blob: 794c3e7f01cf745a607ca3ea26d46c7bd69952c6 [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
Jens Axboeaed3ea92014-12-22 14:04:42 -0700476void blk_set_queue_dying(struct request_queue *q)
477{
478 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
479
480 if (q->mq_ops)
481 blk_mq_wake_waiters(q);
482 else {
483 struct request_list *rl;
484
485 blk_queue_for_each_rl(rl, q) {
486 if (rl->rq_pool) {
487 wake_up(&rl->wait[BLK_RW_SYNC]);
488 wake_up(&rl->wait[BLK_RW_ASYNC]);
489 }
490 }
491 }
492}
493EXPORT_SYMBOL_GPL(blk_set_queue_dying);
494
Tejun Heod7325802012-03-05 13:14:58 -0800495/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200496 * blk_cleanup_queue - shutdown a request queue
497 * @q: request queue to shutdown
498 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100499 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
500 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500501 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100502void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500503{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200504 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700505
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100506 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500507 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700508 blk_set_queue_dying(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200509 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800510
Tejun Heo80fd9972012-04-13 14:50:53 -0700511 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100512 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700513 * that, unlike blk_queue_bypass_start(), we aren't performing
514 * synchronize_rcu() after entering bypass mode to avoid the delay
515 * as some drivers create and destroy a lot of queues while
516 * probing. This is still safe because blk_release_queue() will be
517 * called only after the queue refcnt drops to zero and nothing,
518 * RCU or not, would be traversing the queue by then.
519 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800520 q->bypass_depth++;
521 queue_flag_set(QUEUE_FLAG_BYPASS, q);
522
Tejun Heoc9a929d2011-10-19 14:42:16 +0200523 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
524 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100525 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200526 spin_unlock_irq(lock);
527 mutex_unlock(&q->sysfs_lock);
528
Bart Van Asschec246e802012-12-06 14:32:01 +0100529 /*
530 * Drain all requests queued before DYING marking. Set DEAD flag to
531 * prevent that q->request_fn() gets invoked after draining finished.
532 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800533 if (q->mq_ops) {
Tejun Heo780db202014-07-01 10:31:13 -0600534 blk_mq_freeze_queue(q);
Ming Lei43a5e4e2013-12-26 21:31:35 +0800535 spin_lock_irq(lock);
536 } else {
537 spin_lock_irq(lock);
538 __blk_drain_queue(q, true);
539 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100540 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100541 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200542
543 /* @q won't process any more request, flush async actions */
544 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
545 blk_sync_queue(q);
546
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100547 if (q->mq_ops)
548 blk_mq_free_queue(q);
549
Asias He5e5cfac2012-05-24 23:28:52 +0800550 spin_lock_irq(lock);
551 if (q->queue_lock != &q->__queue_lock)
552 q->queue_lock = &q->__queue_lock;
553 spin_unlock_irq(lock);
554
Tejun Heoc9a929d2011-10-19 14:42:16 +0200555 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500556 blk_put_queue(q);
557}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558EXPORT_SYMBOL(blk_cleanup_queue);
559
Tejun Heo5b788ce2012-06-04 20:40:59 -0700560int blk_init_rl(struct request_list *rl, struct request_queue *q,
561 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400563 if (unlikely(rl->rq_pool))
564 return 0;
565
Tejun Heo5b788ce2012-06-04 20:40:59 -0700566 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200567 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
568 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200569 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
570 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Christoph Lameter19460892005-06-23 00:08:19 -0700572 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700573 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700574 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (!rl->rq_pool)
576 return -ENOMEM;
577
578 return 0;
579}
580
Tejun Heo5b788ce2012-06-04 20:40:59 -0700581void blk_exit_rl(struct request_list *rl)
582{
583 if (rl->rq_pool)
584 mempool_destroy(rl->rq_pool);
585}
586
Jens Axboe165125e2007-07-24 09:28:11 +0200587struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100589 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700590}
591EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Jens Axboe165125e2007-07-24 09:28:11 +0200593struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700594{
Jens Axboe165125e2007-07-24 09:28:11 +0200595 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700596 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700597
Jens Axboe8324aa92008-01-29 14:51:59 +0100598 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700599 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (!q)
601 return NULL;
602
Dan Carpenter00380a42012-03-23 09:58:54 +0100603 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100604 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800605 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100606
Jens Axboe0989a022009-06-12 14:42:56 +0200607 q->backing_dev_info.ra_pages =
608 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
609 q->backing_dev_info.state = 0;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100610 q->backing_dev_info.capabilities = 0;
Jens Axboed9938312009-06-12 14:45:52 +0200611 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100612 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200613
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700614 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100615 if (err)
616 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700617
Matthew Garrett31373d02010-04-06 14:25:14 +0200618 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
619 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700620 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100621 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700622 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100623 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800624#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800625 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800626#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500627 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500628
Jens Axboe8324aa92008-01-29 14:51:59 +0100629 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Al Viro483f4af2006-03-18 18:34:37 -0500631 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700632 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500633
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500634 /*
635 * By default initialize queue_lock to internal lock and driver can
636 * override it later if need be.
637 */
638 q->queue_lock = &q->__queue_lock;
639
Tejun Heob82d4b12012-04-13 13:11:31 -0700640 /*
641 * A queue starts its life with bypass turned on to avoid
642 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700643 * init. The initial bypass will be finished when the queue is
644 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700645 */
646 q->bypass_depth = 1;
647 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
648
Jens Axboe320ae512013-10-24 09:20:05 +0100649 init_waitqueue_head(&q->mq_freeze_wq);
650
Tejun Heo5efd6112012-03-05 13:15:12 -0800651 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400652 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100655
Mikulas Patockafff49962013-10-14 12:11:36 -0400656fail_bdi:
657 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100658fail_id:
659 ida_simple_remove(&blk_queue_ida, q->id);
660fail_q:
661 kmem_cache_free(blk_requestq_cachep, q);
662 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
Christoph Lameter19460892005-06-23 00:08:19 -0700664EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666/**
667 * blk_init_queue - prepare a request queue for use with a block device
668 * @rfn: The function to be called to process requests that have been
669 * placed on the queue.
670 * @lock: Request queue spin lock
671 *
672 * Description:
673 * If a block device wishes to use the standard request handling procedures,
674 * which sorts requests and coalesces adjacent requests, then it must
675 * call blk_init_queue(). The function @rfn will be called when there
676 * are requests on the queue that need to be processed. If the device
677 * supports plugging, then @rfn may not be called immediately when requests
678 * are available on the queue, but may be called at some time later instead.
679 * Plugged queues are generally unplugged when a buffer belonging to one
680 * of the requests on the queue is needed, or due to memory pressure.
681 *
682 * @rfn is not required, or even expected, to remove all requests off the
683 * queue, but only as many as it can handle at a time. If it does leave
684 * requests on the queue, it is responsible for arranging that the requests
685 * get dealt with eventually.
686 *
687 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200688 * request queue; this lock will be taken also from interrupt context, so irq
689 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200691 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 * it didn't succeed.
693 *
694 * Note:
695 * blk_init_queue() must be paired with a blk_cleanup_queue() call
696 * when the block device is deactivated (such as at module unload).
697 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700698
Jens Axboe165125e2007-07-24 09:28:11 +0200699struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100701 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700702}
703EXPORT_SYMBOL(blk_init_queue);
704
Jens Axboe165125e2007-07-24 09:28:11 +0200705struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700706blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
707{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600708 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600710 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
711 if (!uninit_q)
712 return NULL;
713
Mike Snitzer51514122011-11-23 10:59:13 +0100714 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600715 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700716 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700717
Mike Snitzer7982e902014-03-08 17:20:01 -0700718 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200719}
720EXPORT_SYMBOL(blk_init_queue_node);
721
722struct request_queue *
723blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
724 spinlock_t *lock)
725{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (!q)
727 return NULL;
728
Ming Leif70ced02014-09-25 23:23:47 +0800729 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800730 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700731 return NULL;
732
Tejun Heoa0516612012-06-26 15:05:44 -0700733 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600734 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900738 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700739 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500740
741 /* Override internal queue lock with supplied lock pointer */
742 if (lock)
743 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Jens Axboef3b144a2009-03-06 08:48:33 +0100745 /*
746 * This also sets hw/phys segments, boundary and size
747 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200748 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Alan Stern44ec9542007-02-20 11:01:57 -0500750 q->sg_reserved_size = INT_MAX;
751
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600752 /* Protect q->elevator from elevator_change */
753 mutex_lock(&q->sysfs_lock);
754
Tejun Heob82d4b12012-04-13 13:11:31 -0700755 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600756 if (elevator_init(q, NULL)) {
757 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600758 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600759 }
760
761 mutex_unlock(&q->sysfs_lock);
762
Tejun Heob82d4b12012-04-13 13:11:31 -0700763 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600764
765fail:
Ming Leiba483382014-09-25 23:23:44 +0800766 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600767 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
Mike Snitzer51514122011-11-23 10:59:13 +0100769EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Tejun Heo09ac46c2011-12-14 00:33:38 +0100771bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100773 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100774 __blk_get_queue(q);
775 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
777
Tejun Heo09ac46c2011-12-14 00:33:38 +0100778 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}
Jens Axboed86e0e82011-05-27 07:44:43 +0200780EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Tejun Heo5b788ce2012-06-04 20:40:59 -0700782static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100784 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700785 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100786 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100787 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100788 }
789
Tejun Heo5b788ce2012-06-04 20:40:59 -0700790 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793/*
794 * ioc_batching returns true if the ioc is a valid batching request and
795 * should be given priority access to a request.
796 */
Jens Axboe165125e2007-07-24 09:28:11 +0200797static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
799 if (!ioc)
800 return 0;
801
802 /*
803 * Make sure the process is able to allocate at least 1 request
804 * even if the batch times out, otherwise we could theoretically
805 * lose wakeups.
806 */
807 return ioc->nr_batch_requests == q->nr_batching ||
808 (ioc->nr_batch_requests > 0
809 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
810}
811
812/*
813 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
814 * will cause the process to be a "batcher" on all queues in the system. This
815 * is the behaviour we want though - once it gets a wakeup it should be given
816 * a nice run.
817 */
Jens Axboe165125e2007-07-24 09:28:11 +0200818static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
820 if (!ioc || ioc_batching(q, ioc))
821 return;
822
823 ioc->nr_batch_requests = q->nr_batching;
824 ioc->last_waited = jiffies;
825}
826
Tejun Heo5b788ce2012-06-04 20:40:59 -0700827static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700829 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Tejun Heoa0516612012-06-26 15:05:44 -0700831 /*
832 * bdi isn't aware of blkcg yet. As all async IOs end up root
833 * blkcg anyway, just use root blkcg state.
834 */
835 if (rl == &q->root_rl &&
836 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200837 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Jens Axboe1faa16d2009-04-06 14:48:01 +0200839 if (rl->count[sync] + 1 <= q->nr_requests) {
840 if (waitqueue_active(&rl->wait[sync]))
841 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
Tejun Heo5b788ce2012-06-04 20:40:59 -0700843 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 }
845}
846
847/*
848 * A request has just been released. Account for it, update the full and
849 * congestion status, wake up any waiters. Called under q->queue_lock.
850 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700851static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700853 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200854 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700856 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200857 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200858 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700859 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Tejun Heo5b788ce2012-06-04 20:40:59 -0700861 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Jens Axboe1faa16d2009-04-06 14:48:01 +0200863 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700864 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600867int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
868{
869 struct request_list *rl;
870
871 spin_lock_irq(q->queue_lock);
872 q->nr_requests = nr;
873 blk_queue_congestion_threshold(q);
874
875 /* congestion isn't cgroup aware and follows root blkcg for now */
876 rl = &q->root_rl;
877
878 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
879 blk_set_queue_congested(q, BLK_RW_SYNC);
880 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
881 blk_clear_queue_congested(q, BLK_RW_SYNC);
882
883 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
884 blk_set_queue_congested(q, BLK_RW_ASYNC);
885 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
886 blk_clear_queue_congested(q, BLK_RW_ASYNC);
887
888 blk_queue_for_each_rl(rl, q) {
889 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
890 blk_set_rl_full(rl, BLK_RW_SYNC);
891 } else {
892 blk_clear_rl_full(rl, BLK_RW_SYNC);
893 wake_up(&rl->wait[BLK_RW_SYNC]);
894 }
895
896 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
897 blk_set_rl_full(rl, BLK_RW_ASYNC);
898 } else {
899 blk_clear_rl_full(rl, BLK_RW_ASYNC);
900 wake_up(&rl->wait[BLK_RW_ASYNC]);
901 }
902 }
903
904 spin_unlock_irq(q->queue_lock);
905 return 0;
906}
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100909 * Determine if elevator data should be initialized when allocating the
910 * request associated with @bio.
911 */
912static bool blk_rq_should_init_elevator(struct bio *bio)
913{
914 if (!bio)
915 return true;
916
917 /*
918 * Flush requests do not use the elevator so skip initialization.
919 * This allows a request to share the flush and elevator data.
920 */
921 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
922 return false;
923
924 return true;
925}
926
Tejun Heoda8303c2011-10-19 14:33:05 +0200927/**
Tejun Heo852c7882012-03-05 13:15:27 -0800928 * rq_ioc - determine io_context for request allocation
929 * @bio: request being allocated is for this bio (can be %NULL)
930 *
931 * Determine io_context to use for request allocation for @bio. May return
932 * %NULL if %current->io_context doesn't exist.
933 */
934static struct io_context *rq_ioc(struct bio *bio)
935{
936#ifdef CONFIG_BLK_CGROUP
937 if (bio && bio->bi_ioc)
938 return bio->bi_ioc;
939#endif
940 return current->io_context;
941}
942
943/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700944 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700945 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200946 * @rw_flags: RW and SYNC flags
947 * @bio: bio to allocate request for (can be %NULL)
948 * @gfp_mask: allocation mask
949 *
950 * Get a free request from @q. This function may fail under memory
951 * pressure or if @q is dead.
952 *
Masanari Iidada3dae52014-09-09 01:27:23 +0900953 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -0600954 * Returns ERR_PTR on failure, with @q->queue_lock held.
955 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700957static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700958 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700960 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800961 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700962 struct elevator_type *et = q->elevator->type;
963 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100964 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200965 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200966 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100968 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -0600969 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +0200970
Jens Axboe7749a8d2006-12-13 13:02:26 +0100971 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100972 if (may_queue == ELV_MQUEUE_NO)
973 goto rq_starved;
974
Jens Axboe1faa16d2009-04-06 14:48:01 +0200975 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
976 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100977 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100978 * The queue will fill after this allocation, so set
979 * it as full, and mark this process as "batching".
980 * This process will be allowed to complete a batch of
981 * requests, others will be blocked.
982 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700983 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100984 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700985 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100986 } else {
987 if (may_queue != ELV_MQUEUE_MUST
988 && !ioc_batching(q, ioc)) {
989 /*
990 * The queue is full and the allocating
991 * process is not a "batcher", and not
992 * exempted by the IO scheduler
993 */
Joe Lawrencea492f072014-08-28 08:15:21 -0600994 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100995 }
996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 }
Tejun Heoa0516612012-06-26 15:05:44 -0700998 /*
999 * bdi isn't aware of blkcg yet. As all async IOs end up
1000 * root blkcg anyway, just use root blkcg state.
1001 */
1002 if (rl == &q->root_rl)
1003 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005
Jens Axboe082cf692005-06-28 16:35:11 +02001006 /*
1007 * Only allow batching queuers to allocate up to 50% over the defined
1008 * limit of requests, otherwise we could have thousands of requests
1009 * allocated with any setting of ->nr_requests
1010 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001011 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001012 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001013
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001014 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001015 rl->count[is_sync]++;
1016 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001017
Tejun Heof1f8cc92011-12-14 00:33:42 +01001018 /*
1019 * Decide whether the new request will be managed by elevator. If
1020 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1021 * prevent the current elevator from being destroyed until the new
1022 * request is freed. This guarantees icq's won't be destroyed and
1023 * makes creating new ones safe.
1024 *
1025 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1026 * it will be created after releasing queue_lock.
1027 */
Tejun Heod7325802012-03-05 13:14:58 -08001028 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001029 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001030 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001031 if (et->icq_cache && ioc)
1032 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001033 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001034
Jens Axboef253b862010-10-24 22:06:02 +02001035 if (blk_queue_io_stat(q))
1036 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 spin_unlock_irq(q->queue_lock);
1038
Tejun Heo29e2b092012-04-19 16:29:21 -07001039 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001040 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001041 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001042 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Tejun Heo29e2b092012-04-19 16:29:21 -07001044 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001045 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001046 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1047
Tejun Heoaaf7c682012-04-19 16:29:22 -07001048 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001049 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001050 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001051 if (ioc)
1052 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001053 if (!icq)
1054 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001055 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001056
1057 rq->elv.icq = icq;
1058 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1059 goto fail_elvpriv;
1060
1061 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001062 if (icq)
1063 get_io_context(icq->ioc);
1064 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001065out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001066 /*
1067 * ioc may be NULL here, and ioc_batching will be false. That's
1068 * OK, if the queue is under the request limit then requests need
1069 * not count toward the nr_batch_requests limit. There will always
1070 * be some limit enforced by BLK_BATCH_TIME.
1071 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (ioc_batching(q, ioc))
1073 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001074
Jens Axboe1faa16d2009-04-06 14:48:01 +02001075 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001077
Tejun Heoaaf7c682012-04-19 16:29:22 -07001078fail_elvpriv:
1079 /*
1080 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1081 * and may fail indefinitely under memory pressure and thus
1082 * shouldn't stall IO. Treat this request as !elvpriv. This will
1083 * disturb iosched and blkcg but weird is bettern than dead.
1084 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001085 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1086 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001087
1088 rq->cmd_flags &= ~REQ_ELVPRIV;
1089 rq->elv.icq = NULL;
1090
1091 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001092 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001093 spin_unlock_irq(q->queue_lock);
1094 goto out;
1095
Tejun Heob6792812012-03-05 13:15:23 -08001096fail_alloc:
1097 /*
1098 * Allocation failed presumably due to memory. Undo anything we
1099 * might have messed up.
1100 *
1101 * Allocating task should really be put onto the front of the wait
1102 * queue, but this is pretty rare.
1103 */
1104 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001105 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001106
1107 /*
1108 * in the very unlikely event that allocation failed and no
1109 * requests for this direction was pending, mark us starved so that
1110 * freeing of a request in the other direction will notice
1111 * us. another possible fix would be to split the rq mempool into
1112 * READ and WRITE
1113 */
1114rq_starved:
1115 if (unlikely(rl->count[is_sync] == 0))
1116 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001117 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
Tejun Heoda8303c2011-10-19 14:33:05 +02001120/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001121 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001122 * @q: request_queue to allocate request from
1123 * @rw_flags: RW and SYNC flags
1124 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001125 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001126 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001127 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1128 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001129 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001130 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001131 * Returns ERR_PTR on failure, with @q->queue_lock held.
1132 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001134static struct request *get_request(struct request_queue *q, int rw_flags,
1135 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001137 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001138 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001139 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001141
1142 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001143retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001144 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001145 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001146 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001148 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001149 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001150 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Tejun Heoa06e05e2012-06-04 20:40:55 -07001153 /* wait on @rl and retry */
1154 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1155 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001156
Tejun Heoa06e05e2012-06-04 20:40:55 -07001157 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Tejun Heoa06e05e2012-06-04 20:40:55 -07001159 spin_unlock_irq(q->queue_lock);
1160 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Tejun Heoa06e05e2012-06-04 20:40:55 -07001162 /*
1163 * After sleeping, we become a "batching" process and will be able
1164 * to allocate at least one request, and up to a big batch of them
1165 * for a small period time. See ioc_batching, ioc_set_batching
1166 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001167 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Tejun Heoa06e05e2012-06-04 20:40:55 -07001169 spin_lock_irq(q->queue_lock);
1170 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001171
Tejun Heoa06e05e2012-06-04 20:40:55 -07001172 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
Jens Axboe320ae512013-10-24 09:20:05 +01001175static struct request *blk_old_get_request(struct request_queue *q, int rw,
1176 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
1178 struct request *rq;
1179
1180 BUG_ON(rw != READ && rw != WRITE);
1181
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001182 /* create ioc upfront */
1183 create_io_context(gfp_mask, q->node);
1184
Nick Piggind6344532005-06-28 20:45:14 -07001185 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001186 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001187 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001188 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001189 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 return rq;
1192}
Jens Axboe320ae512013-10-24 09:20:05 +01001193
1194struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1195{
1196 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001197 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001198 else
1199 return blk_old_get_request(q, rw, gfp_mask);
1200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201EXPORT_SYMBOL(blk_get_request);
1202
1203/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001204 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001205 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001206 * @bio: The bio describing the memory mappings that will be submitted for IO.
1207 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001208 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001209 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001210 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1211 * type commands. Where the struct request needs to be farther initialized by
1212 * the caller. It is passed a &struct bio, which describes the memory info of
1213 * the I/O transfer.
1214 *
1215 * The caller of blk_make_request must make sure that bi_io_vec
1216 * are set to describe the memory buffers. That bio_data_dir() will return
1217 * the needed direction of the request. (And all bio's in the passed bio-chain
1218 * are properly set accordingly)
1219 *
1220 * If called under none-sleepable conditions, mapped bio buffers must not
1221 * need bouncing, by calling the appropriate masked or flagged allocator,
1222 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1223 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001224 *
1225 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1226 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1227 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1228 * completion of a bio that hasn't been submitted yet, thus resulting in a
1229 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1230 * of bio_alloc(), as that avoids the mempool deadlock.
1231 * If possible a big IO should be split into smaller parts when allocation
1232 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001233 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001234struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1235 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001236{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001237 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1238
Joe Lawrencea492f072014-08-28 08:15:21 -06001239 if (IS_ERR(rq))
1240 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001241
Jens Axboef27b0872014-06-06 07:57:37 -06001242 blk_rq_set_block_pc(rq);
1243
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001244 for_each_bio(bio) {
1245 struct bio *bounce_bio = bio;
1246 int ret;
1247
1248 blk_queue_bounce(q, &bounce_bio);
1249 ret = blk_rq_append_bio(q, rq, bounce_bio);
1250 if (unlikely(ret)) {
1251 blk_put_request(rq);
1252 return ERR_PTR(ret);
1253 }
1254 }
1255
1256 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001257}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001258EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001259
1260/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001261 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001262 * @rq: request to be initialized
1263 *
1264 */
1265void blk_rq_set_block_pc(struct request *rq)
1266{
1267 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1268 rq->__data_len = 0;
1269 rq->__sector = (sector_t) -1;
1270 rq->bio = rq->biotail = NULL;
1271 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001272}
1273EXPORT_SYMBOL(blk_rq_set_block_pc);
1274
1275/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 * blk_requeue_request - put a request back on queue
1277 * @q: request queue where request should be inserted
1278 * @rq: request to be inserted
1279 *
1280 * Description:
1281 * Drivers often keep queueing requests until the hardware cannot accept
1282 * more, when that condition happens we need to put the request back
1283 * on the queue. Must be called with queue lock held.
1284 */
Jens Axboe165125e2007-07-24 09:28:11 +02001285void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001287 blk_delete_timer(rq);
1288 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001289 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001290
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001291 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 blk_queue_end_tag(q, rq);
1293
James Bottomleyba396a62009-05-27 14:17:08 +02001294 BUG_ON(blk_queued_rq(rq));
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 elv_requeue_request(q, rq);
1297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298EXPORT_SYMBOL(blk_requeue_request);
1299
Jens Axboe73c10102011-03-08 13:19:51 +01001300static void add_acct_request(struct request_queue *q, struct request *rq,
1301 int where)
1302{
Jens Axboe320ae512013-10-24 09:20:05 +01001303 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001304 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001305}
1306
Tejun Heo074a7ac2008-08-25 19:56:14 +09001307static void part_round_stats_single(int cpu, struct hd_struct *part,
1308 unsigned long now)
1309{
Jens Axboe7276d022014-05-09 15:48:23 -06001310 int inflight;
1311
Tejun Heo074a7ac2008-08-25 19:56:14 +09001312 if (now == part->stamp)
1313 return;
1314
Jens Axboe7276d022014-05-09 15:48:23 -06001315 inflight = part_in_flight(part);
1316 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001317 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001318 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001319 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1320 }
1321 part->stamp = now;
1322}
1323
1324/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001325 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1326 * @cpu: cpu number for stats access
1327 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 *
1329 * The average IO queue length and utilisation statistics are maintained
1330 * by observing the current state of the queue length and the amount of
1331 * time it has been in this state for.
1332 *
1333 * Normally, that accounting is done on IO completion, but that can result
1334 * in more than a second's worth of IO being accounted for within any one
1335 * second, leading to >100% utilisation. To deal with that, we call this
1336 * function to do a round-off before returning the results when reading
1337 * /proc/diskstats. This accounts immediately for all queue usage up to
1338 * the current jiffies and restarts the counters again.
1339 */
Tejun Heoc9959052008-08-25 19:47:21 +09001340void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001341{
1342 unsigned long now = jiffies;
1343
Tejun Heo074a7ac2008-08-25 19:56:14 +09001344 if (part->partno)
1345 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1346 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001347}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001348EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001349
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001350#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001351static void blk_pm_put_request(struct request *rq)
1352{
1353 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1354 pm_runtime_mark_last_busy(rq->q->dev);
1355}
1356#else
1357static inline void blk_pm_put_request(struct request *rq) {}
1358#endif
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/*
1361 * queue lock must be held
1362 */
Jens Axboe165125e2007-07-24 09:28:11 +02001363void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (unlikely(!q))
1366 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001368 if (q->mq_ops) {
1369 blk_mq_free_request(req);
1370 return;
1371 }
1372
Lin Mingc8158812013-03-23 11:42:27 +08001373 blk_pm_put_request(req);
1374
Tejun Heo8922e162005-10-20 16:23:44 +02001375 elv_completed_request(q, req);
1376
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001377 /* this is a bio leak */
1378 WARN_ON(req->bio != NULL);
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 /*
1381 * Request may not have originated from ll_rw_blk. if not,
1382 * it didn't come out of our reserved rq pools
1383 */
Jens Axboe49171e52006-08-10 08:59:11 +02001384 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001385 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001386 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001389 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Tejun Heoa0516612012-06-26 15:05:44 -07001391 blk_free_request(rl, req);
1392 freed_request(rl, flags);
1393 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 }
1395}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001396EXPORT_SYMBOL_GPL(__blk_put_request);
1397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398void blk_put_request(struct request *req)
1399{
Jens Axboe165125e2007-07-24 09:28:11 +02001400 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Jens Axboe320ae512013-10-24 09:20:05 +01001402 if (q->mq_ops)
1403 blk_mq_free_request(req);
1404 else {
1405 unsigned long flags;
1406
1407 spin_lock_irqsave(q->queue_lock, flags);
1408 __blk_put_request(q, req);
1409 spin_unlock_irqrestore(q->queue_lock, flags);
1410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412EXPORT_SYMBOL(blk_put_request);
1413
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001414/**
1415 * blk_add_request_payload - add a payload to a request
1416 * @rq: request to update
1417 * @page: page backing the payload
1418 * @len: length of the payload.
1419 *
1420 * This allows to later add a payload to an already submitted request by
1421 * a block driver. The driver needs to take care of freeing the payload
1422 * itself.
1423 *
1424 * Note that this is a quite horrible hack and nothing but handling of
1425 * discard requests should ever use it.
1426 */
1427void blk_add_request_payload(struct request *rq, struct page *page,
1428 unsigned int len)
1429{
1430 struct bio *bio = rq->bio;
1431
1432 bio->bi_io_vec->bv_page = page;
1433 bio->bi_io_vec->bv_offset = 0;
1434 bio->bi_io_vec->bv_len = len;
1435
Kent Overstreet4f024f32013-10-11 15:44:27 -07001436 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001437 bio->bi_vcnt = 1;
1438 bio->bi_phys_segments = 1;
1439
1440 rq->__data_len = rq->resid_len = len;
1441 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001442}
1443EXPORT_SYMBOL_GPL(blk_add_request_payload);
1444
Jens Axboe320ae512013-10-24 09:20:05 +01001445bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1446 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001447{
1448 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1449
Jens Axboe73c10102011-03-08 13:19:51 +01001450 if (!ll_back_merge_fn(q, req, bio))
1451 return false;
1452
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001453 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001454
1455 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1456 blk_rq_set_mixed_merge(req);
1457
1458 req->biotail->bi_next = bio;
1459 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001460 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001461 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1462
Jens Axboe320ae512013-10-24 09:20:05 +01001463 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001464 return true;
1465}
1466
Jens Axboe320ae512013-10-24 09:20:05 +01001467bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1468 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001469{
1470 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001471
Jens Axboe73c10102011-03-08 13:19:51 +01001472 if (!ll_front_merge_fn(q, req, bio))
1473 return false;
1474
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001475 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001476
1477 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1478 blk_rq_set_mixed_merge(req);
1479
Jens Axboe73c10102011-03-08 13:19:51 +01001480 bio->bi_next = req->bio;
1481 req->bio = bio;
1482
Kent Overstreet4f024f32013-10-11 15:44:27 -07001483 req->__sector = bio->bi_iter.bi_sector;
1484 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001485 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1486
Jens Axboe320ae512013-10-24 09:20:05 +01001487 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001488 return true;
1489}
1490
Tejun Heobd87b582011-10-19 14:33:08 +02001491/**
Jens Axboe320ae512013-10-24 09:20:05 +01001492 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001493 * @q: request_queue new bio is being queued at
1494 * @bio: new bio being queued
1495 * @request_count: out parameter for number of traversed plugged requests
1496 *
1497 * Determine whether @bio being queued on @q can be merged with a request
1498 * on %current's plugged list. Returns %true if merge was successful,
1499 * otherwise %false.
1500 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001501 * Plugging coalesces IOs from the same issuer for the same purpose without
1502 * going through @q->queue_lock. As such it's more of an issuing mechanism
1503 * than scheduling, and the request, while may have elvpriv data, is not
1504 * added on the elevator at this point. In addition, we don't have
1505 * reliable access to the elevator outside queue lock. Only check basic
1506 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001507 *
1508 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001509 */
Jens Axboe320ae512013-10-24 09:20:05 +01001510bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1511 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001512{
1513 struct blk_plug *plug;
1514 struct request *rq;
1515 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001516 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001517
Tejun Heobd87b582011-10-19 14:33:08 +02001518 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001519 if (!plug)
1520 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001521 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001522
Shaohua Li92f399c2013-10-29 12:01:03 -06001523 if (q->mq_ops)
1524 plug_list = &plug->mq_list;
1525 else
1526 plug_list = &plug->list;
1527
1528 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001529 int el_ret;
1530
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001531 if (rq->q == q)
1532 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001533
Tejun Heo07c2bd32012-02-08 09:19:42 +01001534 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001535 continue;
1536
Tejun Heo050c8ea2012-02-08 09:19:38 +01001537 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001538 if (el_ret == ELEVATOR_BACK_MERGE) {
1539 ret = bio_attempt_back_merge(q, rq, bio);
1540 if (ret)
1541 break;
1542 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1543 ret = bio_attempt_front_merge(q, rq, bio);
1544 if (ret)
1545 break;
1546 }
1547 }
1548out:
1549 return ret;
1550}
1551
Jens Axboe86db1e22008-01-29 14:53:40 +01001552void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001553{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001554 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001555
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001556 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1557 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001558 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001559
Tejun Heo52d9e672006-01-06 09:49:58 +01001560 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001561 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001562 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001563 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001564}
1565
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001566void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001568 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001569 struct blk_plug *plug;
1570 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1571 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001572 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 /*
1575 * low level driver can indicate that it wants pages above a
1576 * certain limit bounced to low memory (ie for highmem, or even
1577 * ISA dma in theory)
1578 */
1579 blk_queue_bounce(q, &bio);
1580
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001581 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1582 bio_endio(bio, -EIO);
1583 return;
1584 }
1585
Tejun Heo4fed9472010-09-03 11:56:17 +02001586 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001587 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001588 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001589 goto get_rq;
1590 }
1591
Jens Axboe73c10102011-03-08 13:19:51 +01001592 /*
1593 * Check if we can merge with the plugged list before grabbing
1594 * any locks.
1595 */
Robert Elliottda41a582014-05-20 16:46:26 -05001596 if (!blk_queue_nomerges(q) &&
1597 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001598 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001599
1600 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001603 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001604 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001605 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001606 if (!attempt_back_merge(q, req))
1607 elv_merged_request(q, req, el_ret);
1608 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001609 }
Jens Axboe73c10102011-03-08 13:19:51 +01001610 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001611 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001612 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001613 if (!attempt_front_merge(q, req))
1614 elv_merged_request(q, req, el_ret);
1615 goto out_unlock;
1616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 }
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001620 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001621 * This sync check and mask will be re-done in init_request_from_bio(),
1622 * but we need to set it earlier to expose the sync flag to the
1623 * rq allocator and io schedulers.
1624 */
1625 rw_flags = bio_data_dir(bio);
1626 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001627 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001628
1629 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001630 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001631 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001632 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001633 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001634 if (IS_ERR(req)) {
1635 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
Tejun Heoda8303c2011-10-19 14:33:05 +02001636 goto out_unlock;
1637 }
Nick Piggind6344532005-06-28 20:45:14 -07001638
Nick Piggin450991b2005-06-28 20:45:13 -07001639 /*
1640 * After dropping the lock and possibly sleeping here, our request
1641 * may now be mergeable after it had proven unmergeable (above).
1642 * We don't worry about that case for efficiency. It won't happen
1643 * often, and the elevators are able to handle it.
1644 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001645 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Tao Ma9562ad92011-10-24 16:11:30 +02001647 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001648 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001649
Jens Axboe73c10102011-03-08 13:19:51 +01001650 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001651 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001652 /*
1653 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001654 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001655 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001656 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001657 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001658 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001659 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001660 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001661 trace_block_plug(q);
1662 }
Jens Axboe73c10102011-03-08 13:19:51 +01001663 }
Shaohua Lia6327162011-08-24 16:04:32 +02001664 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001665 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001666 } else {
1667 spin_lock_irq(q->queue_lock);
1668 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001669 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001670out_unlock:
1671 spin_unlock_irq(q->queue_lock);
1672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
Jens Axboec20e8de2011-09-12 12:03:37 +02001674EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676/*
1677 * If bio->bi_dev is a partition, remap the location
1678 */
1679static inline void blk_partition_remap(struct bio *bio)
1680{
1681 struct block_device *bdev = bio->bi_bdev;
1682
Jens Axboebf2de6f2007-09-27 13:01:25 +02001683 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001685
Kent Overstreet4f024f32013-10-11 15:44:27 -07001686 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001688
Mike Snitzerd07335e2010-11-16 12:52:38 +01001689 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1690 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001691 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
1693}
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695static void handle_bad_sector(struct bio *bio)
1696{
1697 char b[BDEVNAME_SIZE];
1698
1699 printk(KERN_INFO "attempt to access beyond end of device\n");
1700 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1701 bdevname(bio->bi_bdev, b),
1702 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001703 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001704 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
1706 set_bit(BIO_EOF, &bio->bi_flags);
1707}
1708
Akinobu Mitac17bb492006-12-08 02:39:46 -08001709#ifdef CONFIG_FAIL_MAKE_REQUEST
1710
1711static DECLARE_FAULT_ATTR(fail_make_request);
1712
1713static int __init setup_fail_make_request(char *str)
1714{
1715 return setup_fault_attr(&fail_make_request, str);
1716}
1717__setup("fail_make_request=", setup_fail_make_request);
1718
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001719static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001720{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001721 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001722}
1723
1724static int __init fail_make_request_debugfs(void)
1725{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001726 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1727 NULL, &fail_make_request);
1728
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001729 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001730}
1731
1732late_initcall(fail_make_request_debugfs);
1733
1734#else /* CONFIG_FAIL_MAKE_REQUEST */
1735
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001736static inline bool should_fail_request(struct hd_struct *part,
1737 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001738{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001739 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001740}
1741
1742#endif /* CONFIG_FAIL_MAKE_REQUEST */
1743
Jens Axboec07e2b42007-07-18 13:27:58 +02001744/*
1745 * Check whether this bio extends beyond the end of the device.
1746 */
1747static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1748{
1749 sector_t maxsector;
1750
1751 if (!nr_sectors)
1752 return 0;
1753
1754 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001755 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001756 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001757 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001758
1759 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1760 /*
1761 * This may well happen - the kernel calls bread()
1762 * without checking the size of the device, e.g., when
1763 * mounting a device.
1764 */
1765 handle_bad_sector(bio);
1766 return 1;
1767 }
1768 }
1769
1770 return 0;
1771}
1772
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001773static noinline_for_stack bool
1774generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
Jens Axboe165125e2007-07-24 09:28:11 +02001776 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001777 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001778 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001779 char b[BDEVNAME_SIZE];
1780 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
Jens Axboec07e2b42007-07-18 13:27:58 +02001784 if (bio_check_eod(bio, nr_sectors))
1785 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001787 q = bdev_get_queue(bio->bi_bdev);
1788 if (unlikely(!q)) {
1789 printk(KERN_ERR
1790 "generic_make_request: Trying to access "
1791 "nonexistent block-device %s (%Lu)\n",
1792 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001793 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001794 goto end_io;
1795 }
1796
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001797 if (likely(bio_is_rw(bio) &&
1798 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001799 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1800 bdevname(bio->bi_bdev, b),
1801 bio_sectors(bio),
1802 queue_max_hw_sectors(q));
1803 goto end_io;
1804 }
1805
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001806 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001807 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001808 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001809 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001810 goto end_io;
1811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001813 * If this device has partitions, remap block n
1814 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001816 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001818 if (bio_check_eod(bio, nr_sectors))
1819 goto end_io;
1820
1821 /*
1822 * Filter flush bio's early so that make_request based
1823 * drivers without flush support don't have to worry
1824 * about them.
1825 */
1826 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1827 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1828 if (!nr_sectors) {
1829 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001830 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001834 if ((bio->bi_rw & REQ_DISCARD) &&
1835 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001836 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001837 err = -EOPNOTSUPP;
1838 goto end_io;
1839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001841 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 err = -EOPNOTSUPP;
1843 goto end_io;
1844 }
1845
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001846 /*
1847 * Various block parts want %current->io_context and lazy ioc
1848 * allocation ends up trading a lot of pain for a small amount of
1849 * memory. Just allocate it upfront. This may fail and block
1850 * layer knows how to live with it.
1851 */
1852 create_io_context(GFP_ATOMIC, q->node);
1853
Tejun Heobc16a4f2011-10-19 14:33:01 +02001854 if (blk_throtl_bio(q, bio))
1855 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001856
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001857 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001858 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001859
1860end_io:
1861 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001862 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
1864
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001865/**
1866 * generic_make_request - hand a buffer to its device driver for I/O
1867 * @bio: The bio describing the location in memory and on the device.
1868 *
1869 * generic_make_request() is used to make I/O requests of block
1870 * devices. It is passed a &struct bio, which describes the I/O that needs
1871 * to be done.
1872 *
1873 * generic_make_request() does not return any status. The
1874 * success/failure status of the request, along with notification of
1875 * completion, is delivered asynchronously through the bio->bi_end_io
1876 * function described (one day) else where.
1877 *
1878 * The caller of generic_make_request must make sure that bi_io_vec
1879 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1880 * set to describe the device address, and the
1881 * bi_end_io and optionally bi_private are set to describe how
1882 * completion notification should be signaled.
1883 *
1884 * generic_make_request and the drivers it calls may use bi_next if this
1885 * bio happens to be merged with someone else, and may resubmit the bio to
1886 * a lower device by calling into generic_make_request recursively, which
1887 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001888 */
1889void generic_make_request(struct bio *bio)
1890{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001891 struct bio_list bio_list_on_stack;
1892
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001893 if (!generic_make_request_checks(bio))
1894 return;
1895
1896 /*
1897 * We only want one ->make_request_fn to be active at a time, else
1898 * stack usage with stacked devices could be a problem. So use
1899 * current->bio_list to keep a list of requests submited by a
1900 * make_request_fn function. current->bio_list is also used as a
1901 * flag to say if generic_make_request is currently active in this
1902 * task or not. If it is NULL, then no make_request is active. If
1903 * it is non-NULL, then a make_request is active, and new requests
1904 * should be added at the tail
1905 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001906 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001907 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001908 return;
1909 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001910
Neil Brownd89d8792007-05-01 09:53:42 +02001911 /* following loop may be a bit non-obvious, and so deserves some
1912 * explanation.
1913 * Before entering the loop, bio->bi_next is NULL (as all callers
1914 * ensure that) so we have a list with a single bio.
1915 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001916 * we assign bio_list to a pointer to the bio_list_on_stack,
1917 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001918 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001919 * through a recursive call to generic_make_request. If it
1920 * did, we find a non-NULL value in bio_list and re-enter the loop
1921 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001922 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001923 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001924 */
1925 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001926 bio_list_init(&bio_list_on_stack);
1927 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001928 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001929 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1930
1931 q->make_request_fn(q, bio);
1932
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001933 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001934 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001935 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001936}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937EXPORT_SYMBOL(generic_make_request);
1938
1939/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001940 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1942 * @bio: The &struct bio which describes the I/O
1943 *
1944 * submit_bio() is very similar in purpose to generic_make_request(), and
1945 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001946 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 *
1948 */
1949void submit_bio(int rw, struct bio *bio)
1950{
Jens Axboe22e2c502005-06-27 10:55:12 +02001951 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
Jens Axboebf2de6f2007-09-27 13:01:25 +02001953 /*
1954 * If it's a regular read/write or a barrier with data attached,
1955 * go through the normal accounting stuff before submission.
1956 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001957 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001958 unsigned int count;
1959
1960 if (unlikely(rw & REQ_WRITE_SAME))
1961 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1962 else
1963 count = bio_sectors(bio);
1964
Jens Axboebf2de6f2007-09-27 13:01:25 +02001965 if (rw & WRITE) {
1966 count_vm_events(PGPGOUT, count);
1967 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001968 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001969 count_vm_events(PGPGIN, count);
1970 }
1971
1972 if (unlikely(block_dump)) {
1973 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001974 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001975 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001976 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001977 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001978 bdevname(bio->bi_bdev, b),
1979 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001980 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
1982
1983 generic_make_request(bio);
1984}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985EXPORT_SYMBOL(submit_bio);
1986
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001987/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001988 * blk_rq_check_limits - Helper function to check a request for the queue limit
1989 * @q: the queue
1990 * @rq: the request being checked
1991 *
1992 * Description:
1993 * @rq may have been made based on weaker limitations of upper-level queues
1994 * in request stacking drivers, and it may violate the limitation of @q.
1995 * Since the block layer and the underlying device driver trust @rq
1996 * after it is inserted to @q, it should be checked against @q before
1997 * the insertion using this generic function.
1998 *
1999 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02002000 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002001 * Request stacking drivers like request-based dm may change the queue
2002 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09002003 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002004 * the new queue limits again when they dispatch those requests,
2005 * although such checkings are also done against the old queue limits
2006 * when submitting requests.
2007 */
2008int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2009{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002010 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04002011 return 0;
2012
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04002013 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002014 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2015 return -EIO;
2016 }
2017
2018 /*
2019 * queue's settings related to segment counting like q->bounce_pfn
2020 * may differ from that of other stacking queues.
2021 * Recalculate it to check the request correctly on this queue's
2022 * limitation.
2023 */
2024 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002025 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002026 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2027 return -EIO;
2028 }
2029
2030 return 0;
2031}
2032EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2033
2034/**
2035 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2036 * @q: the queue to submit the request
2037 * @rq: the request being queued
2038 */
2039int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2040{
2041 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002042 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002043
2044 if (blk_rq_check_limits(q, rq))
2045 return -EIO;
2046
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002047 if (rq->rq_disk &&
2048 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002049 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002050
Keith Busch7fb48982014-10-17 17:46:38 -06002051 if (q->mq_ops) {
2052 if (blk_queue_io_stat(q))
2053 blk_account_io_start(rq, true);
2054 blk_mq_insert_request(rq, false, true, true);
2055 return 0;
2056 }
2057
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002058 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002059 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002060 spin_unlock_irqrestore(q->queue_lock, flags);
2061 return -ENODEV;
2062 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002063
2064 /*
2065 * Submitting request must be dequeued before calling this function
2066 * because it will be linked to another request_queue
2067 */
2068 BUG_ON(blk_queued_rq(rq));
2069
Jeff Moyer4853aba2011-08-15 21:37:25 +02002070 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2071 where = ELEVATOR_INSERT_FLUSH;
2072
2073 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002074 if (where == ELEVATOR_INSERT_FLUSH)
2075 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002076 spin_unlock_irqrestore(q->queue_lock, flags);
2077
2078 return 0;
2079}
2080EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2081
Tejun Heo80a761f2009-07-03 17:48:17 +09002082/**
2083 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2084 * @rq: request to examine
2085 *
2086 * Description:
2087 * A request could be merge of IOs which require different failure
2088 * handling. This function determines the number of bytes which
2089 * can be failed from the beginning of the request without
2090 * crossing into area which need to be retried further.
2091 *
2092 * Return:
2093 * The number of bytes to fail.
2094 *
2095 * Context:
2096 * queue_lock must be held.
2097 */
2098unsigned int blk_rq_err_bytes(const struct request *rq)
2099{
2100 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2101 unsigned int bytes = 0;
2102 struct bio *bio;
2103
2104 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2105 return blk_rq_bytes(rq);
2106
2107 /*
2108 * Currently the only 'mixing' which can happen is between
2109 * different fastfail types. We can safely fail portions
2110 * which have all the failfast bits that the first one has -
2111 * the ones which are at least as eager to fail as the first
2112 * one.
2113 */
2114 for (bio = rq->bio; bio; bio = bio->bi_next) {
2115 if ((bio->bi_rw & ff) != ff)
2116 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002117 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002118 }
2119
2120 /* this could lead to infinite loop */
2121 BUG_ON(blk_rq_bytes(rq) && !bytes);
2122 return bytes;
2123}
2124EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2125
Jens Axboe320ae512013-10-24 09:20:05 +01002126void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002127{
Jens Axboec2553b52009-04-24 08:10:11 +02002128 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002129 const int rw = rq_data_dir(req);
2130 struct hd_struct *part;
2131 int cpu;
2132
2133 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002134 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002135 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2136 part_stat_unlock();
2137 }
2138}
2139
Jens Axboe320ae512013-10-24 09:20:05 +01002140void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002141{
Jens Axboebc58ba92009-01-23 10:54:44 +01002142 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002143 * Account IO completion. flush_rq isn't accounted as a
2144 * normal IO on queueing nor completion. Accounting the
2145 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002146 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002147 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002148 unsigned long duration = jiffies - req->start_time;
2149 const int rw = rq_data_dir(req);
2150 struct hd_struct *part;
2151 int cpu;
2152
2153 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002154 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002155
2156 part_stat_inc(cpu, part, ios[rw]);
2157 part_stat_add(cpu, part, ticks[rw], duration);
2158 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002159 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002160
Jens Axboe6c23a962011-01-07 08:43:37 +01002161 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002162 part_stat_unlock();
2163 }
2164}
2165
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002166#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002167/*
2168 * Don't process normal requests when queue is suspended
2169 * or in the process of suspending/resuming
2170 */
2171static struct request *blk_pm_peek_request(struct request_queue *q,
2172 struct request *rq)
2173{
2174 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2175 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2176 return NULL;
2177 else
2178 return rq;
2179}
2180#else
2181static inline struct request *blk_pm_peek_request(struct request_queue *q,
2182 struct request *rq)
2183{
2184 return rq;
2185}
2186#endif
2187
Jens Axboe320ae512013-10-24 09:20:05 +01002188void blk_account_io_start(struct request *rq, bool new_io)
2189{
2190 struct hd_struct *part;
2191 int rw = rq_data_dir(rq);
2192 int cpu;
2193
2194 if (!blk_do_io_stat(rq))
2195 return;
2196
2197 cpu = part_stat_lock();
2198
2199 if (!new_io) {
2200 part = rq->part;
2201 part_stat_inc(cpu, part, merges[rw]);
2202 } else {
2203 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2204 if (!hd_struct_try_get(part)) {
2205 /*
2206 * The partition is already being removed,
2207 * the request will be accounted on the disk only
2208 *
2209 * We take a reference on disk->part0 although that
2210 * partition will never be deleted, so we can treat
2211 * it as any other partition.
2212 */
2213 part = &rq->rq_disk->part0;
2214 hd_struct_get(part);
2215 }
2216 part_round_stats(cpu, part);
2217 part_inc_in_flight(part, rw);
2218 rq->part = part;
2219 }
2220
2221 part_stat_unlock();
2222}
2223
Tejun Heo53a08802008-12-03 12:41:26 +01002224/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002225 * blk_peek_request - peek at the top of a request queue
2226 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002227 *
2228 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002229 * Return the request at the top of @q. The returned request
2230 * should be started using blk_start_request() before LLD starts
2231 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002232 *
2233 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002234 * Pointer to the request at the top of @q if available. Null
2235 * otherwise.
2236 *
2237 * Context:
2238 * queue_lock must be held.
2239 */
2240struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002241{
2242 struct request *rq;
2243 int ret;
2244
2245 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002246
2247 rq = blk_pm_peek_request(q, rq);
2248 if (!rq)
2249 break;
2250
Tejun Heo158dbda2009-04-23 11:05:18 +09002251 if (!(rq->cmd_flags & REQ_STARTED)) {
2252 /*
2253 * This is the first time the device driver
2254 * sees this request (possibly after
2255 * requeueing). Notify IO scheduler.
2256 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002257 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002258 elv_activate_rq(q, rq);
2259
2260 /*
2261 * just mark as started even if we don't start
2262 * it, a request that has been delayed should
2263 * not be passed by new incoming requests
2264 */
2265 rq->cmd_flags |= REQ_STARTED;
2266 trace_block_rq_issue(q, rq);
2267 }
2268
2269 if (!q->boundary_rq || q->boundary_rq == rq) {
2270 q->end_sector = rq_end_sector(rq);
2271 q->boundary_rq = NULL;
2272 }
2273
2274 if (rq->cmd_flags & REQ_DONTPREP)
2275 break;
2276
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002277 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002278 /*
2279 * make sure space for the drain appears we
2280 * know we can do this because max_hw_segments
2281 * has been adjusted to be one fewer than the
2282 * device can handle
2283 */
2284 rq->nr_phys_segments++;
2285 }
2286
2287 if (!q->prep_rq_fn)
2288 break;
2289
2290 ret = q->prep_rq_fn(q, rq);
2291 if (ret == BLKPREP_OK) {
2292 break;
2293 } else if (ret == BLKPREP_DEFER) {
2294 /*
2295 * the request may have been (partially) prepped.
2296 * we need to keep this request in the front to
2297 * avoid resource deadlock. REQ_STARTED will
2298 * prevent other fs requests from passing this one.
2299 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002300 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002301 !(rq->cmd_flags & REQ_DONTPREP)) {
2302 /*
2303 * remove the space for the drain we added
2304 * so that we don't add it again
2305 */
2306 --rq->nr_phys_segments;
2307 }
2308
2309 rq = NULL;
2310 break;
2311 } else if (ret == BLKPREP_KILL) {
2312 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002313 /*
2314 * Mark this request as started so we don't trigger
2315 * any debug logic in the end I/O path.
2316 */
2317 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002318 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002319 } else {
2320 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2321 break;
2322 }
2323 }
2324
2325 return rq;
2326}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002327EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002328
Tejun Heo9934c8c2009-05-08 11:54:16 +09002329void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002330{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002331 struct request_queue *q = rq->q;
2332
Tejun Heo158dbda2009-04-23 11:05:18 +09002333 BUG_ON(list_empty(&rq->queuelist));
2334 BUG_ON(ELV_ON_HASH(rq));
2335
2336 list_del_init(&rq->queuelist);
2337
2338 /*
2339 * the time frame between a request being removed from the lists
2340 * and to it is freed is accounted as io that is in progress at
2341 * the driver side.
2342 */
Divyesh Shah91952912010-04-01 15:01:41 -07002343 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002344 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002345 set_io_start_time_ns(rq);
2346 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002347}
2348
Tejun Heo5efccd12009-04-23 11:05:18 +09002349/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002350 * blk_start_request - start request processing on the driver
2351 * @req: request to dequeue
2352 *
2353 * Description:
2354 * Dequeue @req and start timeout timer on it. This hands off the
2355 * request to the driver.
2356 *
2357 * Block internal functions which don't want to start timer should
2358 * call blk_dequeue_request().
2359 *
2360 * Context:
2361 * queue_lock must be held.
2362 */
2363void blk_start_request(struct request *req)
2364{
2365 blk_dequeue_request(req);
2366
2367 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002368 * We are now handing the request to the hardware, initialize
2369 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002370 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002371 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002372 if (unlikely(blk_bidi_rq(req)))
2373 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2374
Jeff Moyer4912aa62013-10-08 14:36:41 -04002375 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002376 blk_add_timer(req);
2377}
2378EXPORT_SYMBOL(blk_start_request);
2379
2380/**
2381 * blk_fetch_request - fetch a request from a request queue
2382 * @q: request queue to fetch a request from
2383 *
2384 * Description:
2385 * Return the request at the top of @q. The request is started on
2386 * return and LLD can start processing it immediately.
2387 *
2388 * Return:
2389 * Pointer to the request at the top of @q if available. Null
2390 * otherwise.
2391 *
2392 * Context:
2393 * queue_lock must be held.
2394 */
2395struct request *blk_fetch_request(struct request_queue *q)
2396{
2397 struct request *rq;
2398
2399 rq = blk_peek_request(q);
2400 if (rq)
2401 blk_start_request(rq);
2402 return rq;
2403}
2404EXPORT_SYMBOL(blk_fetch_request);
2405
2406/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002407 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002408 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002409 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002410 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002411 *
2412 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002413 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2414 * the request structure even if @req doesn't have leftover.
2415 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002416 *
2417 * This special helper function is only for request stacking drivers
2418 * (e.g. request-based dm) so that they can handle partial completion.
2419 * Actual device drivers should use blk_end_request instead.
2420 *
2421 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2422 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002423 *
2424 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002425 * %false - this request doesn't have any more data
2426 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002427 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002428bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002430 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002432 trace_block_rq_complete(req->q, req, nr_bytes);
2433
Tejun Heo2e60e022009-04-23 11:05:18 +09002434 if (!req->bio)
2435 return false;
2436
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002438 * For fs requests, rq is just carrier of independent bio's
2439 * and each partial completion should be handled separately.
2440 * Reset per-request error on each partial completion.
2441 *
2442 * TODO: tj: This is too subtle. It would be better to let
2443 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002445 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 req->errors = 0;
2447
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002448 if (error && req->cmd_type == REQ_TYPE_FS &&
2449 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002450 char *error_type;
2451
2452 switch (error) {
2453 case -ENOLINK:
2454 error_type = "recoverable transport";
2455 break;
2456 case -EREMOTEIO:
2457 error_type = "critical target";
2458 break;
2459 case -EBADE:
2460 error_type = "critical nexus";
2461 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002462 case -ETIMEDOUT:
2463 error_type = "timeout";
2464 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002465 case -ENOSPC:
2466 error_type = "critical space allocation";
2467 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002468 case -ENODATA:
2469 error_type = "critical medium";
2470 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002471 case -EIO:
2472 default:
2473 error_type = "I/O";
2474 break;
2475 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002476 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2477 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002478 req->rq_disk->disk_name : "?",
2479 (unsigned long long)blk_rq_pos(req));
2480
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 }
2482
Jens Axboebc58ba92009-01-23 10:54:44 +01002483 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002484
Kent Overstreetf79ea412012-09-20 16:38:30 -07002485 total_bytes = 0;
2486 while (req->bio) {
2487 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002488 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Kent Overstreet4f024f32013-10-11 15:44:27 -07002490 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Kent Overstreetf79ea412012-09-20 16:38:30 -07002493 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494
Kent Overstreetf79ea412012-09-20 16:38:30 -07002495 total_bytes += bio_bytes;
2496 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497
Kent Overstreetf79ea412012-09-20 16:38:30 -07002498 if (!nr_bytes)
2499 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 }
2501
2502 /*
2503 * completely done
2504 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002505 if (!req->bio) {
2506 /*
2507 * Reset counters so that the request stacking driver
2508 * can find how many bytes remain in the request
2509 * later.
2510 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002511 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002512 return false;
2513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002515 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002516
2517 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002518 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002519 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002520
Tejun Heo80a761f2009-07-03 17:48:17 +09002521 /* mixed attributes always follow the first bio */
2522 if (req->cmd_flags & REQ_MIXED_MERGE) {
2523 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2524 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2525 }
2526
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002527 /*
2528 * If total number of sectors is less than the first segment
2529 * size, something has gone terribly wrong.
2530 */
2531 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002532 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002533 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002534 }
2535
2536 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002538
Tejun Heo2e60e022009-04-23 11:05:18 +09002539 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540}
Tejun Heo2e60e022009-04-23 11:05:18 +09002541EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Tejun Heo2e60e022009-04-23 11:05:18 +09002543static bool blk_update_bidi_request(struct request *rq, int error,
2544 unsigned int nr_bytes,
2545 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002546{
Tejun Heo2e60e022009-04-23 11:05:18 +09002547 if (blk_update_request(rq, error, nr_bytes))
2548 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002549
Tejun Heo2e60e022009-04-23 11:05:18 +09002550 /* Bidi request must be completed as a whole */
2551 if (unlikely(blk_bidi_rq(rq)) &&
2552 blk_update_request(rq->next_rq, error, bidi_bytes))
2553 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002554
Jens Axboee2e1a142010-06-09 10:42:09 +02002555 if (blk_queue_add_random(rq->q))
2556 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002557
2558 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559}
2560
James Bottomley28018c22010-07-01 19:49:17 +09002561/**
2562 * blk_unprep_request - unprepare a request
2563 * @req: the request
2564 *
2565 * This function makes a request ready for complete resubmission (or
2566 * completion). It happens only after all error handling is complete,
2567 * so represents the appropriate moment to deallocate any resources
2568 * that were allocated to the request in the prep_rq_fn. The queue
2569 * lock is held when calling this.
2570 */
2571void blk_unprep_request(struct request *req)
2572{
2573 struct request_queue *q = req->q;
2574
2575 req->cmd_flags &= ~REQ_DONTPREP;
2576 if (q->unprep_rq_fn)
2577 q->unprep_rq_fn(q, req);
2578}
2579EXPORT_SYMBOL_GPL(blk_unprep_request);
2580
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581/*
2582 * queue lock must be held
2583 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002584void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002586 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002587 blk_queue_end_tag(req->q, req);
2588
James Bottomleyba396a62009-05-27 14:17:08 +02002589 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002591 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002592 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
Mike Andersone78042e2008-10-30 02:16:20 -07002594 blk_delete_timer(req);
2595
James Bottomley28018c22010-07-01 19:49:17 +09002596 if (req->cmd_flags & REQ_DONTPREP)
2597 blk_unprep_request(req);
2598
Jens Axboebc58ba92009-01-23 10:54:44 +01002599 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002602 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002603 else {
2604 if (blk_bidi_rq(req))
2605 __blk_put_request(req->next_rq->q, req->next_rq);
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 }
2609}
Christoph Hellwig12120072014-04-16 09:44:59 +02002610EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002612/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002613 * blk_end_bidi_request - Complete a bidi request
2614 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002615 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002616 * @nr_bytes: number of bytes to complete @rq
2617 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002618 *
2619 * Description:
2620 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002621 * Drivers that supports bidi can safely call this member for any
2622 * type of request, bidi or uni. In the later case @bidi_bytes is
2623 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002624 *
2625 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002626 * %false - we are done with this request
2627 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002628 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002629static bool blk_end_bidi_request(struct request *rq, int error,
2630 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002631{
2632 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002633 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002634
Tejun Heo2e60e022009-04-23 11:05:18 +09002635 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2636 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002637
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002638 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002639 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002640 spin_unlock_irqrestore(q->queue_lock, flags);
2641
Tejun Heo2e60e022009-04-23 11:05:18 +09002642 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002643}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002644
2645/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002646 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2647 * @rq: the request to complete
2648 * @error: %0 for success, < %0 for error
2649 * @nr_bytes: number of bytes to complete @rq
2650 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002651 *
2652 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002653 * Identical to blk_end_bidi_request() except that queue lock is
2654 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002655 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002656 * Return:
2657 * %false - we are done with this request
2658 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002659 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002660bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002661 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002662{
Tejun Heo2e60e022009-04-23 11:05:18 +09002663 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2664 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002665
Tejun Heo2e60e022009-04-23 11:05:18 +09002666 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002667
Tejun Heo2e60e022009-04-23 11:05:18 +09002668 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002669}
2670
2671/**
2672 * blk_end_request - Helper function for drivers to complete the request.
2673 * @rq: the request being processed
2674 * @error: %0 for success, < %0 for error
2675 * @nr_bytes: number of bytes to complete
2676 *
2677 * Description:
2678 * Ends I/O on a number of bytes attached to @rq.
2679 * If @rq has leftover, sets it up for the next range of segments.
2680 *
2681 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002682 * %false - we are done with this request
2683 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002684 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002685bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002686{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002687 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002688}
Jens Axboe56ad1742009-07-28 22:11:24 +02002689EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002690
2691/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002692 * blk_end_request_all - Helper function for drives to finish the request.
2693 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002694 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002695 *
2696 * Description:
2697 * Completely finish @rq.
2698 */
2699void blk_end_request_all(struct request *rq, int error)
2700{
2701 bool pending;
2702 unsigned int bidi_bytes = 0;
2703
2704 if (unlikely(blk_bidi_rq(rq)))
2705 bidi_bytes = blk_rq_bytes(rq->next_rq);
2706
2707 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2708 BUG_ON(pending);
2709}
Jens Axboe56ad1742009-07-28 22:11:24 +02002710EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002711
2712/**
2713 * blk_end_request_cur - Helper function to finish the current request chunk.
2714 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002715 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002716 *
2717 * Description:
2718 * Complete the current consecutively mapped chunk from @rq.
2719 *
2720 * Return:
2721 * %false - we are done with this request
2722 * %true - still buffers pending for this request
2723 */
2724bool blk_end_request_cur(struct request *rq, int error)
2725{
2726 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2727}
Jens Axboe56ad1742009-07-28 22:11:24 +02002728EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002729
2730/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002731 * blk_end_request_err - Finish a request till the next failure boundary.
2732 * @rq: the request to finish till the next failure boundary for
2733 * @error: must be negative errno
2734 *
2735 * Description:
2736 * Complete @rq till the next failure boundary.
2737 *
2738 * Return:
2739 * %false - we are done with this request
2740 * %true - still buffers pending for this request
2741 */
2742bool blk_end_request_err(struct request *rq, int error)
2743{
2744 WARN_ON(error >= 0);
2745 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2746}
2747EXPORT_SYMBOL_GPL(blk_end_request_err);
2748
2749/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002750 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002751 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002752 * @error: %0 for success, < %0 for error
2753 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002754 *
2755 * Description:
2756 * Must be called with queue lock held unlike blk_end_request().
2757 *
2758 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002759 * %false - we are done with this request
2760 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002761 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002762bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002763{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002764 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002765}
Jens Axboe56ad1742009-07-28 22:11:24 +02002766EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002767
2768/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002769 * __blk_end_request_all - Helper function for drives to finish the request.
2770 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002771 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002772 *
2773 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002774 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002775 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002776void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002777{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002778 bool pending;
2779 unsigned int bidi_bytes = 0;
2780
2781 if (unlikely(blk_bidi_rq(rq)))
2782 bidi_bytes = blk_rq_bytes(rq->next_rq);
2783
2784 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2785 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002786}
Jens Axboe56ad1742009-07-28 22:11:24 +02002787EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002788
2789/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002790 * __blk_end_request_cur - Helper function to finish the current request chunk.
2791 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002792 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002793 *
2794 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002795 * Complete the current consecutively mapped chunk from @rq. Must
2796 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002797 *
2798 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002799 * %false - we are done with this request
2800 * %true - still buffers pending for this request
2801 */
2802bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002803{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002804 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002805}
Jens Axboe56ad1742009-07-28 22:11:24 +02002806EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002807
Tejun Heo80a761f2009-07-03 17:48:17 +09002808/**
2809 * __blk_end_request_err - Finish a request till the next failure boundary.
2810 * @rq: the request to finish till the next failure boundary for
2811 * @error: must be negative errno
2812 *
2813 * Description:
2814 * Complete @rq till the next failure boundary. Must be called
2815 * with queue lock held.
2816 *
2817 * Return:
2818 * %false - we are done with this request
2819 * %true - still buffers pending for this request
2820 */
2821bool __blk_end_request_err(struct request *rq, int error)
2822{
2823 WARN_ON(error >= 0);
2824 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2825}
2826EXPORT_SYMBOL_GPL(__blk_end_request_err);
2827
Jens Axboe86db1e22008-01-29 14:53:40 +01002828void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2829 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002831 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002832 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
Jens Axboeb4f42e22014-04-10 09:46:28 -06002834 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002835 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002836
Kent Overstreet4f024f32013-10-11 15:44:27 -07002837 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
NeilBrown66846572007-08-16 13:31:28 +02002840 if (bio->bi_bdev)
2841 rq->rq_disk = bio->bi_bdev->bd_disk;
2842}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002844#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2845/**
2846 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2847 * @rq: the request to be flushed
2848 *
2849 * Description:
2850 * Flush all pages in @rq.
2851 */
2852void rq_flush_dcache_pages(struct request *rq)
2853{
2854 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002855 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002856
2857 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002858 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002859}
2860EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2861#endif
2862
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002863/**
2864 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2865 * @q : the queue of the device being checked
2866 *
2867 * Description:
2868 * Check if underlying low-level drivers of a device are busy.
2869 * If the drivers want to export their busy state, they must set own
2870 * exporting function using blk_queue_lld_busy() first.
2871 *
2872 * Basically, this function is used only by request stacking drivers
2873 * to stop dispatching requests to underlying devices when underlying
2874 * devices are busy. This behavior helps more I/O merging on the queue
2875 * of the request stacking driver and prevents I/O throughput regression
2876 * on burst I/O load.
2877 *
2878 * Return:
2879 * 0 - Not busy (The request stacking driver should dispatch request)
2880 * 1 - Busy (The request stacking driver should stop dispatching request)
2881 */
2882int blk_lld_busy(struct request_queue *q)
2883{
2884 if (q->lld_busy_fn)
2885 return q->lld_busy_fn(q);
2886
2887 return 0;
2888}
2889EXPORT_SYMBOL_GPL(blk_lld_busy);
2890
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002891/**
2892 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2893 * @rq: the clone request to be cleaned up
2894 *
2895 * Description:
2896 * Free all bios in @rq for a cloned request.
2897 */
2898void blk_rq_unprep_clone(struct request *rq)
2899{
2900 struct bio *bio;
2901
2902 while ((bio = rq->bio) != NULL) {
2903 rq->bio = bio->bi_next;
2904
2905 bio_put(bio);
2906 }
2907}
2908EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2909
2910/*
2911 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002912 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002913 */
2914static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2915{
2916 dst->cpu = src->cpu;
Keith Busch77a08682015-01-12 15:04:06 -05002917 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002918 dst->cmd_type = src->cmd_type;
2919 dst->__sector = blk_rq_pos(src);
2920 dst->__data_len = blk_rq_bytes(src);
2921 dst->nr_phys_segments = src->nr_phys_segments;
2922 dst->ioprio = src->ioprio;
2923 dst->extra_len = src->extra_len;
2924}
2925
2926/**
2927 * blk_rq_prep_clone - Helper function to setup clone request
2928 * @rq: the request to be setup
2929 * @rq_src: original request to be cloned
2930 * @bs: bio_set that bios for clone are allocated from
2931 * @gfp_mask: memory allocation mask for bio
2932 * @bio_ctr: setup function to be called for each clone bio.
2933 * Returns %0 for success, non %0 for failure.
2934 * @data: private data to be passed to @bio_ctr
2935 *
2936 * Description:
2937 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002938 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002939 * are not copied, and copying such parts is the caller's responsibility.
2940 * Also, pages which the original bios are pointing to are not copied
2941 * and the cloned bios just point same pages.
2942 * So cloned bios must be completed before original bios, which means
2943 * the caller must complete @rq before @rq_src.
2944 */
2945int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2946 struct bio_set *bs, gfp_t gfp_mask,
2947 int (*bio_ctr)(struct bio *, struct bio *, void *),
2948 void *data)
2949{
2950 struct bio *bio, *bio_src;
2951
2952 if (!bs)
2953 bs = fs_bio_set;
2954
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002955 __rq_for_each_bio(bio_src, rq_src) {
Junichi Nomura11dfce52014-10-03 17:27:11 -04002956 bio = bio_clone_fast(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002957 if (!bio)
2958 goto free_and_out;
2959
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002960 if (bio_ctr && bio_ctr(bio, bio_src, data))
2961 goto free_and_out;
2962
2963 if (rq->bio) {
2964 rq->biotail->bi_next = bio;
2965 rq->biotail = bio;
2966 } else
2967 rq->bio = rq->biotail = bio;
2968 }
2969
2970 __blk_rq_prep_clone(rq, rq_src);
2971
2972 return 0;
2973
2974free_and_out:
2975 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002976 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002977 blk_rq_unprep_clone(rq);
2978
2979 return -ENOMEM;
2980}
2981EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2982
Jens Axboe59c3d452014-04-08 09:15:35 -06002983int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984{
2985 return queue_work(kblockd_workqueue, work);
2986}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987EXPORT_SYMBOL(kblockd_schedule_work);
2988
Jens Axboe59c3d452014-04-08 09:15:35 -06002989int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2990 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04002991{
2992 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2993}
2994EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2995
Jens Axboe8ab14592014-04-08 09:17:40 -06002996int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2997 unsigned long delay)
2998{
2999 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3000}
3001EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3002
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003003/**
3004 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3005 * @plug: The &struct blk_plug that needs to be initialized
3006 *
3007 * Description:
3008 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3009 * pending I/O should the task end up blocking between blk_start_plug() and
3010 * blk_finish_plug(). This is important from a performance perspective, but
3011 * also ensures that we don't deadlock. For instance, if the task is blocking
3012 * for a memory allocation, memory reclaim could end up wanting to free a
3013 * page belonging to that request that is currently residing in our private
3014 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3015 * this kind of deadlock.
3016 */
Jens Axboe73c10102011-03-08 13:19:51 +01003017void blk_start_plug(struct blk_plug *plug)
3018{
3019 struct task_struct *tsk = current;
3020
Jens Axboe73c10102011-03-08 13:19:51 +01003021 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003022 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003023 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003024
3025 /*
3026 * If this is a nested plug, don't actually assign it. It will be
3027 * flushed on its own.
3028 */
3029 if (!tsk->plug) {
3030 /*
3031 * Store ordering should not be needed here, since a potential
3032 * preempt will imply a full memory barrier
3033 */
3034 tsk->plug = plug;
3035 }
3036}
3037EXPORT_SYMBOL(blk_start_plug);
3038
3039static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3040{
3041 struct request *rqa = container_of(a, struct request, queuelist);
3042 struct request *rqb = container_of(b, struct request, queuelist);
3043
Jianpeng Ma975927b2012-10-25 21:58:17 +02003044 return !(rqa->q < rqb->q ||
3045 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003046}
3047
Jens Axboe49cac012011-04-16 13:51:05 +02003048/*
3049 * If 'from_schedule' is true, then postpone the dispatch of requests
3050 * until a safe kblockd context. We due this to avoid accidental big
3051 * additional stack usage in driver dispatch, in places where the originally
3052 * plugger did not intend it.
3053 */
Jens Axboef6603782011-04-15 15:49:07 +02003054static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003055 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003056 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003057{
Jens Axboe49cac012011-04-16 13:51:05 +02003058 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003059
Bart Van Assche70460572012-11-28 13:45:56 +01003060 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003061 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003062 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003063 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003064 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003065}
3066
NeilBrown74018dc2012-07-31 09:08:15 +02003067static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003068{
3069 LIST_HEAD(callbacks);
3070
Shaohua Li2a7d5552012-07-31 09:08:15 +02003071 while (!list_empty(&plug->cb_list)) {
3072 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003073
Shaohua Li2a7d5552012-07-31 09:08:15 +02003074 while (!list_empty(&callbacks)) {
3075 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003076 struct blk_plug_cb,
3077 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003078 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003079 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003080 }
NeilBrown048c9372011-04-18 09:52:22 +02003081 }
3082}
3083
NeilBrown9cbb1752012-07-31 09:08:14 +02003084struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3085 int size)
3086{
3087 struct blk_plug *plug = current->plug;
3088 struct blk_plug_cb *cb;
3089
3090 if (!plug)
3091 return NULL;
3092
3093 list_for_each_entry(cb, &plug->cb_list, list)
3094 if (cb->callback == unplug && cb->data == data)
3095 return cb;
3096
3097 /* Not currently on the callback list */
3098 BUG_ON(size < sizeof(*cb));
3099 cb = kzalloc(size, GFP_ATOMIC);
3100 if (cb) {
3101 cb->data = data;
3102 cb->callback = unplug;
3103 list_add(&cb->list, &plug->cb_list);
3104 }
3105 return cb;
3106}
3107EXPORT_SYMBOL(blk_check_plugged);
3108
Jens Axboe49cac012011-04-16 13:51:05 +02003109void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003110{
3111 struct request_queue *q;
3112 unsigned long flags;
3113 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003114 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003115 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003116
NeilBrown74018dc2012-07-31 09:08:15 +02003117 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003118
3119 if (!list_empty(&plug->mq_list))
3120 blk_mq_flush_plug_list(plug, from_schedule);
3121
Jens Axboe73c10102011-03-08 13:19:51 +01003122 if (list_empty(&plug->list))
3123 return;
3124
NeilBrown109b8122011-04-11 14:13:10 +02003125 list_splice_init(&plug->list, &list);
3126
Jianpeng Ma422765c2013-01-11 14:46:09 +01003127 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003128
3129 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003130 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003131
3132 /*
3133 * Save and disable interrupts here, to avoid doing it for every
3134 * queue lock we have to take.
3135 */
Jens Axboe73c10102011-03-08 13:19:51 +01003136 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003137 while (!list_empty(&list)) {
3138 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003139 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003140 BUG_ON(!rq->q);
3141 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003142 /*
3143 * This drops the queue lock
3144 */
3145 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003146 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003147 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003148 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003149 spin_lock(q->queue_lock);
3150 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003151
3152 /*
3153 * Short-circuit if @q is dead
3154 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003155 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003156 __blk_end_request_all(rq, -ENODEV);
3157 continue;
3158 }
3159
Jens Axboe73c10102011-03-08 13:19:51 +01003160 /*
3161 * rq is already accounted, so use raw insert
3162 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003163 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3164 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3165 else
3166 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003167
3168 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003169 }
3170
Jens Axboe99e22592011-04-18 09:59:55 +02003171 /*
3172 * This drops the queue lock
3173 */
3174 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003175 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003176
Jens Axboe73c10102011-03-08 13:19:51 +01003177 local_irq_restore(flags);
3178}
Jens Axboe73c10102011-03-08 13:19:51 +01003179
3180void blk_finish_plug(struct blk_plug *plug)
3181{
Jens Axboef6603782011-04-15 15:49:07 +02003182 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003183
3184 if (plug == current->plug)
3185 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003186}
3187EXPORT_SYMBOL(blk_finish_plug);
3188
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003189#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003190/**
3191 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3192 * @q: the queue of the device
3193 * @dev: the device the queue belongs to
3194 *
3195 * Description:
3196 * Initialize runtime-PM-related fields for @q and start auto suspend for
3197 * @dev. Drivers that want to take advantage of request-based runtime PM
3198 * should call this function after @dev has been initialized, and its
3199 * request queue @q has been allocated, and runtime PM for it can not happen
3200 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3201 * cases, driver should call this function before any I/O has taken place.
3202 *
3203 * This function takes care of setting up using auto suspend for the device,
3204 * the autosuspend delay is set to -1 to make runtime suspend impossible
3205 * until an updated value is either set by user or by driver. Drivers do
3206 * not need to touch other autosuspend settings.
3207 *
3208 * The block layer runtime PM is request based, so only works for drivers
3209 * that use request as their IO unit instead of those directly use bio's.
3210 */
3211void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3212{
3213 q->dev = dev;
3214 q->rpm_status = RPM_ACTIVE;
3215 pm_runtime_set_autosuspend_delay(q->dev, -1);
3216 pm_runtime_use_autosuspend(q->dev);
3217}
3218EXPORT_SYMBOL(blk_pm_runtime_init);
3219
3220/**
3221 * blk_pre_runtime_suspend - Pre runtime suspend check
3222 * @q: the queue of the device
3223 *
3224 * Description:
3225 * This function will check if runtime suspend is allowed for the device
3226 * by examining if there are any requests pending in the queue. If there
3227 * are requests pending, the device can not be runtime suspended; otherwise,
3228 * the queue's status will be updated to SUSPENDING and the driver can
3229 * proceed to suspend the device.
3230 *
3231 * For the not allowed case, we mark last busy for the device so that
3232 * runtime PM core will try to autosuspend it some time later.
3233 *
3234 * This function should be called near the start of the device's
3235 * runtime_suspend callback.
3236 *
3237 * Return:
3238 * 0 - OK to runtime suspend the device
3239 * -EBUSY - Device should not be runtime suspended
3240 */
3241int blk_pre_runtime_suspend(struct request_queue *q)
3242{
3243 int ret = 0;
3244
3245 spin_lock_irq(q->queue_lock);
3246 if (q->nr_pending) {
3247 ret = -EBUSY;
3248 pm_runtime_mark_last_busy(q->dev);
3249 } else {
3250 q->rpm_status = RPM_SUSPENDING;
3251 }
3252 spin_unlock_irq(q->queue_lock);
3253 return ret;
3254}
3255EXPORT_SYMBOL(blk_pre_runtime_suspend);
3256
3257/**
3258 * blk_post_runtime_suspend - Post runtime suspend processing
3259 * @q: the queue of the device
3260 * @err: return value of the device's runtime_suspend function
3261 *
3262 * Description:
3263 * Update the queue's runtime status according to the return value of the
3264 * device's runtime suspend function and mark last busy for the device so
3265 * that PM core will try to auto suspend the device at a later time.
3266 *
3267 * This function should be called near the end of the device's
3268 * runtime_suspend callback.
3269 */
3270void blk_post_runtime_suspend(struct request_queue *q, int err)
3271{
3272 spin_lock_irq(q->queue_lock);
3273 if (!err) {
3274 q->rpm_status = RPM_SUSPENDED;
3275 } else {
3276 q->rpm_status = RPM_ACTIVE;
3277 pm_runtime_mark_last_busy(q->dev);
3278 }
3279 spin_unlock_irq(q->queue_lock);
3280}
3281EXPORT_SYMBOL(blk_post_runtime_suspend);
3282
3283/**
3284 * blk_pre_runtime_resume - Pre runtime resume processing
3285 * @q: the queue of the device
3286 *
3287 * Description:
3288 * Update the queue's runtime status to RESUMING in preparation for the
3289 * runtime resume of the device.
3290 *
3291 * This function should be called near the start of the device's
3292 * runtime_resume callback.
3293 */
3294void blk_pre_runtime_resume(struct request_queue *q)
3295{
3296 spin_lock_irq(q->queue_lock);
3297 q->rpm_status = RPM_RESUMING;
3298 spin_unlock_irq(q->queue_lock);
3299}
3300EXPORT_SYMBOL(blk_pre_runtime_resume);
3301
3302/**
3303 * blk_post_runtime_resume - Post runtime resume processing
3304 * @q: the queue of the device
3305 * @err: return value of the device's runtime_resume function
3306 *
3307 * Description:
3308 * Update the queue's runtime status according to the return value of the
3309 * device's runtime_resume function. If it is successfully resumed, process
3310 * the requests that are queued into the device's queue when it is resuming
3311 * and then mark last busy and initiate autosuspend for it.
3312 *
3313 * This function should be called near the end of the device's
3314 * runtime_resume callback.
3315 */
3316void blk_post_runtime_resume(struct request_queue *q, int err)
3317{
3318 spin_lock_irq(q->queue_lock);
3319 if (!err) {
3320 q->rpm_status = RPM_ACTIVE;
3321 __blk_run_queue(q);
3322 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003323 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003324 } else {
3325 q->rpm_status = RPM_SUSPENDED;
3326 }
3327 spin_unlock_irq(q->queue_lock);
3328}
3329EXPORT_SYMBOL(blk_post_runtime_resume);
3330#endif
3331
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332int __init blk_dev_init(void)
3333{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003334 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3335 sizeof(((struct request *)0)->cmd_flags));
3336
Tejun Heo89b90be2011-01-03 15:01:47 +01003337 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3338 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003339 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340 if (!kblockd_workqueue)
3341 panic("Failed to create kblockd\n");
3342
3343 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003344 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345
Jens Axboe8324aa92008-01-29 14:51:59 +01003346 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003347 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 return 0;
3350}