IDA(blk_queue_ida);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * For the allocated request tables
53 */
Jens Axboe320ae512013-10-24 09:20:05 +010054struct kmem_cache *request_cachep = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
57 * For queue allocation
58 */
Jens Axboe6728cb02008-01-31 13:03:55 +010059struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * Controlling structure to kblockd
63 */
Jens Axboeff856ba2006-01-09 16:02:34 +010064static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Jens Axboe8324aa92008-01-29 14:51:59 +010066void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067{
68 int nr;
69
70 nr = q->nr_requests - (q->nr_requests / 8) + 1;
71 if (nr > q->nr_requests)
72 nr = q->nr_requests;
73 q->nr_congestion_on = nr;
74
75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
76 if (nr < 1)
77 nr = 1;
78 q->nr_congestion_off = nr;
79}
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/**
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
83 * @bdev: device
84 *
85 * Locates the passed device's request queue and returns the address of its
Tejun Heoff9ea322014-09-08 08:03:56 +090086 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
89struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
90{
Jens Axboe165125e2007-07-24 09:28:11 +020091 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Tejun Heoff9ea322014-09-08 08:03:56 +090093 return &q->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
Linus Torvalds1da177e2005-04-16 15:20:36 -070095EXPORT_SYMBOL(blk_get_backing_dev_info);
96
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +020097void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +020099 memset(rq, 0, sizeof(*rq));
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700102 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200103 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100104 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900105 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200106 INIT_HLIST_NODE(&rq->hash);
107 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200108 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800109 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100110 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900111 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700112 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100113 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200115EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
NeilBrown5bb23a62007-09-27 12:46:13 +0200117static void req_bio_endio(struct request *rq, struct bio *bio,
118 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100119{
Tejun Heo143a87f2011-01-25 12:43:52 +0100120 if (error)
121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
123 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100124
Tejun Heo143a87f2011-01-25 12:43:52 +0100125 if (unlikely(rq->cmd_flags & REQ_QUIET))
126 set_bit(BIO_QUIET, &bio->bi_flags);
127
Kent Overstreetf79ea412012-09-20 16:38:30 -0700128 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100129
130 /* don't actually finish bio if it's part of flush sequence */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700131 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Tejun Heo143a87f2011-01-25 12:43:52 +0100132 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100133}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135void blk_dump_rq_flags(struct request *rq, char *msg)
136{
137 int bit;
138
Jens Axboe59533162013-05-23 12:25:08 +0200139 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200140 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200141 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Tejun Heo83096eb2009-05-07 22:24:39 +0900143 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
144 (unsigned long long)blk_rq_pos(rq),
145 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600146 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
147 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200149 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100150 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200151 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 printk("%02x ", rq->cmd[bit]);
153 printk("\n");
154 }
155}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156EXPORT_SYMBOL(blk_dump_rq_flags);
157
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500158static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200159{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500160 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200161
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500162 q = container_of(work, struct request_queue, delay_work.work);
163 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200164 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500165 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500169 * blk_delay_queue - restart queueing after defined interval
170 * @q: The &struct request_queue in question
171 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 *
173 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500174 * Sometimes queueing needs to be postponed for a little while, to allow
175 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100176 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500177 */
178void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Bart Van Assche70460572012-11-28 13:45:56 +0100180 if (likely(!blk_queue_dead(q)))
181 queue_delayed_work(kblockd_workqueue, &q->delay_work,
182 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500184EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/**
187 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200188 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 *
190 * Description:
191 * blk_start_queue() will clear the stop flag on the queue, and call
192 * the request_fn for the queue if it was in a stopped state when
193 * entered. Also see blk_stop_queue(). Queue lock must be held.
194 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200195void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200197 WARN_ON(!irqs_disabled());
198
Nick Piggin75ad23b2008-04-29 14:48:33 +0200199 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200200 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_start_queue);
203
204/**
205 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200206 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 *
208 * Description:
209 * The Linux block layer assumes that a block driver will consume all
210 * entries on the request queue when the request_fn strategy is called.
211 * Often this will not happen, because of hardware limitations (queue
212 * depth settings). If a device driver gets a 'queue full' response,
213 * or if it simply chooses not to queue more I/O at one point, it can
214 * call this function to prevent the request_fn from being called until
215 * the driver has signalled it's ready to go again. This happens by calling
216 * blk_start_queue() to restart queue operations. Queue lock must be held.
217 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200218void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
Tejun Heo136b5722012-08-21 13:18:24 -0700220 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200221 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223EXPORT_SYMBOL(blk_stop_queue);
224
225/**
226 * blk_sync_queue - cancel any pending callbacks on a queue
227 * @q: the queue
228 *
229 * Description:
230 * The block layer may perform asynchronous callback activity
231 * on a queue, such as calling the unplug function after a timeout.
232 * A block device may call blk_sync_queue to ensure that any
233 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200234 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * that its ->make_request_fn will not re-add plugging prior to calling
236 * this function.
237 *
Vivek Goyalda527772011-03-02 19:05:33 -0500238 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900239 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800240 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500241 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
243void blk_sync_queue(struct request_queue *q)
244{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100245 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800246
247 if (q->mq_ops) {
248 struct blk_mq_hw_ctx *hctx;
249 int i;
250
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600251 queue_for_each_hw_ctx(q, hctx, i) {
252 cancel_delayed_work_sync(&hctx->run_work);
253 cancel_delayed_work_sync(&hctx->delay_work);
254 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800255 } else {
256 cancel_delayed_work_sync(&q->delay_work);
257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259EXPORT_SYMBOL(blk_sync_queue);
260
261/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100262 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
263 * @q: The queue to run
264 *
265 * Description:
266 * Invoke request handling on a queue if there are any pending requests.
267 * May be used to restart request handling after a request has completed.
268 * This variant runs the queue whether or not the queue has been
269 * stopped. Must be called with the queue lock held and interrupts
270 * disabled. See also @blk_run_queue.
271 */
272inline void __blk_run_queue_uncond(struct request_queue *q)
273{
274 if (unlikely(blk_queue_dead(q)))
275 return;
276
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100277 /*
278 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
279 * the queue lock internally. As a result multiple threads may be
280 * running such a request function concurrently. Keep track of the
281 * number of active request_fn invocations such that blk_drain_queue()
282 * can wait until all these request_fn calls have finished.
283 */
284 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100285 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100286 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100287}
288
289/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200290 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200292 *
293 * Description:
294 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200295 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200297void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Tejun Heoa538cd02009-04-23 11:05:17 +0900299 if (unlikely(blk_queue_stopped(q)))
300 return;
301
Bart Van Asschec246e802012-12-06 14:32:01 +0100302 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200303}
304EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200305
Nick Piggin75ad23b2008-04-29 14:48:33 +0200306/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200307 * blk_run_queue_async - run a single device queue in workqueue context
308 * @q: The queue to run
309 *
310 * Description:
311 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100312 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200313 */
314void blk_run_queue_async(struct request_queue *q)
315{
Bart Van Assche70460572012-11-28 13:45:56 +0100316 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700317 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200318}
Jens Axboec21e6be2011-04-19 13:32:46 +0200319EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200320
321/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200322 * blk_run_queue - run a single device queue
323 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200324 *
325 * Description:
326 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900327 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328 */
329void blk_run_queue(struct request_queue *q)
330{
331 unsigned long flags;
332
333 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200334 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 spin_unlock_irqrestore(q->queue_lock, flags);
336}
337EXPORT_SYMBOL(blk_run_queue);
338
Jens Axboe165125e2007-07-24 09:28:11 +0200339void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500340{
341 kobject_put(&q->kobj);
342}
Jens Axboed86e0e82011-05-27 07:44:43 +0200343EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500344
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200345/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100346 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200347 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200348 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200349 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200350 * Drain requests from @q. If @drain_all is set, all requests are drained.
351 * If not, only ELVPRIV requests are drained. The caller is responsible
352 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100354static void __blk_drain_queue(struct request_queue *q, bool drain_all)
355 __releases(q->queue_lock)
356 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200357{
Asias He458f27a2012-06-15 08:45:25 +0200358 int i;
359
Bart Van Assche807592a2012-11-28 13:43:38 +0100360 lockdep_assert_held(q->queue_lock);
361
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200362 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100363 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200364
Tejun Heob855b042012-03-06 21:24:55 +0100365 /*
366 * The caller might be trying to drain @q before its
367 * elevator is initialized.
368 */
369 if (q->elevator)
370 elv_drain_elevator(q);
371
Tejun Heo5efd6112012-03-05 13:15:12 -0800372 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200373
Tejun Heo4eabc942011-12-15 20:03:04 +0100374 /*
375 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100376 * driver init after queue creation or is not yet fully
377 * active yet. Some drivers (e.g. fd and loop) get unhappy
378 * in such cases. Kick queue iff dispatch queue has
379 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100380 */
Tejun Heob855b042012-03-06 21:24:55 +0100381 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100382 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200383
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700384 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100385 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100386
387 /*
388 * Unfortunately, requests are queued at and tracked from
389 * multiple places and there's no single counter which can
390 * be drained. Check all the queues and counters.
391 */
392 if (drain_all) {
Ming Leie97c2932014-09-25 23:23:46 +0800393 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
Tejun Heo481a7d62011-12-14 00:33:37 +0100394 drain |= !list_empty(&q->queue_head);
395 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700396 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100397 drain |= q->in_flight[i];
Ming Lei7c94e1c2014-09-25 23:23:43 +0800398 if (fq)
399 drain |= !list_empty(&fq->flush_queue[i]);
Tejun Heo481a7d62011-12-14 00:33:37 +0100400 }
401 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200402
Tejun Heo481a7d62011-12-14 00:33:37 +0100403 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100405
406 spin_unlock_irq(q->queue_lock);
407
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200408 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100409
410 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200411 }
Asias He458f27a2012-06-15 08:45:25 +0200412
413 /*
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
417 */
418 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700419 struct request_list *rl;
420
Tejun Heoa0516612012-06-26 15:05:44 -0700421 blk_queue_for_each_rl(rl, q)
422 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
423 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200424 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200425}
426
Tejun Heoc9a929d2011-10-19 14:42:16 +0200427/**
Tejun Heod7325802012-03-05 13:14:58 -0800428 * blk_queue_bypass_start - enter queue bypass mode
429 * @q: queue of interest
430 *
431 * In bypass mode, only the dispatch FIFO queue of @q is used. This
432 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800433 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700434 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
435 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800436 */
437void blk_queue_bypass_start(struct request_queue *q)
438{
439 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600440 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800441 queue_flag_set(QUEUE_FLAG_BYPASS, q);
442 spin_unlock_irq(q->queue_lock);
443
Tejun Heo776687b2014-07-01 10:29:17 -0600444 /*
445 * Queues start drained. Skip actual draining till init is
446 * complete. This avoids lenghty delays during queue init which
447 * can happen many times during boot.
448 */
449 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100450 spin_lock_irq(q->queue_lock);
451 __blk_drain_queue(q, false);
452 spin_unlock_irq(q->queue_lock);
453
Tejun Heob82d4b12012-04-13 13:11:31 -0700454 /* ensure blk_queue_bypass() is %true inside RCU read lock */
455 synchronize_rcu();
456 }
Tejun Heod7325802012-03-05 13:14:58 -0800457}
458EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
459
460/**
461 * blk_queue_bypass_end - leave queue bypass mode
462 * @q: queue of interest
463 *
464 * Leave bypass mode and restore the normal queueing behavior.
465 */
466void blk_queue_bypass_end(struct request_queue *q)
467{
468 spin_lock_irq(q->queue_lock);
469 if (!--q->bypass_depth)
470 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
471 WARN_ON_ONCE(q->bypass_depth < 0);
472 spin_unlock_irq(q->queue_lock);
473}
474EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
475
476/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200477 * blk_cleanup_queue - shutdown a request queue
478 * @q: request queue to shutdown
479 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100480 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
481 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500482 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100483void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500484{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200485 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700486
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100487 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500488 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100489 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200490 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800491
Tejun Heo80fd9972012-04-13 14:50:53 -0700492 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100493 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700494 * that, unlike blk_queue_bypass_start(), we aren't performing
495 * synchronize_rcu() after entering bypass mode to avoid the delay
496 * as some drivers create and destroy a lot of queues while
497 * probing. This is still safe because blk_release_queue() will be
498 * called only after the queue refcnt drops to zero and nothing,
499 * RCU or not, would be traversing the queue by then.
500 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800501 q->bypass_depth++;
502 queue_flag_set(QUEUE_FLAG_BYPASS, q);
503
Tejun Heoc9a929d2011-10-19 14:42:16 +0200504 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
505 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100506 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200507 spin_unlock_irq(lock);
508 mutex_unlock(&q->sysfs_lock);
509
Bart Van Asschec246e802012-12-06 14:32:01 +0100510 /*
511 * Drain all requests queued before DYING marking. Set DEAD flag to
512 * prevent that q->request_fn() gets invoked after draining finished.
513 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800514 if (q->mq_ops) {
Tejun Heo780db202014-07-01 10:31:13 -0600515 blk_mq_freeze_queue(q);
Ming Lei43a5e4e2013-12-26 21:31:35 +0800516 spin_lock_irq(lock);
517 } else {
518 spin_lock_irq(lock);
519 __blk_drain_queue(q, true);
520 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100521 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100522 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200523
524 /* @q won't process any more request, flush async actions */
525 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
526 blk_sync_queue(q);
527
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100528 if (q->mq_ops)
529 blk_mq_free_queue(q);
530
Asias He5e5cfac2012-05-24 23:28:52 +0800531 spin_lock_irq(lock);
532 if (q->queue_lock != &q->__queue_lock)
533 q->queue_lock = &q->__queue_lock;
534 spin_unlock_irq(lock);
535
Tejun Heoc9a929d2011-10-19 14:42:16 +0200536 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500537 blk_put_queue(q);
538}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539EXPORT_SYMBOL(blk_cleanup_queue);
540
Tejun Heo5b788ce2012-06-04 20:40:59 -0700541int blk_init_rl(struct request_list *rl, struct request_queue *q,
542 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400544 if (unlikely(rl->rq_pool))
545 return 0;
546
Tejun Heo5b788ce2012-06-04 20:40:59 -0700547 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200548 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
549 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200550 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
551 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Christoph Lameter19460892005-06-23 00:08:19 -0700553 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700554 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700555 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 if (!rl->rq_pool)
557 return -ENOMEM;
558
559 return 0;
560}
561
Tejun Heo5b788ce2012-06-04 20:40:59 -0700562void blk_exit_rl(struct request_list *rl)
563{
564 if (rl->rq_pool)
565 mempool_destroy(rl->rq_pool);
566}
567
Jens Axboe165125e2007-07-24 09:28:11 +0200568struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100570 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700571}
572EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Jens Axboe165125e2007-07-24 09:28:11 +0200574struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700575{
Jens Axboe165125e2007-07-24 09:28:11 +0200576 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700577 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700578
Jens Axboe8324aa92008-01-29 14:51:59 +0100579 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700580 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (!q)
582 return NULL;
583
Dan Carpenter00380a42012-03-23 09:58:54 +0100584 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100585 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800586 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100587
Jens Axboe0989a022009-06-12 14:42:56 +0200588 q->backing_dev_info.ra_pages =
589 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
590 q->backing_dev_info.state = 0;
591 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200592 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100593 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200594
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700595 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100596 if (err)
597 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700598
Matthew Garrett31373d02010-04-06 14:25:14 +0200599 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
600 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700601 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100602 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700603 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100604 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800605#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800606 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800607#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500608 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500609
Jens Axboe8324aa92008-01-29 14:51:59 +0100610 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Al Viro483f4af2006-03-18 18:34:37 -0500612 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700613 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500614
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500615 /*
616 * By default initialize queue_lock to internal lock and driver can
617 * override it later if need be.
618 */
619 q->queue_lock = &q->__queue_lock;
620
Tejun Heob82d4b12012-04-13 13:11:31 -0700621 /*
622 * A queue starts its life with bypass turned on to avoid
623 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700624 * init. The initial bypass will be finished when the queue is
625 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700626 */
627 q->bypass_depth = 1;
628 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
629
Jens Axboe320ae512013-10-24 09:20:05 +0100630 init_waitqueue_head(&q->mq_freeze_wq);
631
Tejun Heo5efd6112012-03-05 13:15:12 -0800632 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400633 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100636
Mikulas Patockafff49962013-10-14 12:11:36 -0400637fail_bdi:
638 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100639fail_id:
640 ida_simple_remove(&blk_queue_ida, q->id);
641fail_q:
642 kmem_cache_free(blk_requestq_cachep, q);
643 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
Christoph Lameter19460892005-06-23 00:08:19 -0700645EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647/**
648 * blk_init_queue - prepare a request queue for use with a block device
649 * @rfn: The function to be called to process requests that have been
650 * placed on the queue.
651 * @lock: Request queue spin lock
652 *
653 * Description:
654 * If a block device wishes to use the standard request handling procedures,
655 * which sorts requests and coalesces adjacent requests, then it must
656 * call blk_init_queue(). The function @rfn will be called when there
657 * are requests on the queue that need to be processed. If the device
658 * supports plugging, then @rfn may not be called immediately when requests
659 * are available on the queue, but may be called at some time later instead.
660 * Plugged queues are generally unplugged when a buffer belonging to one
661 * of the requests on the queue is needed, or due to memory pressure.
662 *
663 * @rfn is not required, or even expected, to remove all requests off the
664 * queue, but only as many as it can handle at a time. If it does leave
665 * requests on the queue, it is responsible for arranging that the requests
666 * get dealt with eventually.
667 *
668 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200669 * request queue; this lock will be taken also from interrupt context, so irq
670 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200672 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 * it didn't succeed.
674 *
675 * Note:
676 * blk_init_queue() must be paired with a blk_cleanup_queue() call
677 * when the block device is deactivated (such as at module unload).
678 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700679
Jens Axboe165125e2007-07-24 09:28:11 +0200680struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100682 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700683}
684EXPORT_SYMBOL(blk_init_queue);
685
Jens Axboe165125e2007-07-24 09:28:11 +0200686struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700687blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
688{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600689 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600691 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
692 if (!uninit_q)
693 return NULL;
694
Mike Snitzer51514122011-11-23 10:59:13 +0100695 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600696 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700697 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700698
Mike Snitzer7982e902014-03-08 17:20:01 -0700699 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200700}
701EXPORT_SYMBOL(blk_init_queue_node);
702
703struct request_queue *
704blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
705 spinlock_t *lock)
706{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 if (!q)
708 return NULL;
709
Ming Leif70ced02014-09-25 23:23:47 +0800710 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800711 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700712 return NULL;
713
Tejun Heoa0516612012-06-26 15:05:44 -0700714 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600715 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
717 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900719 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700720 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500721
722 /* Override internal queue lock with supplied lock pointer */
723 if (lock)
724 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Jens Axboef3b144a2009-03-06 08:48:33 +0100726 /*
727 * This also sets hw/phys segments, boundary and size
728 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200729 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Alan Stern44ec9542007-02-20 11:01:57 -0500731 q->sg_reserved_size = INT_MAX;
732
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600733 /* Protect q->elevator from elevator_change */
734 mutex_lock(&q->sysfs_lock);
735
Tejun Heob82d4b12012-04-13 13:11:31 -0700736 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600737 if (elevator_init(q, NULL)) {
738 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600739 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600740 }
741
742 mutex_unlock(&q->sysfs_lock);
743
Tejun Heob82d4b12012-04-13 13:11:31 -0700744 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600745
746fail:
Ming Leiba483382014-09-25 23:23:44 +0800747 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600748 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
Mike Snitzer51514122011-11-23 10:59:13 +0100750EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Tejun Heo09ac46c2011-12-14 00:33:38 +0100752bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100754 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100755 __blk_get_queue(q);
756 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
758
Tejun Heo09ac46c2011-12-14 00:33:38 +0100759 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
Jens Axboed86e0e82011-05-27 07:44:43 +0200761EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Tejun Heo5b788ce2012-06-04 20:40:59 -0700763static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100765 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700766 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100767 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100768 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100769 }
770
Tejun Heo5b788ce2012-06-04 20:40:59 -0700771 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774/*
775 * ioc_batching returns true if the ioc is a valid batching request and
776 * should be given priority access to a request.
777 */
Jens Axboe165125e2007-07-24 09:28:11 +0200778static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
780 if (!ioc)
781 return 0;
782
783 /*
784 * Make sure the process is able to allocate at least 1 request
785 * even if the batch times out, otherwise we could theoretically
786 * lose wakeups.
787 */
788 return ioc->nr_batch_requests == q->nr_batching ||
789 (ioc->nr_batch_requests > 0
790 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
791}
792
793/*
794 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
795 * will cause the process to be a "batcher" on all queues in the system. This
796 * is the behaviour we want though - once it gets a wakeup it should be given
797 * a nice run.
798 */
Jens Axboe165125e2007-07-24 09:28:11 +0200799static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
801 if (!ioc || ioc_batching(q, ioc))
802 return;
803
804 ioc->nr_batch_requests = q->nr_batching;
805 ioc->last_waited = jiffies;
806}
807
Tejun Heo5b788ce2012-06-04 20:40:59 -0700808static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700810 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Tejun Heoa0516612012-06-26 15:05:44 -0700812 /*
813 * bdi isn't aware of blkcg yet. As all async IOs end up root
814 * blkcg anyway, just use root blkcg state.
815 */
816 if (rl == &q->root_rl &&
817 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200818 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Jens Axboe1faa16d2009-04-06 14:48:01 +0200820 if (rl->count[sync] + 1 <= q->nr_requests) {
821 if (waitqueue_active(&rl->wait[sync]))
822 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Tejun Heo5b788ce2012-06-04 20:40:59 -0700824 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 }
826}
827
828/*
829 * A request has just been released. Account for it, update the full and
830 * congestion status, wake up any waiters. Called under q->queue_lock.
831 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700832static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700834 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200835 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700837 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200838 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200839 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700840 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Tejun Heo5b788ce2012-06-04 20:40:59 -0700842 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Jens Axboe1faa16d2009-04-06 14:48:01 +0200844 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700845 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
847
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600848int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
849{
850 struct request_list *rl;
851
852 spin_lock_irq(q->queue_lock);
853 q->nr_requests = nr;
854 blk_queue_congestion_threshold(q);
855
856 /* congestion isn't cgroup aware and follows root blkcg for now */
857 rl = &q->root_rl;
858
859 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
860 blk_set_queue_congested(q, BLK_RW_SYNC);
861 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
862 blk_clear_queue_congested(q, BLK_RW_SYNC);
863
864 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
865 blk_set_queue_congested(q, BLK_RW_ASYNC);
866 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
867 blk_clear_queue_congested(q, BLK_RW_ASYNC);
868
869 blk_queue_for_each_rl(rl, q) {
870 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
871 blk_set_rl_full(rl, BLK_RW_SYNC);
872 } else {
873 blk_clear_rl_full(rl, BLK_RW_SYNC);
874 wake_up(&rl->wait[BLK_RW_SYNC]);
875 }
876
877 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
878 blk_set_rl_full(rl, BLK_RW_ASYNC);
879 } else {
880 blk_clear_rl_full(rl, BLK_RW_ASYNC);
881 wake_up(&rl->wait[BLK_RW_ASYNC]);
882 }
883 }
884
885 spin_unlock_irq(q->queue_lock);
886 return 0;
887}
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100890 * Determine if elevator data should be initialized when allocating the
891 * request associated with @bio.
892 */
893static bool blk_rq_should_init_elevator(struct bio *bio)
894{
895 if (!bio)
896 return true;
897
898 /*
899 * Flush requests do not use the elevator so skip initialization.
900 * This allows a request to share the flush and elevator data.
901 */
902 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
903 return false;
904
905 return true;
906}
907
Tejun Heoda8303c2011-10-19 14:33:05 +0200908/**
Tejun Heo852c7882012-03-05 13:15:27 -0800909 * rq_ioc - determine io_context for request allocation
910 * @bio: request being allocated is for this bio (can be %NULL)
911 *
912 * Determine io_context to use for request allocation for @bio. May return
913 * %NULL if %current->io_context doesn't exist.
914 */
915static struct io_context *rq_ioc(struct bio *bio)
916{
917#ifdef CONFIG_BLK_CGROUP
918 if (bio && bio->bi_ioc)
919 return bio->bi_ioc;
920#endif
921 return current->io_context;
922}
923
924/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700925 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700926 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200927 * @rw_flags: RW and SYNC flags
928 * @bio: bio to allocate request for (can be %NULL)
929 * @gfp_mask: allocation mask
930 *
931 * Get a free request from @q. This function may fail under memory
932 * pressure or if @q is dead.
933 *
Masanari Iidada3dae52014-09-09 01:27:23 +0900934 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -0600935 * Returns ERR_PTR on failure, with @q->queue_lock held.
936 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700938static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700939 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700941 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800942 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700943 struct elevator_type *et = q->elevator->type;
944 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100945 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200946 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200947 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100949 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -0600950 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +0200951
Jens Axboe7749a8d2006-12-13 13:02:26 +0100952 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100953 if (may_queue == ELV_MQUEUE_NO)
954 goto rq_starved;
955
Jens Axboe1faa16d2009-04-06 14:48:01 +0200956 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
957 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100958 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100959 * The queue will fill after this allocation, so set
960 * it as full, and mark this process as "batching".
961 * This process will be allowed to complete a batch of
962 * requests, others will be blocked.
963 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700964 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100965 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700966 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100967 } else {
968 if (may_queue != ELV_MQUEUE_MUST
969 && !ioc_batching(q, ioc)) {
970 /*
971 * The queue is full and the allocating
972 * process is not a "batcher", and not
973 * exempted by the IO scheduler
974 */
Joe Lawrencea492f072014-08-28 08:15:21 -0600975 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100976 }
977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 }
Tejun Heoa0516612012-06-26 15:05:44 -0700979 /*
980 * bdi isn't aware of blkcg yet. As all async IOs end up
981 * root blkcg anyway, just use root blkcg state.
982 */
983 if (rl == &q->root_rl)
984 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 }
986
Jens Axboe082cf692005-06-28 16:35:11 +0200987 /*
988 * Only allow batching queuers to allocate up to 50% over the defined
989 * limit of requests, otherwise we could have thousands of requests
990 * allocated with any setting of ->nr_requests
991 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200992 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -0600993 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100994
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700995 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200996 rl->count[is_sync]++;
997 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200998
Tejun Heof1f8cc92011-12-14 00:33:42 +0100999 /*
1000 * Decide whether the new request will be managed by elevator. If
1001 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1002 * prevent the current elevator from being destroyed until the new
1003 * request is freed. This guarantees icq's won't be destroyed and
1004 * makes creating new ones safe.
1005 *
1006 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1007 * it will be created after releasing queue_lock.
1008 */
Tejun Heod7325802012-03-05 13:14:58 -08001009 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001010 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001011 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001012 if (et->icq_cache && ioc)
1013 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001014 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001015
Jens Axboef253b862010-10-24 22:06:02 +02001016 if (blk_queue_io_stat(q))
1017 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 spin_unlock_irq(q->queue_lock);
1019
Tejun Heo29e2b092012-04-19 16:29:21 -07001020 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001021 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001022 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001023 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Tejun Heo29e2b092012-04-19 16:29:21 -07001025 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001026 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001027 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1028
Tejun Heoaaf7c682012-04-19 16:29:22 -07001029 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001030 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001031 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001032 if (ioc)
1033 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001034 if (!icq)
1035 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001036 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001037
1038 rq->elv.icq = icq;
1039 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1040 goto fail_elvpriv;
1041
1042 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001043 if (icq)
1044 get_io_context(icq->ioc);
1045 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001046out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001047 /*
1048 * ioc may be NULL here, and ioc_batching will be false. That's
1049 * OK, if the queue is under the request limit then requests need
1050 * not count toward the nr_batch_requests limit. There will always
1051 * be some limit enforced by BLK_BATCH_TIME.
1052 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 if (ioc_batching(q, ioc))
1054 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001055
Jens Axboe1faa16d2009-04-06 14:48:01 +02001056 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001058
Tejun Heoaaf7c682012-04-19 16:29:22 -07001059fail_elvpriv:
1060 /*
1061 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1062 * and may fail indefinitely under memory pressure and thus
1063 * shouldn't stall IO. Treat this request as !elvpriv. This will
1064 * disturb iosched and blkcg but weird is bettern than dead.
1065 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001066 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1067 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001068
1069 rq->cmd_flags &= ~REQ_ELVPRIV;
1070 rq->elv.icq = NULL;
1071
1072 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001073 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001074 spin_unlock_irq(q->queue_lock);
1075 goto out;
1076
Tejun Heob6792812012-03-05 13:15:23 -08001077fail_alloc:
1078 /*
1079 * Allocation failed presumably due to memory. Undo anything we
1080 * might have messed up.
1081 *
1082 * Allocating task should really be put onto the front of the wait
1083 * queue, but this is pretty rare.
1084 */
1085 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001086 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001087
1088 /*
1089 * in the very unlikely event that allocation failed and no
1090 * requests for this direction was pending, mark us starved so that
1091 * freeing of a request in the other direction will notice
1092 * us. another possible fix would be to split the rq mempool into
1093 * READ and WRITE
1094 */
1095rq_starved:
1096 if (unlikely(rl->count[is_sync] == 0))
1097 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001098 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099}
1100
Tejun Heoda8303c2011-10-19 14:33:05 +02001101/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001102 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001103 * @q: request_queue to allocate request from
1104 * @rw_flags: RW and SYNC flags
1105 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001106 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001107 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001108 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1109 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001110 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001111 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001112 * Returns ERR_PTR on failure, with @q->queue_lock held.
1113 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001115static struct request *get_request(struct request_queue *q, int rw_flags,
1116 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001118 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001119 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001120 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001122
1123 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001124retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001125 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001126 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001127 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001129 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001130 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001131 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Tejun Heoa06e05e2012-06-04 20:40:55 -07001134 /* wait on @rl and retry */
1135 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1136 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001137
Tejun Heoa06e05e2012-06-04 20:40:55 -07001138 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Tejun Heoa06e05e2012-06-04 20:40:55 -07001140 spin_unlock_irq(q->queue_lock);
1141 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Tejun Heoa06e05e2012-06-04 20:40:55 -07001143 /*
1144 * After sleeping, we become a "batching" process and will be able
1145 * to allocate at least one request, and up to a big batch of them
1146 * for a small period time. See ioc_batching, ioc_set_batching
1147 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001148 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Tejun Heoa06e05e2012-06-04 20:40:55 -07001150 spin_lock_irq(q->queue_lock);
1151 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001152
Tejun Heoa06e05e2012-06-04 20:40:55 -07001153 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154}
1155
Jens Axboe320ae512013-10-24 09:20:05 +01001156static struct request *blk_old_get_request(struct request_queue *q, int rw,
1157 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
1159 struct request *rq;
1160
1161 BUG_ON(rw != READ && rw != WRITE);
1162
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001163 /* create ioc upfront */
1164 create_io_context(gfp_mask, q->node);
1165
Nick Piggind6344532005-06-28 20:45:14 -07001166 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001167 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001168 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001169 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001170 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 return rq;
1173}
Jens Axboe320ae512013-10-24 09:20:05 +01001174
1175struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1176{
1177 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001178 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001179 else
1180 return blk_old_get_request(q, rw, gfp_mask);
1181}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182EXPORT_SYMBOL(blk_get_request);
1183
1184/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001185 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001186 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001187 * @bio: The bio describing the memory mappings that will be submitted for IO.
1188 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001189 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001190 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001191 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1192 * type commands. Where the struct request needs to be farther initialized by
1193 * the caller. It is passed a &struct bio, which describes the memory info of
1194 * the I/O transfer.
1195 *
1196 * The caller of blk_make_request must make sure that bi_io_vec
1197 * are set to describe the memory buffers. That bio_data_dir() will return
1198 * the needed direction of the request. (And all bio's in the passed bio-chain
1199 * are properly set accordingly)
1200 *
1201 * If called under none-sleepable conditions, mapped bio buffers must not
1202 * need bouncing, by calling the appropriate masked or flagged allocator,
1203 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1204 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001205 *
1206 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1207 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1208 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1209 * completion of a bio that hasn't been submitted yet, thus resulting in a
1210 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1211 * of bio_alloc(), as that avoids the mempool deadlock.
1212 * If possible a big IO should be split into smaller parts when allocation
1213 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001214 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001215struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1216 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001217{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001218 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1219
Joe Lawrencea492f072014-08-28 08:15:21 -06001220 if (IS_ERR(rq))
1221 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001222
Jens Axboef27b0872014-06-06 07:57:37 -06001223 blk_rq_set_block_pc(rq);
1224
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001225 for_each_bio(bio) {
1226 struct bio *bounce_bio = bio;
1227 int ret;
1228
1229 blk_queue_bounce(q, &bounce_bio);
1230 ret = blk_rq_append_bio(q, rq, bounce_bio);
1231 if (unlikely(ret)) {
1232 blk_put_request(rq);
1233 return ERR_PTR(ret);
1234 }
1235 }
1236
1237 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001238}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001239EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001240
1241/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001242 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001243 * @rq: request to be initialized
1244 *
1245 */
1246void blk_rq_set_block_pc(struct request *rq)
1247{
1248 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1249 rq->__data_len = 0;
1250 rq->__sector = (sector_t) -1;
1251 rq->bio = rq->biotail = NULL;
1252 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001253}
1254EXPORT_SYMBOL(blk_rq_set_block_pc);
1255
1256/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 * blk_requeue_request - put a request back on queue
1258 * @q: request queue where request should be inserted
1259 * @rq: request to be inserted
1260 *
1261 * Description:
1262 * Drivers often keep queueing requests until the hardware cannot accept
1263 * more, when that condition happens we need to put the request back
1264 * on the queue. Must be called with queue lock held.
1265 */
Jens Axboe165125e2007-07-24 09:28:11 +02001266void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001268 blk_delete_timer(rq);
1269 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001270 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001271
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001272 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 blk_queue_end_tag(q, rq);
1274
James Bottomleyba396a62009-05-27 14:17:08 +02001275 BUG_ON(blk_queued_rq(rq));
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 elv_requeue_request(q, rq);
1278}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279EXPORT_SYMBOL(blk_requeue_request);
1280
Jens Axboe73c10102011-03-08 13:19:51 +01001281static void add_acct_request(struct request_queue *q, struct request *rq,
1282 int where)
1283{
Jens Axboe320ae512013-10-24 09:20:05 +01001284 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001285 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001286}
1287
Tejun Heo074a7ac2008-08-25 19:56:14 +09001288static void part_round_stats_single(int cpu, struct hd_struct *part,
1289 unsigned long now)
1290{
Jens Axboe7276d022014-05-09 15:48:23 -06001291 int inflight;
1292
Tejun Heo074a7ac2008-08-25 19:56:14 +09001293 if (now == part->stamp)
1294 return;
1295
Jens Axboe7276d022014-05-09 15:48:23 -06001296 inflight = part_in_flight(part);
1297 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001298 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001299 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001300 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1301 }
1302 part->stamp = now;
1303}
1304
1305/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001306 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1307 * @cpu: cpu number for stats access
1308 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 *
1310 * The average IO queue length and utilisation statistics are maintained
1311 * by observing the current state of the queue length and the amount of
1312 * time it has been in this state for.
1313 *
1314 * Normally, that accounting is done on IO completion, but that can result
1315 * in more than a second's worth of IO being accounted for within any one
1316 * second, leading to >100% utilisation. To deal with that, we call this
1317 * function to do a round-off before returning the results when reading
1318 * /proc/diskstats. This accounts immediately for all queue usage up to
1319 * the current jiffies and restarts the counters again.
1320 */
Tejun Heoc9959052008-08-25 19:47:21 +09001321void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001322{
1323 unsigned long now = jiffies;
1324
Tejun Heo074a7ac2008-08-25 19:56:14 +09001325 if (part->partno)
1326 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1327 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001328}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001329EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001330
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001331#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001332static void blk_pm_put_request(struct request *rq)
1333{
1334 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1335 pm_runtime_mark_last_busy(rq->q->dev);
1336}
1337#else
1338static inline void blk_pm_put_request(struct request *rq) {}
1339#endif
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341/*
1342 * queue lock must be held
1343 */
Jens Axboe165125e2007-07-24 09:28:11 +02001344void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 if (unlikely(!q))
1347 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001349 if (q->mq_ops) {
1350 blk_mq_free_request(req);
1351 return;
1352 }
1353
Lin Mingc8158812013-03-23 11:42:27 +08001354 blk_pm_put_request(req);
1355
Tejun Heo8922e162005-10-20 16:23:44 +02001356 elv_completed_request(q, req);
1357
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001358 /* this is a bio leak */
1359 WARN_ON(req->bio != NULL);
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 /*
1362 * Request may not have originated from ll_rw_blk. if not,
1363 * it didn't come out of our reserved rq pools
1364 */
Jens Axboe49171e52006-08-10 08:59:11 +02001365 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001366 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001367 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001370 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Tejun Heoa0516612012-06-26 15:05:44 -07001372 blk_free_request(rl, req);
1373 freed_request(rl, flags);
1374 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 }
1376}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001377EXPORT_SYMBOL_GPL(__blk_put_request);
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379void blk_put_request(struct request *req)
1380{
Jens Axboe165125e2007-07-24 09:28:11 +02001381 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Jens Axboe320ae512013-10-24 09:20:05 +01001383 if (q->mq_ops)
1384 blk_mq_free_request(req);
1385 else {
1386 unsigned long flags;
1387
1388 spin_lock_irqsave(q->queue_lock, flags);
1389 __blk_put_request(q, req);
1390 spin_unlock_irqrestore(q->queue_lock, flags);
1391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393EXPORT_SYMBOL(blk_put_request);
1394
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001395/**
1396 * blk_add_request_payload - add a payload to a request
1397 * @rq: request to update
1398 * @page: page backing the payload
1399 * @len: length of the payload.
1400 *
1401 * This allows to later add a payload to an already submitted request by
1402 * a block driver. The driver needs to take care of freeing the payload
1403 * itself.
1404 *
1405 * Note that this is a quite horrible hack and nothing but handling of
1406 * discard requests should ever use it.
1407 */
1408void blk_add_request_payload(struct request *rq, struct page *page,
1409 unsigned int len)
1410{
1411 struct bio *bio = rq->bio;
1412
1413 bio->bi_io_vec->bv_page = page;
1414 bio->bi_io_vec->bv_offset = 0;
1415 bio->bi_io_vec->bv_len = len;
1416
Kent Overstreet4f024f32013-10-11 15:44:27 -07001417 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001418 bio->bi_vcnt = 1;
1419 bio->bi_phys_segments = 1;
1420
1421 rq->__data_len = rq->resid_len = len;
1422 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001423}
1424EXPORT_SYMBOL_GPL(blk_add_request_payload);
1425
Jens Axboe320ae512013-10-24 09:20:05 +01001426bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1427 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001428{
1429 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1430
Jens Axboe73c10102011-03-08 13:19:51 +01001431 if (!ll_back_merge_fn(q, req, bio))
1432 return false;
1433
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001434 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001435
1436 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1437 blk_rq_set_mixed_merge(req);
1438
1439 req->biotail->bi_next = bio;
1440 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001441 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001442 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1443
Jens Axboe320ae512013-10-24 09:20:05 +01001444 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001445 return true;
1446}
1447
Jens Axboe320ae512013-10-24 09:20:05 +01001448bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1449 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001450{
1451 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001452
Jens Axboe73c10102011-03-08 13:19:51 +01001453 if (!ll_front_merge_fn(q, req, bio))
1454 return false;
1455
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001456 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001457
1458 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1459 blk_rq_set_mixed_merge(req);
1460
Jens Axboe73c10102011-03-08 13:19:51 +01001461 bio->bi_next = req->bio;
1462 req->bio = bio;
1463
Kent Overstreet4f024f32013-10-11 15:44:27 -07001464 req->__sector = bio->bi_iter.bi_sector;
1465 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001466 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1467
Jens Axboe320ae512013-10-24 09:20:05 +01001468 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001469 return true;
1470}
1471
Tejun Heobd87b582011-10-19 14:33:08 +02001472/**
Jens Axboe320ae512013-10-24 09:20:05 +01001473 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001474 * @q: request_queue new bio is being queued at
1475 * @bio: new bio being queued
1476 * @request_count: out parameter for number of traversed plugged requests
1477 *
1478 * Determine whether @bio being queued on @q can be merged with a request
1479 * on %current's plugged list. Returns %true if merge was successful,
1480 * otherwise %false.
1481 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001482 * Plugging coalesces IOs from the same issuer for the same purpose without
1483 * going through @q->queue_lock. As such it's more of an issuing mechanism
1484 * than scheduling, and the request, while may have elvpriv data, is not
1485 * added on the elevator at this point. In addition, we don't have
1486 * reliable access to the elevator outside queue lock. Only check basic
1487 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001488 *
1489 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001490 */
Jens Axboe320ae512013-10-24 09:20:05 +01001491bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1492 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001493{
1494 struct blk_plug *plug;
1495 struct request *rq;
1496 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001497 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001498
Tejun Heobd87b582011-10-19 14:33:08 +02001499 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001500 if (!plug)
1501 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001502 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001503
Shaohua Li92f399c2013-10-29 12:01:03 -06001504 if (q->mq_ops)
1505 plug_list = &plug->mq_list;
1506 else
1507 plug_list = &plug->list;
1508
1509 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001510 int el_ret;
1511
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001512 if (rq->q == q)
1513 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001514
Tejun Heo07c2bd32012-02-08 09:19:42 +01001515 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001516 continue;
1517
Tejun Heo050c8ea2012-02-08 09:19:38 +01001518 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001519 if (el_ret == ELEVATOR_BACK_MERGE) {
1520 ret = bio_attempt_back_merge(q, rq, bio);
1521 if (ret)
1522 break;
1523 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1524 ret = bio_attempt_front_merge(q, rq, bio);
1525 if (ret)
1526 break;
1527 }
1528 }
1529out:
1530 return ret;
1531}
1532
Jens Axboe86db1e22008-01-29 14:53:40 +01001533void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001534{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001535 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001536
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001537 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1538 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001539 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001540
Tejun Heo52d9e672006-01-06 09:49:58 +01001541 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001542 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001543 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001544 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001545}
1546
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001547void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001549 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001550 struct blk_plug *plug;
1551 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1552 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001553 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 /*
1556 * low level driver can indicate that it wants pages above a
1557 * certain limit bounced to low memory (ie for highmem, or even
1558 * ISA dma in theory)
1559 */
1560 blk_queue_bounce(q, &bio);
1561
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001562 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1563 bio_endio(bio, -EIO);
1564 return;
1565 }
1566
Tejun Heo4fed9472010-09-03 11:56:17 +02001567 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001568 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001569 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001570 goto get_rq;
1571 }
1572
Jens Axboe73c10102011-03-08 13:19:51 +01001573 /*
1574 * Check if we can merge with the plugged list before grabbing
1575 * any locks.
1576 */
Robert Elliottda41a582014-05-20 16:46:26 -05001577 if (!blk_queue_nomerges(q) &&
1578 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001579 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001580
1581 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
1583 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001584 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001585 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001586 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001587 if (!attempt_back_merge(q, req))
1588 elv_merged_request(q, req, el_ret);
1589 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001590 }
Jens Axboe73c10102011-03-08 13:19:51 +01001591 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001592 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001593 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001594 if (!attempt_front_merge(q, req))
1595 elv_merged_request(q, req, el_ret);
1596 goto out_unlock;
1597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001601 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001602 * This sync check and mask will be re-done in init_request_from_bio(),
1603 * but we need to set it earlier to expose the sync flag to the
1604 * rq allocator and io schedulers.
1605 */
1606 rw_flags = bio_data_dir(bio);
1607 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001608 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001609
1610 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001611 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001612 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001613 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001614 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001615 if (IS_ERR(req)) {
1616 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
Tejun Heoda8303c2011-10-19 14:33:05 +02001617 goto out_unlock;
1618 }
Nick Piggind6344532005-06-28 20:45:14 -07001619
Nick Piggin450991b2005-06-28 20:45:13 -07001620 /*
1621 * After dropping the lock and possibly sleeping here, our request
1622 * may now be mergeable after it had proven unmergeable (above).
1623 * We don't worry about that case for efficiency. It won't happen
1624 * often, and the elevators are able to handle it.
1625 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001626 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
Tao Ma9562ad92011-10-24 16:11:30 +02001628 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001629 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001630
Jens Axboe73c10102011-03-08 13:19:51 +01001631 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001632 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001633 /*
1634 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001635 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001636 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001637 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001638 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001639 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001640 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001641 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001642 trace_block_plug(q);
1643 }
Jens Axboe73c10102011-03-08 13:19:51 +01001644 }
Shaohua Lia6327162011-08-24 16:04:32 +02001645 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001646 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001647 } else {
1648 spin_lock_irq(q->queue_lock);
1649 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001650 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001651out_unlock:
1652 spin_unlock_irq(q->queue_lock);
1653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654}
Jens Axboec20e8de2011-09-12 12:03:37 +02001655EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657/*
1658 * If bio->bi_dev is a partition, remap the location
1659 */
1660static inline void blk_partition_remap(struct bio *bio)
1661{
1662 struct block_device *bdev = bio->bi_bdev;
1663
Jens Axboebf2de6f2007-09-27 13:01:25 +02001664 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001666
Kent Overstreet4f024f32013-10-11 15:44:27 -07001667 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001669
Mike Snitzerd07335e2010-11-16 12:52:38 +01001670 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1671 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001672 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
1674}
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676static void handle_bad_sector(struct bio *bio)
1677{
1678 char b[BDEVNAME_SIZE];
1679
1680 printk(KERN_INFO "attempt to access beyond end of device\n");
1681 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1682 bdevname(bio->bi_bdev, b),
1683 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001684 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001685 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687 set_bit(BIO_EOF, &bio->bi_flags);
1688}
1689
Akinobu Mitac17bb492006-12-08 02:39:46 -08001690#ifdef CONFIG_FAIL_MAKE_REQUEST
1691
1692static DECLARE_FAULT_ATTR(fail_make_request);
1693
1694static int __init setup_fail_make_request(char *str)
1695{
1696 return setup_fault_attr(&fail_make_request, str);
1697}
1698__setup("fail_make_request=", setup_fail_make_request);
1699
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001700static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001701{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001702 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001703}
1704
1705static int __init fail_make_request_debugfs(void)
1706{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001707 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1708 NULL, &fail_make_request);
1709
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001710 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001711}
1712
1713late_initcall(fail_make_request_debugfs);
1714
1715#else /* CONFIG_FAIL_MAKE_REQUEST */
1716
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001717static inline bool should_fail_request(struct hd_struct *part,
1718 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001719{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001720 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001721}
1722
1723#endif /* CONFIG_FAIL_MAKE_REQUEST */
1724
Jens Axboec07e2b42007-07-18 13:27:58 +02001725/*
1726 * Check whether this bio extends beyond the end of the device.
1727 */
1728static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1729{
1730 sector_t maxsector;
1731
1732 if (!nr_sectors)
1733 return 0;
1734
1735 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001736 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001737 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001738 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001739
1740 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1741 /*
1742 * This may well happen - the kernel calls bread()
1743 * without checking the size of the device, e.g., when
1744 * mounting a device.
1745 */
1746 handle_bad_sector(bio);
1747 return 1;
1748 }
1749 }
1750
1751 return 0;
1752}
1753
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001754static noinline_for_stack bool
1755generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756{
Jens Axboe165125e2007-07-24 09:28:11 +02001757 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001758 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001759 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001760 char b[BDEVNAME_SIZE];
1761 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
1763 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Jens Axboec07e2b42007-07-18 13:27:58 +02001765 if (bio_check_eod(bio, nr_sectors))
1766 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001768 q = bdev_get_queue(bio->bi_bdev);
1769 if (unlikely(!q)) {
1770 printk(KERN_ERR
1771 "generic_make_request: Trying to access "
1772 "nonexistent block-device %s (%Lu)\n",
1773 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001774 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001775 goto end_io;
1776 }
1777
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001778 if (likely(bio_is_rw(bio) &&
1779 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001780 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1781 bdevname(bio->bi_bdev, b),
1782 bio_sectors(bio),
1783 queue_max_hw_sectors(q));
1784 goto end_io;
1785 }
1786
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001787 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001788 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001789 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001790 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001791 goto end_io;
1792
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001794 * If this device has partitions, remap block n
1795 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001797 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001799 if (bio_check_eod(bio, nr_sectors))
1800 goto end_io;
1801
1802 /*
1803 * Filter flush bio's early so that make_request based
1804 * drivers without flush support don't have to worry
1805 * about them.
1806 */
1807 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1808 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1809 if (!nr_sectors) {
1810 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001811 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001815 if ((bio->bi_rw & REQ_DISCARD) &&
1816 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001817 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001818 err = -EOPNOTSUPP;
1819 goto end_io;
1820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001822 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 err = -EOPNOTSUPP;
1824 goto end_io;
1825 }
1826
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001827 /*
1828 * Various block parts want %current->io_context and lazy ioc
1829 * allocation ends up trading a lot of pain for a small amount of
1830 * memory. Just allocate it upfront. This may fail and block
1831 * layer knows how to live with it.
1832 */
1833 create_io_context(GFP_ATOMIC, q->node);
1834
Tejun Heobc16a4f2011-10-19 14:33:01 +02001835 if (blk_throtl_bio(q, bio))
1836 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001837
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001838 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001839 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001840
1841end_io:
1842 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001843 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001846/**
1847 * generic_make_request - hand a buffer to its device driver for I/O
1848 * @bio: The bio describing the location in memory and on the device.
1849 *
1850 * generic_make_request() is used to make I/O requests of block
1851 * devices. It is passed a &struct bio, which describes the I/O that needs
1852 * to be done.
1853 *
1854 * generic_make_request() does not return any status. The
1855 * success/failure status of the request, along with notification of
1856 * completion, is delivered asynchronously through the bio->bi_end_io
1857 * function described (one day) else where.
1858 *
1859 * The caller of generic_make_request must make sure that bi_io_vec
1860 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1861 * set to describe the device address, and the
1862 * bi_end_io and optionally bi_private are set to describe how
1863 * completion notification should be signaled.
1864 *
1865 * generic_make_request and the drivers it calls may use bi_next if this
1866 * bio happens to be merged with someone else, and may resubmit the bio to
1867 * a lower device by calling into generic_make_request recursively, which
1868 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001869 */
1870void generic_make_request(struct bio *bio)
1871{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001872 struct bio_list bio_list_on_stack;
1873
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001874 if (!generic_make_request_checks(bio))
1875 return;
1876
1877 /*
1878 * We only want one ->make_request_fn to be active at a time, else
1879 * stack usage with stacked devices could be a problem. So use
1880 * current->bio_list to keep a list of requests submited by a
1881 * make_request_fn function. current->bio_list is also used as a
1882 * flag to say if generic_make_request is currently active in this
1883 * task or not. If it is NULL, then no make_request is active. If
1884 * it is non-NULL, then a make_request is active, and new requests
1885 * should be added at the tail
1886 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001887 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001888 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001889 return;
1890 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001891
Neil Brownd89d8792007-05-01 09:53:42 +02001892 /* following loop may be a bit non-obvious, and so deserves some
1893 * explanation.
1894 * Before entering the loop, bio->bi_next is NULL (as all callers
1895 * ensure that) so we have a list with a single bio.
1896 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001897 * we assign bio_list to a pointer to the bio_list_on_stack,
1898 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001899 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001900 * through a recursive call to generic_make_request. If it
1901 * did, we find a non-NULL value in bio_list and re-enter the loop
1902 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001903 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001904 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001905 */
1906 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001907 bio_list_init(&bio_list_on_stack);
1908 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001909 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001910 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1911
1912 q->make_request_fn(q, bio);
1913
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001914 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001915 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001916 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001917}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918EXPORT_SYMBOL(generic_make_request);
1919
1920/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001921 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1923 * @bio: The &struct bio which describes the I/O
1924 *
1925 * submit_bio() is very similar in purpose to generic_make_request(), and
1926 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001927 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 *
1929 */
1930void submit_bio(int rw, struct bio *bio)
1931{
Jens Axboe22e2c502005-06-27 10:55:12 +02001932 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
Jens Axboebf2de6f2007-09-27 13:01:25 +02001934 /*
1935 * If it's a regular read/write or a barrier with data attached,
1936 * go through the normal accounting stuff before submission.
1937 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001938 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001939 unsigned int count;
1940
1941 if (unlikely(rw & REQ_WRITE_SAME))
1942 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1943 else
1944 count = bio_sectors(bio);
1945
Jens Axboebf2de6f2007-09-27 13:01:25 +02001946 if (rw & WRITE) {
1947 count_vm_events(PGPGOUT, count);
1948 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001949 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001950 count_vm_events(PGPGIN, count);
1951 }
1952
1953 if (unlikely(block_dump)) {
1954 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001955 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001956 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001957 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001958 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001959 bdevname(bio->bi_bdev, b),
1960 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963
1964 generic_make_request(bio);
1965}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966EXPORT_SYMBOL(submit_bio);
1967
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001968/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001969 * blk_rq_check_limits - Helper function to check a request for the queue limit
1970 * @q: the queue
1971 * @rq: the request being checked
1972 *
1973 * Description:
1974 * @rq may have been made based on weaker limitations of upper-level queues
1975 * in request stacking drivers, and it may violate the limitation of @q.
1976 * Since the block layer and the underlying device driver trust @rq
1977 * after it is inserted to @q, it should be checked against @q before
1978 * the insertion using this generic function.
1979 *
1980 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001981 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001982 * Request stacking drivers like request-based dm may change the queue
1983 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09001984 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001985 * the new queue limits again when they dispatch those requests,
1986 * although such checkings are also done against the old queue limits
1987 * when submitting requests.
1988 */
1989int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1990{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001991 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001992 return 0;
1993
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001994 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001995 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1996 return -EIO;
1997 }
1998
1999 /*
2000 * queue's settings related to segment counting like q->bounce_pfn
2001 * may differ from that of other stacking queues.
2002 * Recalculate it to check the request correctly on this queue's
2003 * limitation.
2004 */
2005 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002006 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002007 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2008 return -EIO;
2009 }
2010
2011 return 0;
2012}
2013EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2014
2015/**
2016 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2017 * @q: the queue to submit the request
2018 * @rq: the request being queued
2019 */
2020int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2021{
2022 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002023 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002024
2025 if (blk_rq_check_limits(q, rq))
2026 return -EIO;
2027
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002028 if (rq->rq_disk &&
2029 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002030 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002031
2032 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002033 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002034 spin_unlock_irqrestore(q->queue_lock, flags);
2035 return -ENODEV;
2036 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002037
2038 /*
2039 * Submitting request must be dequeued before calling this function
2040 * because it will be linked to another request_queue
2041 */
2042 BUG_ON(blk_queued_rq(rq));
2043
Jeff Moyer4853aba2011-08-15 21:37:25 +02002044 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2045 where = ELEVATOR_INSERT_FLUSH;
2046
2047 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002048 if (where == ELEVATOR_INSERT_FLUSH)
2049 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002050 spin_unlock_irqrestore(q->queue_lock, flags);
2051
2052 return 0;
2053}
2054EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2055
Tejun Heo80a761f2009-07-03 17:48:17 +09002056/**
2057 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2058 * @rq: request to examine
2059 *
2060 * Description:
2061 * A request could be merge of IOs which require different failure
2062 * handling. This function determines the number of bytes which
2063 * can be failed from the beginning of the request without
2064 * crossing into area which need to be retried further.
2065 *
2066 * Return:
2067 * The number of bytes to fail.
2068 *
2069 * Context:
2070 * queue_lock must be held.
2071 */
2072unsigned int blk_rq_err_bytes(const struct request *rq)
2073{
2074 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2075 unsigned int bytes = 0;
2076 struct bio *bio;
2077
2078 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2079 return blk_rq_bytes(rq);
2080
2081 /*
2082 * Currently the only 'mixing' which can happen is between
2083 * different fastfail types. We can safely fail portions
2084 * which have all the failfast bits that the first one has -
2085 * the ones which are at least as eager to fail as the first
2086 * one.
2087 */
2088 for (bio = rq->bio; bio; bio = bio->bi_next) {
2089 if ((bio->bi_rw & ff) != ff)
2090 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002091 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002092 }
2093
2094 /* this could lead to infinite loop */
2095 BUG_ON(blk_rq_bytes(rq) && !bytes);
2096 return bytes;
2097}
2098EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2099
Jens Axboe320ae512013-10-24 09:20:05 +01002100void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002101{
Jens Axboec2553b52009-04-24 08:10:11 +02002102 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002103 const int rw = rq_data_dir(req);
2104 struct hd_struct *part;
2105 int cpu;
2106
2107 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002108 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002109 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2110 part_stat_unlock();
2111 }
2112}
2113
Jens Axboe320ae512013-10-24 09:20:05 +01002114void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002115{
Jens Axboebc58ba92009-01-23 10:54:44 +01002116 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002117 * Account IO completion. flush_rq isn't accounted as a
2118 * normal IO on queueing nor completion. Accounting the
2119 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002120 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002121 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002122 unsigned long duration = jiffies - req->start_time;
2123 const int rw = rq_data_dir(req);
2124 struct hd_struct *part;
2125 int cpu;
2126
2127 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002128 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002129
2130 part_stat_inc(cpu, part, ios[rw]);
2131 part_stat_add(cpu, part, ticks[rw], duration);
2132 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002133 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002134
Jens Axboe6c23a962011-01-07 08:43:37 +01002135 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002136 part_stat_unlock();
2137 }
2138}
2139
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002140#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002141/*
2142 * Don't process normal requests when queue is suspended
2143 * or in the process of suspending/resuming
2144 */
2145static struct request *blk_pm_peek_request(struct request_queue *q,
2146 struct request *rq)
2147{
2148 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2149 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2150 return NULL;
2151 else
2152 return rq;
2153}
2154#else
2155static inline struct request *blk_pm_peek_request(struct request_queue *q,
2156 struct request *rq)
2157{
2158 return rq;
2159}
2160#endif
2161
Jens Axboe320ae512013-10-24 09:20:05 +01002162void blk_account_io_start(struct request *rq, bool new_io)
2163{
2164 struct hd_struct *part;
2165 int rw = rq_data_dir(rq);
2166 int cpu;
2167
2168 if (!blk_do_io_stat(rq))
2169 return;
2170
2171 cpu = part_stat_lock();
2172
2173 if (!new_io) {
2174 part = rq->part;
2175 part_stat_inc(cpu, part, merges[rw]);
2176 } else {
2177 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2178 if (!hd_struct_try_get(part)) {
2179 /*
2180 * The partition is already being removed,
2181 * the request will be accounted on the disk only
2182 *
2183 * We take a reference on disk->part0 although that
2184 * partition will never be deleted, so we can treat
2185 * it as any other partition.
2186 */
2187 part = &rq->rq_disk->part0;
2188 hd_struct_get(part);
2189 }
2190 part_round_stats(cpu, part);
2191 part_inc_in_flight(part, rw);
2192 rq->part = part;
2193 }
2194
2195 part_stat_unlock();
2196}
2197
Tejun Heo53a08802008-12-03 12:41:26 +01002198/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002199 * blk_peek_request - peek at the top of a request queue
2200 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002201 *
2202 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002203 * Return the request at the top of @q. The returned request
2204 * should be started using blk_start_request() before LLD starts
2205 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002206 *
2207 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002208 * Pointer to the request at the top of @q if available. Null
2209 * otherwise.
2210 *
2211 * Context:
2212 * queue_lock must be held.
2213 */
2214struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002215{
2216 struct request *rq;
2217 int ret;
2218
2219 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002220
2221 rq = blk_pm_peek_request(q, rq);
2222 if (!rq)
2223 break;
2224
Tejun Heo158dbda2009-04-23 11:05:18 +09002225 if (!(rq->cmd_flags & REQ_STARTED)) {
2226 /*
2227 * This is the first time the device driver
2228 * sees this request (possibly after
2229 * requeueing). Notify IO scheduler.
2230 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002231 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002232 elv_activate_rq(q, rq);
2233
2234 /*
2235 * just mark as started even if we don't start
2236 * it, a request that has been delayed should
2237 * not be passed by new incoming requests
2238 */
2239 rq->cmd_flags |= REQ_STARTED;
2240 trace_block_rq_issue(q, rq);
2241 }
2242
2243 if (!q->boundary_rq || q->boundary_rq == rq) {
2244 q->end_sector = rq_end_sector(rq);
2245 q->boundary_rq = NULL;
2246 }
2247
2248 if (rq->cmd_flags & REQ_DONTPREP)
2249 break;
2250
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002251 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002252 /*
2253 * make sure space for the drain appears we
2254 * know we can do this because max_hw_segments
2255 * has been adjusted to be one fewer than the
2256 * device can handle
2257 */
2258 rq->nr_phys_segments++;
2259 }
2260
2261 if (!q->prep_rq_fn)
2262 break;
2263
2264 ret = q->prep_rq_fn(q, rq);
2265 if (ret == BLKPREP_OK) {
2266 break;
2267 } else if (ret == BLKPREP_DEFER) {
2268 /*
2269 * the request may have been (partially) prepped.
2270 * we need to keep this request in the front to
2271 * avoid resource deadlock. REQ_STARTED will
2272 * prevent other fs requests from passing this one.
2273 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002274 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002275 !(rq->cmd_flags & REQ_DONTPREP)) {
2276 /*
2277 * remove the space for the drain we added
2278 * so that we don't add it again
2279 */
2280 --rq->nr_phys_segments;
2281 }
2282
2283 rq = NULL;
2284 break;
2285 } else if (ret == BLKPREP_KILL) {
2286 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002287 /*
2288 * Mark this request as started so we don't trigger
2289 * any debug logic in the end I/O path.
2290 */
2291 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002292 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002293 } else {
2294 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2295 break;
2296 }
2297 }
2298
2299 return rq;
2300}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002301EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002302
Tejun Heo9934c8c2009-05-08 11:54:16 +09002303void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002304{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002305 struct request_queue *q = rq->q;
2306
Tejun Heo158dbda2009-04-23 11:05:18 +09002307 BUG_ON(list_empty(&rq->queuelist));
2308 BUG_ON(ELV_ON_HASH(rq));
2309
2310 list_del_init(&rq->queuelist);
2311
2312 /*
2313 * the time frame between a request being removed from the lists
2314 * and to it is freed is accounted as io that is in progress at
2315 * the driver side.
2316 */
Divyesh Shah91952912010-04-01 15:01:41 -07002317 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002318 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002319 set_io_start_time_ns(rq);
2320 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002321}
2322
Tejun Heo5efccd12009-04-23 11:05:18 +09002323/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002324 * blk_start_request - start request processing on the driver
2325 * @req: request to dequeue
2326 *
2327 * Description:
2328 * Dequeue @req and start timeout timer on it. This hands off the
2329 * request to the driver.
2330 *
2331 * Block internal functions which don't want to start timer should
2332 * call blk_dequeue_request().
2333 *
2334 * Context:
2335 * queue_lock must be held.
2336 */
2337void blk_start_request(struct request *req)
2338{
2339 blk_dequeue_request(req);
2340
2341 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002342 * We are now handing the request to the hardware, initialize
2343 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002344 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002345 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002346 if (unlikely(blk_bidi_rq(req)))
2347 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2348
Jeff Moyer4912aa62013-10-08 14:36:41 -04002349 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002350 blk_add_timer(req);
2351}
2352EXPORT_SYMBOL(blk_start_request);
2353
2354/**
2355 * blk_fetch_request - fetch a request from a request queue
2356 * @q: request queue to fetch a request from
2357 *
2358 * Description:
2359 * Return the request at the top of @q. The request is started on
2360 * return and LLD can start processing it immediately.
2361 *
2362 * Return:
2363 * Pointer to the request at the top of @q if available. Null
2364 * otherwise.
2365 *
2366 * Context:
2367 * queue_lock must be held.
2368 */
2369struct request *blk_fetch_request(struct request_queue *q)
2370{
2371 struct request *rq;
2372
2373 rq = blk_peek_request(q);
2374 if (rq)
2375 blk_start_request(rq);
2376 return rq;
2377}
2378EXPORT_SYMBOL(blk_fetch_request);
2379
2380/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002381 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002382 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002383 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002384 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002385 *
2386 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002387 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2388 * the request structure even if @req doesn't have leftover.
2389 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002390 *
2391 * This special helper function is only for request stacking drivers
2392 * (e.g. request-based dm) so that they can handle partial completion.
2393 * Actual device drivers should use blk_end_request instead.
2394 *
2395 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2396 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002397 *
2398 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002399 * %false - this request doesn't have any more data
2400 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002401 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002402bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002404 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002406 trace_block_rq_complete(req->q, req, nr_bytes);
2407
Tejun Heo2e60e022009-04-23 11:05:18 +09002408 if (!req->bio)
2409 return false;
2410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002412 * For fs requests, rq is just carrier of independent bio's
2413 * and each partial completion should be handled separately.
2414 * Reset per-request error on each partial completion.
2415 *
2416 * TODO: tj: This is too subtle. It would be better to let
2417 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002419 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 req->errors = 0;
2421
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002422 if (error && req->cmd_type == REQ_TYPE_FS &&
2423 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002424 char *error_type;
2425
2426 switch (error) {
2427 case -ENOLINK:
2428 error_type = "recoverable transport";
2429 break;
2430 case -EREMOTEIO:
2431 error_type = "critical target";
2432 break;
2433 case -EBADE:
2434 error_type = "critical nexus";
2435 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002436 case -ETIMEDOUT:
2437 error_type = "timeout";
2438 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002439 case -ENOSPC:
2440 error_type = "critical space allocation";
2441 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002442 case -ENODATA:
2443 error_type = "critical medium";
2444 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002445 case -EIO:
2446 default:
2447 error_type = "I/O";
2448 break;
2449 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002450 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2451 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002452 req->rq_disk->disk_name : "?",
2453 (unsigned long long)blk_rq_pos(req));
2454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 }
2456
Jens Axboebc58ba92009-01-23 10:54:44 +01002457 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002458
Kent Overstreetf79ea412012-09-20 16:38:30 -07002459 total_bytes = 0;
2460 while (req->bio) {
2461 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002462 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Kent Overstreet4f024f32013-10-11 15:44:27 -07002464 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Kent Overstreetf79ea412012-09-20 16:38:30 -07002467 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Kent Overstreetf79ea412012-09-20 16:38:30 -07002469 total_bytes += bio_bytes;
2470 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Kent Overstreetf79ea412012-09-20 16:38:30 -07002472 if (!nr_bytes)
2473 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 }
2475
2476 /*
2477 * completely done
2478 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002479 if (!req->bio) {
2480 /*
2481 * Reset counters so that the request stacking driver
2482 * can find how many bytes remain in the request
2483 * later.
2484 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002485 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002486 return false;
2487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002489 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002490
2491 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002492 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002493 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002494
Tejun Heo80a761f2009-07-03 17:48:17 +09002495 /* mixed attributes always follow the first bio */
2496 if (req->cmd_flags & REQ_MIXED_MERGE) {
2497 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2498 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2499 }
2500
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002501 /*
2502 * If total number of sectors is less than the first segment
2503 * size, something has gone terribly wrong.
2504 */
2505 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002506 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002507 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002508 }
2509
2510 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002512
Tejun Heo2e60e022009-04-23 11:05:18 +09002513 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514}
Tejun Heo2e60e022009-04-23 11:05:18 +09002515EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
Tejun Heo2e60e022009-04-23 11:05:18 +09002517static bool blk_update_bidi_request(struct request *rq, int error,
2518 unsigned int nr_bytes,
2519 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002520{
Tejun Heo2e60e022009-04-23 11:05:18 +09002521 if (blk_update_request(rq, error, nr_bytes))
2522 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002523
Tejun Heo2e60e022009-04-23 11:05:18 +09002524 /* Bidi request must be completed as a whole */
2525 if (unlikely(blk_bidi_rq(rq)) &&
2526 blk_update_request(rq->next_rq, error, bidi_bytes))
2527 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002528
Jens Axboee2e1a142010-06-09 10:42:09 +02002529 if (blk_queue_add_random(rq->q))
2530 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002531
2532 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533}
2534
James Bottomley28018c22010-07-01 19:49:17 +09002535/**
2536 * blk_unprep_request - unprepare a request
2537 * @req: the request
2538 *
2539 * This function makes a request ready for complete resubmission (or
2540 * completion). It happens only after all error handling is complete,
2541 * so represents the appropriate moment to deallocate any resources
2542 * that were allocated to the request in the prep_rq_fn. The queue
2543 * lock is held when calling this.
2544 */
2545void blk_unprep_request(struct request *req)
2546{
2547 struct request_queue *q = req->q;
2548
2549 req->cmd_flags &= ~REQ_DONTPREP;
2550 if (q->unprep_rq_fn)
2551 q->unprep_rq_fn(q, req);
2552}
2553EXPORT_SYMBOL_GPL(blk_unprep_request);
2554
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555/*
2556 * queue lock must be held
2557 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002558void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002560 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002561 blk_queue_end_tag(req->q, req);
2562
James Bottomleyba396a62009-05-27 14:17:08 +02002563 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002565 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002566 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Mike Andersone78042e2008-10-30 02:16:20 -07002568 blk_delete_timer(req);
2569
James Bottomley28018c22010-07-01 19:49:17 +09002570 if (req->cmd_flags & REQ_DONTPREP)
2571 blk_unprep_request(req);
2572
Jens Axboebc58ba92009-01-23 10:54:44 +01002573 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002574
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002576 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002577 else {
2578 if (blk_bidi_rq(req))
2579 __blk_put_request(req->next_rq->q, req->next_rq);
2580
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 }
2583}
Christoph Hellwig12120072014-04-16 09:44:59 +02002584EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002586/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002587 * blk_end_bidi_request - Complete a bidi request
2588 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002589 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002590 * @nr_bytes: number of bytes to complete @rq
2591 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002592 *
2593 * Description:
2594 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002595 * Drivers that supports bidi can safely call this member for any
2596 * type of request, bidi or uni. In the later case @bidi_bytes is
2597 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002598 *
2599 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002600 * %false - we are done with this request
2601 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002602 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002603static bool blk_end_bidi_request(struct request *rq, int error,
2604 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002605{
2606 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002607 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002608
Tejun Heo2e60e022009-04-23 11:05:18 +09002609 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2610 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002611
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002612 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002613 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002614 spin_unlock_irqrestore(q->queue_lock, flags);
2615
Tejun Heo2e60e022009-04-23 11:05:18 +09002616 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002617}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002618
2619/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002620 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2621 * @rq: the request to complete
2622 * @error: %0 for success, < %0 for error
2623 * @nr_bytes: number of bytes to complete @rq
2624 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002625 *
2626 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002627 * Identical to blk_end_bidi_request() except that queue lock is
2628 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002629 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002630 * Return:
2631 * %false - we are done with this request
2632 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002633 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002634bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002635 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002636{
Tejun Heo2e60e022009-04-23 11:05:18 +09002637 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2638 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002639
Tejun Heo2e60e022009-04-23 11:05:18 +09002640 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002641
Tejun Heo2e60e022009-04-23 11:05:18 +09002642 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002643}
2644
2645/**
2646 * blk_end_request - Helper function for drivers to complete the request.
2647 * @rq: the request being processed
2648 * @error: %0 for success, < %0 for error
2649 * @nr_bytes: number of bytes to complete
2650 *
2651 * Description:
2652 * Ends I/O on a number of bytes attached to @rq.
2653 * If @rq has leftover, sets it up for the next range of segments.
2654 *
2655 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002656 * %false - we are done with this request
2657 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002658 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002659bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002660{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002661 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002662}
Jens Axboe56ad1742009-07-28 22:11:24 +02002663EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002664
2665/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002666 * blk_end_request_all - Helper function for drives to finish the request.
2667 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002668 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002669 *
2670 * Description:
2671 * Completely finish @rq.
2672 */
2673void blk_end_request_all(struct request *rq, int error)
2674{
2675 bool pending;
2676 unsigned int bidi_bytes = 0;
2677
2678 if (unlikely(blk_bidi_rq(rq)))
2679 bidi_bytes = blk_rq_bytes(rq->next_rq);
2680
2681 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2682 BUG_ON(pending);
2683}
Jens Axboe56ad1742009-07-28 22:11:24 +02002684EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002685
2686/**
2687 * blk_end_request_cur - Helper function to finish the current request chunk.
2688 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002689 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002690 *
2691 * Description:
2692 * Complete the current consecutively mapped chunk from @rq.
2693 *
2694 * Return:
2695 * %false - we are done with this request
2696 * %true - still buffers pending for this request
2697 */
2698bool blk_end_request_cur(struct request *rq, int error)
2699{
2700 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2701}
Jens Axboe56ad1742009-07-28 22:11:24 +02002702EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002703
2704/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002705 * blk_end_request_err - Finish a request till the next failure boundary.
2706 * @rq: the request to finish till the next failure boundary for
2707 * @error: must be negative errno
2708 *
2709 * Description:
2710 * Complete @rq till the next failure boundary.
2711 *
2712 * Return:
2713 * %false - we are done with this request
2714 * %true - still buffers pending for this request
2715 */
2716bool blk_end_request_err(struct request *rq, int error)
2717{
2718 WARN_ON(error >= 0);
2719 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2720}
2721EXPORT_SYMBOL_GPL(blk_end_request_err);
2722
2723/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002724 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002725 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002726 * @error: %0 for success, < %0 for error
2727 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002728 *
2729 * Description:
2730 * Must be called with queue lock held unlike blk_end_request().
2731 *
2732 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002733 * %false - we are done with this request
2734 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002735 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002736bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002737{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002738 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002739}
Jens Axboe56ad1742009-07-28 22:11:24 +02002740EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002741
2742/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002743 * __blk_end_request_all - Helper function for drives to finish the request.
2744 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002745 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002746 *
2747 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002748 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002749 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002750void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002751{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002752 bool pending;
2753 unsigned int bidi_bytes = 0;
2754
2755 if (unlikely(blk_bidi_rq(rq)))
2756 bidi_bytes = blk_rq_bytes(rq->next_rq);
2757
2758 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2759 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002760}
Jens Axboe56ad1742009-07-28 22:11:24 +02002761EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002762
2763/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002764 * __blk_end_request_cur - Helper function to finish the current request chunk.
2765 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002766 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002767 *
2768 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002769 * Complete the current consecutively mapped chunk from @rq. Must
2770 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002771 *
2772 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002773 * %false - we are done with this request
2774 * %true - still buffers pending for this request
2775 */
2776bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002777{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002778 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002779}
Jens Axboe56ad1742009-07-28 22:11:24 +02002780EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002781
Tejun Heo80a761f2009-07-03 17:48:17 +09002782/**
2783 * __blk_end_request_err - Finish a request till the next failure boundary.
2784 * @rq: the request to finish till the next failure boundary for
2785 * @error: must be negative errno
2786 *
2787 * Description:
2788 * Complete @rq till the next failure boundary. Must be called
2789 * with queue lock held.
2790 *
2791 * Return:
2792 * %false - we are done with this request
2793 * %true - still buffers pending for this request
2794 */
2795bool __blk_end_request_err(struct request *rq, int error)
2796{
2797 WARN_ON(error >= 0);
2798 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2799}
2800EXPORT_SYMBOL_GPL(__blk_end_request_err);
2801
Jens Axboe86db1e22008-01-29 14:53:40 +01002802void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2803 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002805 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002806 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807
Jens Axboeb4f42e22014-04-10 09:46:28 -06002808 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002809 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002810
Kent Overstreet4f024f32013-10-11 15:44:27 -07002811 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
NeilBrown66846572007-08-16 13:31:28 +02002814 if (bio->bi_bdev)
2815 rq->rq_disk = bio->bi_bdev->bd_disk;
2816}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002818#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2819/**
2820 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2821 * @rq: the request to be flushed
2822 *
2823 * Description:
2824 * Flush all pages in @rq.
2825 */
2826void rq_flush_dcache_pages(struct request *rq)
2827{
2828 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002829 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002830
2831 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002832 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002833}
2834EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2835#endif
2836
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002837/**
2838 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2839 * @q : the queue of the device being checked
2840 *
2841 * Description:
2842 * Check if underlying low-level drivers of a device are busy.
2843 * If the drivers want to export their busy state, they must set own
2844 * exporting function using blk_queue_lld_busy() first.
2845 *
2846 * Basically, this function is used only by request stacking drivers
2847 * to stop dispatching requests to underlying devices when underlying
2848 * devices are busy. This behavior helps more I/O merging on the queue
2849 * of the request stacking driver and prevents I/O throughput regression
2850 * on burst I/O load.
2851 *
2852 * Return:
2853 * 0 - Not busy (The request stacking driver should dispatch request)
2854 * 1 - Busy (The request stacking driver should stop dispatching request)
2855 */
2856int blk_lld_busy(struct request_queue *q)
2857{
2858 if (q->lld_busy_fn)
2859 return q->lld_busy_fn(q);
2860
2861 return 0;
2862}
2863EXPORT_SYMBOL_GPL(blk_lld_busy);
2864
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002865/**
2866 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2867 * @rq: the clone request to be cleaned up
2868 *
2869 * Description:
2870 * Free all bios in @rq for a cloned request.
2871 */
2872void blk_rq_unprep_clone(struct request *rq)
2873{
2874 struct bio *bio;
2875
2876 while ((bio = rq->bio) != NULL) {
2877 rq->bio = bio->bi_next;
2878
2879 bio_put(bio);
2880 }
2881}
2882EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2883
2884/*
2885 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002886 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002887 */
2888static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2889{
2890 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002891 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002892 dst->cmd_type = src->cmd_type;
2893 dst->__sector = blk_rq_pos(src);
2894 dst->__data_len = blk_rq_bytes(src);
2895 dst->nr_phys_segments = src->nr_phys_segments;
2896 dst->ioprio = src->ioprio;
2897 dst->extra_len = src->extra_len;
2898}
2899
2900/**
2901 * blk_rq_prep_clone - Helper function to setup clone request
2902 * @rq: the request to be setup
2903 * @rq_src: original request to be cloned
2904 * @bs: bio_set that bios for clone are allocated from
2905 * @gfp_mask: memory allocation mask for bio
2906 * @bio_ctr: setup function to be called for each clone bio.
2907 * Returns %0 for success, non %0 for failure.
2908 * @data: private data to be passed to @bio_ctr
2909 *
2910 * Description:
2911 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002912 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002913 * are not copied, and copying such parts is the caller's responsibility.
2914 * Also, pages which the original bios are pointing to are not copied
2915 * and the cloned bios just point same pages.
2916 * So cloned bios must be completed before original bios, which means
2917 * the caller must complete @rq before @rq_src.
2918 */
2919int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2920 struct bio_set *bs, gfp_t gfp_mask,
2921 int (*bio_ctr)(struct bio *, struct bio *, void *),
2922 void *data)
2923{
2924 struct bio *bio, *bio_src;
2925
2926 if (!bs)
2927 bs = fs_bio_set;
2928
2929 blk_rq_init(NULL, rq);
2930
2931 __rq_for_each_bio(bio_src, rq_src) {
Junichi Nomura11dfce52014-10-03 17:27:11 -04002932 bio = bio_clone_fast(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002933 if (!bio)
2934 goto free_and_out;
2935
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002936 if (bio_ctr && bio_ctr(bio, bio_src, data))
2937 goto free_and_out;
2938
2939 if (rq->bio) {
2940 rq->biotail->bi_next = bio;
2941 rq->biotail = bio;
2942 } else
2943 rq->bio = rq->biotail = bio;
2944 }
2945
2946 __blk_rq_prep_clone(rq, rq_src);
2947
2948 return 0;
2949
2950free_and_out:
2951 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002952 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002953 blk_rq_unprep_clone(rq);
2954
2955 return -ENOMEM;
2956}
2957EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2958
Jens Axboe59c3d452014-04-08 09:15:35 -06002959int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960{
2961 return queue_work(kblockd_workqueue, work);
2962}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963EXPORT_SYMBOL(kblockd_schedule_work);
2964
Jens Axboe59c3d452014-04-08 09:15:35 -06002965int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2966 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04002967{
2968 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2969}
2970EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2971
Jens Axboe8ab14592014-04-08 09:17:40 -06002972int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2973 unsigned long delay)
2974{
2975 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2976}
2977EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2978
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002979/**
2980 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2981 * @plug: The &struct blk_plug that needs to be initialized
2982 *
2983 * Description:
2984 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2985 * pending I/O should the task end up blocking between blk_start_plug() and
2986 * blk_finish_plug(). This is important from a performance perspective, but
2987 * also ensures that we don't deadlock. For instance, if the task is blocking
2988 * for a memory allocation, memory reclaim could end up wanting to free a
2989 * page belonging to that request that is currently residing in our private
2990 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2991 * this kind of deadlock.
2992 */
Jens Axboe73c10102011-03-08 13:19:51 +01002993void blk_start_plug(struct blk_plug *plug)
2994{
2995 struct task_struct *tsk = current;
2996
Jens Axboe73c10102011-03-08 13:19:51 +01002997 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01002998 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02002999 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003000
3001 /*
3002 * If this is a nested plug, don't actually assign it. It will be
3003 * flushed on its own.
3004 */
3005 if (!tsk->plug) {
3006 /*
3007 * Store ordering should not be needed here, since a potential
3008 * preempt will imply a full memory barrier
3009 */
3010 tsk->plug = plug;
3011 }
3012}
3013EXPORT_SYMBOL(blk_start_plug);
3014
3015static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3016{
3017 struct request *rqa = container_of(a, struct request, queuelist);
3018 struct request *rqb = container_of(b, struct request, queuelist);
3019
Jianpeng Ma975927b2012-10-25 21:58:17 +02003020 return !(rqa->q < rqb->q ||
3021 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003022}
3023
Jens Axboe49cac012011-04-16 13:51:05 +02003024/*
3025 * If 'from_schedule' is true, then postpone the dispatch of requests
3026 * until a safe kblockd context. We due this to avoid accidental big
3027 * additional stack usage in driver dispatch, in places where the originally
3028 * plugger did not intend it.
3029 */
Jens Axboef6603782011-04-15 15:49:07 +02003030static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003031 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003032 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003033{
Jens Axboe49cac012011-04-16 13:51:05 +02003034 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003035
Bart Van Assche70460572012-11-28 13:45:56 +01003036 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003037 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003038 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003039 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003040 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003041}
3042
NeilBrown74018dc2012-07-31 09:08:15 +02003043static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003044{
3045 LIST_HEAD(callbacks);
3046
Shaohua Li2a7d5552012-07-31 09:08:15 +02003047 while (!list_empty(&plug->cb_list)) {
3048 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003049
Shaohua Li2a7d5552012-07-31 09:08:15 +02003050 while (!list_empty(&callbacks)) {
3051 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003052 struct blk_plug_cb,
3053 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003054 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003055 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003056 }
NeilBrown048c9372011-04-18 09:52:22 +02003057 }
3058}
3059
NeilBrown9cbb1752012-07-31 09:08:14 +02003060struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3061 int size)
3062{
3063 struct blk_plug *plug = current->plug;
3064 struct blk_plug_cb *cb;
3065
3066 if (!plug)
3067 return NULL;
3068
3069 list_for_each_entry(cb, &plug->cb_list, list)
3070 if (cb->callback == unplug && cb->data == data)
3071 return cb;
3072
3073 /* Not currently on the callback list */
3074 BUG_ON(size < sizeof(*cb));
3075 cb = kzalloc(size, GFP_ATOMIC);
3076 if (cb) {
3077 cb->data = data;
3078 cb->callback = unplug;
3079 list_add(&cb->list, &plug->cb_list);
3080 }
3081 return cb;
3082}
3083EXPORT_SYMBOL(blk_check_plugged);
3084
Jens Axboe49cac012011-04-16 13:51:05 +02003085void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003086{
3087 struct request_queue *q;
3088 unsigned long flags;
3089 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003090 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003091 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003092
NeilBrown74018dc2012-07-31 09:08:15 +02003093 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003094
3095 if (!list_empty(&plug->mq_list))
3096 blk_mq_flush_plug_list(plug, from_schedule);
3097
Jens Axboe73c10102011-03-08 13:19:51 +01003098 if (list_empty(&plug->list))
3099 return;
3100
NeilBrown109b8122011-04-11 14:13:10 +02003101 list_splice_init(&plug->list, &list);
3102
Jianpeng Ma422765c2013-01-11 14:46:09 +01003103 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003104
3105 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003106 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003107
3108 /*
3109 * Save and disable interrupts here, to avoid doing it for every
3110 * queue lock we have to take.
3111 */
Jens Axboe73c10102011-03-08 13:19:51 +01003112 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003113 while (!list_empty(&list)) {
3114 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003115 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003116 BUG_ON(!rq->q);
3117 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003118 /*
3119 * This drops the queue lock
3120 */
3121 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003122 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003123 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003124 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003125 spin_lock(q->queue_lock);
3126 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003127
3128 /*
3129 * Short-circuit if @q is dead
3130 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003131 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003132 __blk_end_request_all(rq, -ENODEV);
3133 continue;
3134 }
3135
Jens Axboe73c10102011-03-08 13:19:51 +01003136 /*
3137 * rq is already accounted, so use raw insert
3138 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003139 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3140 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3141 else
3142 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003143
3144 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003145 }
3146
Jens Axboe99e22592011-04-18 09:59:55 +02003147 /*
3148 * This drops the queue lock
3149 */
3150 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003151 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003152
Jens Axboe73c10102011-03-08 13:19:51 +01003153 local_irq_restore(flags);
3154}
Jens Axboe73c10102011-03-08 13:19:51 +01003155
3156void blk_finish_plug(struct blk_plug *plug)
3157{
Jens Axboef6603782011-04-15 15:49:07 +02003158 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003159
3160 if (plug == current->plug)
3161 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003162}
3163EXPORT_SYMBOL(blk_finish_plug);
3164
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003165#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003166/**
3167 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3168 * @q: the queue of the device
3169 * @dev: the device the queue belongs to
3170 *
3171 * Description:
3172 * Initialize runtime-PM-related fields for @q and start auto suspend for
3173 * @dev. Drivers that want to take advantage of request-based runtime PM
3174 * should call this function after @dev has been initialized, and its
3175 * request queue @q has been allocated, and runtime PM for it can not happen
3176 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3177 * cases, driver should call this function before any I/O has taken place.
3178 *
3179 * This function takes care of setting up using auto suspend for the device,
3180 * the autosuspend delay is set to -1 to make runtime suspend impossible
3181 * until an updated value is either set by user or by driver. Drivers do
3182 * not need to touch other autosuspend settings.
3183 *
3184 * The block layer runtime PM is request based, so only works for drivers
3185 * that use request as their IO unit instead of those directly use bio's.
3186 */
3187void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3188{
3189 q->dev = dev;
3190 q->rpm_status = RPM_ACTIVE;
3191 pm_runtime_set_autosuspend_delay(q->dev, -1);
3192 pm_runtime_use_autosuspend(q->dev);
3193}
3194EXPORT_SYMBOL(blk_pm_runtime_init);
3195
3196/**
3197 * blk_pre_runtime_suspend - Pre runtime suspend check
3198 * @q: the queue of the device
3199 *
3200 * Description:
3201 * This function will check if runtime suspend is allowed for the device
3202 * by examining if there are any requests pending in the queue. If there
3203 * are requests pending, the device can not be runtime suspended; otherwise,
3204 * the queue's status will be updated to SUSPENDING and the driver can
3205 * proceed to suspend the device.
3206 *
3207 * For the not allowed case, we mark last busy for the device so that
3208 * runtime PM core will try to autosuspend it some time later.
3209 *
3210 * This function should be called near the start of the device's
3211 * runtime_suspend callback.
3212 *
3213 * Return:
3214 * 0 - OK to runtime suspend the device
3215 * -EBUSY - Device should not be runtime suspended
3216 */
3217int blk_pre_runtime_suspend(struct request_queue *q)
3218{
3219 int ret = 0;
3220
3221 spin_lock_irq(q->queue_lock);
3222 if (q->nr_pending) {
3223 ret = -EBUSY;
3224 pm_runtime_mark_last_busy(q->dev);
3225 } else {
3226 q->rpm_status = RPM_SUSPENDING;
3227 }
3228 spin_unlock_irq(q->queue_lock);
3229 return ret;
3230}
3231EXPORT_SYMBOL(blk_pre_runtime_suspend);
3232
3233/**
3234 * blk_post_runtime_suspend - Post runtime suspend processing
3235 * @q: the queue of the device
3236 * @err: return value of the device's runtime_suspend function
3237 *
3238 * Description:
3239 * Update the queue's runtime status according to the return value of the
3240 * device's runtime suspend function and mark last busy for the device so
3241 * that PM core will try to auto suspend the device at a later time.
3242 *
3243 * This function should be called near the end of the device's
3244 * runtime_suspend callback.
3245 */
3246void blk_post_runtime_suspend(struct request_queue *q, int err)
3247{
3248 spin_lock_irq(q->queue_lock);
3249 if (!err) {
3250 q->rpm_status = RPM_SUSPENDED;
3251 } else {
3252 q->rpm_status = RPM_ACTIVE;
3253 pm_runtime_mark_last_busy(q->dev);
3254 }
3255 spin_unlock_irq(q->queue_lock);
3256}
3257EXPORT_SYMBOL(blk_post_runtime_suspend);
3258
3259/**
3260 * blk_pre_runtime_resume - Pre runtime resume processing
3261 * @q: the queue of the device
3262 *
3263 * Description:
3264 * Update the queue's runtime status to RESUMING in preparation for the
3265 * runtime resume of the device.
3266 *
3267 * This function should be called near the start of the device's
3268 * runtime_resume callback.
3269 */
3270void blk_pre_runtime_resume(struct request_queue *q)
3271{
3272 spin_lock_irq(q->queue_lock);
3273 q->rpm_status = RPM_RESUMING;
3274 spin_unlock_irq(q->queue_lock);
3275}
3276EXPORT_SYMBOL(blk_pre_runtime_resume);
3277
3278/**
3279 * blk_post_runtime_resume - Post runtime resume processing
3280 * @q: the queue of the device
3281 * @err: return value of the device's runtime_resume function
3282 *
3283 * Description:
3284 * Update the queue's runtime status according to the return value of the
3285 * device's runtime_resume function. If it is successfully resumed, process
3286 * the requests that are queued into the device's queue when it is resuming
3287 * and then mark last busy and initiate autosuspend for it.
3288 *
3289 * This function should be called near the end of the device's
3290 * runtime_resume callback.
3291 */
3292void blk_post_runtime_resume(struct request_queue *q, int err)
3293{
3294 spin_lock_irq(q->queue_lock);
3295 if (!err) {
3296 q->rpm_status = RPM_ACTIVE;
3297 __blk_run_queue(q);
3298 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003299 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003300 } else {
3301 q->rpm_status = RPM_SUSPENDED;
3302 }
3303 spin_unlock_irq(q->queue_lock);
3304}
3305EXPORT_SYMBOL(blk_post_runtime_resume);
3306#endif
3307
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308int __init blk_dev_init(void)
3309{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003310 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3311 sizeof(((struct request *)0)->cmd_flags));
3312
Tejun Heo89b90be2011-01-03 15:01:47 +01003313 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3314 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003315 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316 if (!kblockd_workqueue)
3317 panic("Failed to create kblockd\n");
3318
3319 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003320 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321
Jens Axboe8324aa92008-01-29 14:51:59 +01003322 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003323 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 return 0;
3326}