blob: fd154b94447a25788f48d5e8cc04bc803d1efdb8 [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
David Rientjes271508d2015-03-24 16:21:16 -0700560/* Allocate memory local to the request queue */
561static void *alloc_request_struct(gfp_t gfp_mask, void *data)
562{
563 int nid = (int)(long)data;
564 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
565}
566
567static void free_request_struct(void *element, void *unused)
568{
569 kmem_cache_free(request_cachep, element);
570}
571
Tejun Heo5b788ce2012-06-04 20:40:59 -0700572int blk_init_rl(struct request_list *rl, struct request_queue *q,
573 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400575 if (unlikely(rl->rq_pool))
576 return 0;
577
Tejun Heo5b788ce2012-06-04 20:40:59 -0700578 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200579 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
580 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200581 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
582 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
David Rientjes271508d2015-03-24 16:21:16 -0700584 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
585 free_request_struct,
586 (void *)(long)q->node, gfp_mask,
587 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 if (!rl->rq_pool)
589 return -ENOMEM;
590
591 return 0;
592}
593
Tejun Heo5b788ce2012-06-04 20:40:59 -0700594void blk_exit_rl(struct request_list *rl)
595{
596 if (rl->rq_pool)
597 mempool_destroy(rl->rq_pool);
598}
599
Jens Axboe165125e2007-07-24 09:28:11 +0200600struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100602 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700603}
604EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
Jens Axboe165125e2007-07-24 09:28:11 +0200606struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700607{
Jens Axboe165125e2007-07-24 09:28:11 +0200608 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700609 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700610
Jens Axboe8324aa92008-01-29 14:51:59 +0100611 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700612 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (!q)
614 return NULL;
615
Dan Carpenter00380a42012-03-23 09:58:54 +0100616 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100617 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800618 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100619
Jens Axboe0989a022009-06-12 14:42:56 +0200620 q->backing_dev_info.ra_pages =
621 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
622 q->backing_dev_info.state = 0;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100623 q->backing_dev_info.capabilities = 0;
Jens Axboed9938312009-06-12 14:45:52 +0200624 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100625 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200626
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700627 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100628 if (err)
629 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700630
Matthew Garrett31373d02010-04-06 14:25:14 +0200631 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
632 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700633 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100634 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700635 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100636 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800637#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800638 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800639#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500640 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500641
Jens Axboe8324aa92008-01-29 14:51:59 +0100642 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Al Viro483f4af2006-03-18 18:34:37 -0500644 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700645 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500646
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500647 /*
648 * By default initialize queue_lock to internal lock and driver can
649 * override it later if need be.
650 */
651 q->queue_lock = &q->__queue_lock;
652
Tejun Heob82d4b12012-04-13 13:11:31 -0700653 /*
654 * A queue starts its life with bypass turned on to avoid
655 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700656 * init. The initial bypass will be finished when the queue is
657 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700658 */
659 q->bypass_depth = 1;
660 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
661
Jens Axboe320ae512013-10-24 09:20:05 +0100662 init_waitqueue_head(&q->mq_freeze_wq);
663
Tejun Heo5efd6112012-03-05 13:15:12 -0800664 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400665 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100668
Mikulas Patockafff49962013-10-14 12:11:36 -0400669fail_bdi:
670 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100671fail_id:
672 ida_simple_remove(&blk_queue_ida, q->id);
673fail_q:
674 kmem_cache_free(blk_requestq_cachep, q);
675 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
Christoph Lameter19460892005-06-23 00:08:19 -0700677EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679/**
680 * blk_init_queue - prepare a request queue for use with a block device
681 * @rfn: The function to be called to process requests that have been
682 * placed on the queue.
683 * @lock: Request queue spin lock
684 *
685 * Description:
686 * If a block device wishes to use the standard request handling procedures,
687 * which sorts requests and coalesces adjacent requests, then it must
688 * call blk_init_queue(). The function @rfn will be called when there
689 * are requests on the queue that need to be processed. If the device
690 * supports plugging, then @rfn may not be called immediately when requests
691 * are available on the queue, but may be called at some time later instead.
692 * Plugged queues are generally unplugged when a buffer belonging to one
693 * of the requests on the queue is needed, or due to memory pressure.
694 *
695 * @rfn is not required, or even expected, to remove all requests off the
696 * queue, but only as many as it can handle at a time. If it does leave
697 * requests on the queue, it is responsible for arranging that the requests
698 * get dealt with eventually.
699 *
700 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200701 * request queue; this lock will be taken also from interrupt context, so irq
702 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200704 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 * it didn't succeed.
706 *
707 * Note:
708 * blk_init_queue() must be paired with a blk_cleanup_queue() call
709 * when the block device is deactivated (such as at module unload).
710 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700711
Jens Axboe165125e2007-07-24 09:28:11 +0200712struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100714 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700715}
716EXPORT_SYMBOL(blk_init_queue);
717
Jens Axboe165125e2007-07-24 09:28:11 +0200718struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700719blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
720{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600721 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600723 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
724 if (!uninit_q)
725 return NULL;
726
Mike Snitzer51514122011-11-23 10:59:13 +0100727 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600728 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700729 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700730
Mike Snitzer7982e902014-03-08 17:20:01 -0700731 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200732}
733EXPORT_SYMBOL(blk_init_queue_node);
734
735struct request_queue *
736blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
737 spinlock_t *lock)
738{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (!q)
740 return NULL;
741
Ming Leif70ced02014-09-25 23:23:47 +0800742 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800743 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700744 return NULL;
745
Tejun Heoa0516612012-06-26 15:05:44 -0700746 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600747 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
749 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900751 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700752 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500753
754 /* Override internal queue lock with supplied lock pointer */
755 if (lock)
756 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Jens Axboef3b144a2009-03-06 08:48:33 +0100758 /*
759 * This also sets hw/phys segments, boundary and size
760 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200761 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Alan Stern44ec9542007-02-20 11:01:57 -0500763 q->sg_reserved_size = INT_MAX;
764
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600765 /* Protect q->elevator from elevator_change */
766 mutex_lock(&q->sysfs_lock);
767
Tejun Heob82d4b12012-04-13 13:11:31 -0700768 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600769 if (elevator_init(q, NULL)) {
770 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600771 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600772 }
773
774 mutex_unlock(&q->sysfs_lock);
775
Tejun Heob82d4b12012-04-13 13:11:31 -0700776 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600777
778fail:
Ming Leiba483382014-09-25 23:23:44 +0800779 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600780 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
Mike Snitzer51514122011-11-23 10:59:13 +0100782EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Tejun Heo09ac46c2011-12-14 00:33:38 +0100784bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100786 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100787 __blk_get_queue(q);
788 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
790
Tejun Heo09ac46c2011-12-14 00:33:38 +0100791 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
Jens Axboed86e0e82011-05-27 07:44:43 +0200793EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Tejun Heo5b788ce2012-06-04 20:40:59 -0700795static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100797 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700798 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100799 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100800 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100801 }
802
Tejun Heo5b788ce2012-06-04 20:40:59 -0700803 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806/*
807 * ioc_batching returns true if the ioc is a valid batching request and
808 * should be given priority access to a request.
809 */
Jens Axboe165125e2007-07-24 09:28:11 +0200810static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
812 if (!ioc)
813 return 0;
814
815 /*
816 * Make sure the process is able to allocate at least 1 request
817 * even if the batch times out, otherwise we could theoretically
818 * lose wakeups.
819 */
820 return ioc->nr_batch_requests == q->nr_batching ||
821 (ioc->nr_batch_requests > 0
822 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
823}
824
825/*
826 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
827 * will cause the process to be a "batcher" on all queues in the system. This
828 * is the behaviour we want though - once it gets a wakeup it should be given
829 * a nice run.
830 */
Jens Axboe165125e2007-07-24 09:28:11 +0200831static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 if (!ioc || ioc_batching(q, ioc))
834 return;
835
836 ioc->nr_batch_requests = q->nr_batching;
837 ioc->last_waited = jiffies;
838}
839
Tejun Heo5b788ce2012-06-04 20:40:59 -0700840static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700842 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Tejun Heoa0516612012-06-26 15:05:44 -0700844 /*
845 * bdi isn't aware of blkcg yet. As all async IOs end up root
846 * blkcg anyway, just use root blkcg state.
847 */
848 if (rl == &q->root_rl &&
849 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200850 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Jens Axboe1faa16d2009-04-06 14:48:01 +0200852 if (rl->count[sync] + 1 <= q->nr_requests) {
853 if (waitqueue_active(&rl->wait[sync]))
854 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Tejun Heo5b788ce2012-06-04 20:40:59 -0700856 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 }
858}
859
860/*
861 * A request has just been released. Account for it, update the full and
862 * congestion status, wake up any waiters. Called under q->queue_lock.
863 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700864static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700866 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200867 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700869 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200870 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200871 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700872 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Tejun Heo5b788ce2012-06-04 20:40:59 -0700874 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Jens Axboe1faa16d2009-04-06 14:48:01 +0200876 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700877 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600880int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
881{
882 struct request_list *rl;
883
884 spin_lock_irq(q->queue_lock);
885 q->nr_requests = nr;
886 blk_queue_congestion_threshold(q);
887
888 /* congestion isn't cgroup aware and follows root blkcg for now */
889 rl = &q->root_rl;
890
891 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
892 blk_set_queue_congested(q, BLK_RW_SYNC);
893 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
894 blk_clear_queue_congested(q, BLK_RW_SYNC);
895
896 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
897 blk_set_queue_congested(q, BLK_RW_ASYNC);
898 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
899 blk_clear_queue_congested(q, BLK_RW_ASYNC);
900
901 blk_queue_for_each_rl(rl, q) {
902 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
903 blk_set_rl_full(rl, BLK_RW_SYNC);
904 } else {
905 blk_clear_rl_full(rl, BLK_RW_SYNC);
906 wake_up(&rl->wait[BLK_RW_SYNC]);
907 }
908
909 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
910 blk_set_rl_full(rl, BLK_RW_ASYNC);
911 } else {
912 blk_clear_rl_full(rl, BLK_RW_ASYNC);
913 wake_up(&rl->wait[BLK_RW_ASYNC]);
914 }
915 }
916
917 spin_unlock_irq(q->queue_lock);
918 return 0;
919}
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100922 * Determine if elevator data should be initialized when allocating the
923 * request associated with @bio.
924 */
925static bool blk_rq_should_init_elevator(struct bio *bio)
926{
927 if (!bio)
928 return true;
929
930 /*
931 * Flush requests do not use the elevator so skip initialization.
932 * This allows a request to share the flush and elevator data.
933 */
934 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
935 return false;
936
937 return true;
938}
939
Tejun Heoda8303c2011-10-19 14:33:05 +0200940/**
Tejun Heo852c7882012-03-05 13:15:27 -0800941 * rq_ioc - determine io_context for request allocation
942 * @bio: request being allocated is for this bio (can be %NULL)
943 *
944 * Determine io_context to use for request allocation for @bio. May return
945 * %NULL if %current->io_context doesn't exist.
946 */
947static struct io_context *rq_ioc(struct bio *bio)
948{
949#ifdef CONFIG_BLK_CGROUP
950 if (bio && bio->bi_ioc)
951 return bio->bi_ioc;
952#endif
953 return current->io_context;
954}
955
956/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700957 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700958 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200959 * @rw_flags: RW and SYNC flags
960 * @bio: bio to allocate request for (can be %NULL)
961 * @gfp_mask: allocation mask
962 *
963 * Get a free request from @q. This function may fail under memory
964 * pressure or if @q is dead.
965 *
Masanari Iidada3dae52014-09-09 01:27:23 +0900966 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -0600967 * Returns ERR_PTR on failure, with @q->queue_lock held.
968 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700970static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700971 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700973 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800974 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700975 struct elevator_type *et = q->elevator->type;
976 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100977 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200978 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200979 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100981 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -0600982 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +0200983
Jens Axboe7749a8d2006-12-13 13:02:26 +0100984 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100985 if (may_queue == ELV_MQUEUE_NO)
986 goto rq_starved;
987
Jens Axboe1faa16d2009-04-06 14:48:01 +0200988 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
989 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100990 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100991 * The queue will fill after this allocation, so set
992 * it as full, and mark this process as "batching".
993 * This process will be allowed to complete a batch of
994 * requests, others will be blocked.
995 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700996 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100997 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700998 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100999 } else {
1000 if (may_queue != ELV_MQUEUE_MUST
1001 && !ioc_batching(q, ioc)) {
1002 /*
1003 * The queue is full and the allocating
1004 * process is not a "batcher", and not
1005 * exempted by the IO scheduler
1006 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001007 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001008 }
1009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
Tejun Heoa0516612012-06-26 15:05:44 -07001011 /*
1012 * bdi isn't aware of blkcg yet. As all async IOs end up
1013 * root blkcg anyway, just use root blkcg state.
1014 */
1015 if (rl == &q->root_rl)
1016 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
1018
Jens Axboe082cf692005-06-28 16:35:11 +02001019 /*
1020 * Only allow batching queuers to allocate up to 50% over the defined
1021 * limit of requests, otherwise we could have thousands of requests
1022 * allocated with any setting of ->nr_requests
1023 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001024 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001025 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001026
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001027 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001028 rl->count[is_sync]++;
1029 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001030
Tejun Heof1f8cc92011-12-14 00:33:42 +01001031 /*
1032 * Decide whether the new request will be managed by elevator. If
1033 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1034 * prevent the current elevator from being destroyed until the new
1035 * request is freed. This guarantees icq's won't be destroyed and
1036 * makes creating new ones safe.
1037 *
1038 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1039 * it will be created after releasing queue_lock.
1040 */
Tejun Heod7325802012-03-05 13:14:58 -08001041 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001042 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001043 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001044 if (et->icq_cache && ioc)
1045 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001046 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001047
Jens Axboef253b862010-10-24 22:06:02 +02001048 if (blk_queue_io_stat(q))
1049 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 spin_unlock_irq(q->queue_lock);
1051
Tejun Heo29e2b092012-04-19 16:29:21 -07001052 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001053 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001054 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001055 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Tejun Heo29e2b092012-04-19 16:29:21 -07001057 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001058 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001059 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1060
Tejun Heoaaf7c682012-04-19 16:29:22 -07001061 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001062 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001063 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001064 if (ioc)
1065 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001066 if (!icq)
1067 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001068 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001069
1070 rq->elv.icq = icq;
1071 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1072 goto fail_elvpriv;
1073
1074 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001075 if (icq)
1076 get_io_context(icq->ioc);
1077 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001078out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001079 /*
1080 * ioc may be NULL here, and ioc_batching will be false. That's
1081 * OK, if the queue is under the request limit then requests need
1082 * not count toward the nr_batch_requests limit. There will always
1083 * be some limit enforced by BLK_BATCH_TIME.
1084 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (ioc_batching(q, ioc))
1086 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001087
Jens Axboe1faa16d2009-04-06 14:48:01 +02001088 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001090
Tejun Heoaaf7c682012-04-19 16:29:22 -07001091fail_elvpriv:
1092 /*
1093 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1094 * and may fail indefinitely under memory pressure and thus
1095 * shouldn't stall IO. Treat this request as !elvpriv. This will
1096 * disturb iosched and blkcg but weird is bettern than dead.
1097 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001098 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1099 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001100
1101 rq->cmd_flags &= ~REQ_ELVPRIV;
1102 rq->elv.icq = NULL;
1103
1104 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001105 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001106 spin_unlock_irq(q->queue_lock);
1107 goto out;
1108
Tejun Heob6792812012-03-05 13:15:23 -08001109fail_alloc:
1110 /*
1111 * Allocation failed presumably due to memory. Undo anything we
1112 * might have messed up.
1113 *
1114 * Allocating task should really be put onto the front of the wait
1115 * queue, but this is pretty rare.
1116 */
1117 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001118 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001119
1120 /*
1121 * in the very unlikely event that allocation failed and no
1122 * requests for this direction was pending, mark us starved so that
1123 * freeing of a request in the other direction will notice
1124 * us. another possible fix would be to split the rq mempool into
1125 * READ and WRITE
1126 */
1127rq_starved:
1128 if (unlikely(rl->count[is_sync] == 0))
1129 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001130 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131}
1132
Tejun Heoda8303c2011-10-19 14:33:05 +02001133/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001134 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001135 * @q: request_queue to allocate request from
1136 * @rw_flags: RW and SYNC flags
1137 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001138 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001139 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001140 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1141 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001142 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001143 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001144 * Returns ERR_PTR on failure, with @q->queue_lock held.
1145 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001147static struct request *get_request(struct request_queue *q, int rw_flags,
1148 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001150 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001151 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001152 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001154
1155 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001156retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001157 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001158 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001159 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001161 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001162 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001163 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Tejun Heoa06e05e2012-06-04 20:40:55 -07001166 /* wait on @rl and retry */
1167 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1168 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001169
Tejun Heoa06e05e2012-06-04 20:40:55 -07001170 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Tejun Heoa06e05e2012-06-04 20:40:55 -07001172 spin_unlock_irq(q->queue_lock);
1173 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Tejun Heoa06e05e2012-06-04 20:40:55 -07001175 /*
1176 * After sleeping, we become a "batching" process and will be able
1177 * to allocate at least one request, and up to a big batch of them
1178 * for a small period time. See ioc_batching, ioc_set_batching
1179 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001180 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Tejun Heoa06e05e2012-06-04 20:40:55 -07001182 spin_lock_irq(q->queue_lock);
1183 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001184
Tejun Heoa06e05e2012-06-04 20:40:55 -07001185 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
Jens Axboe320ae512013-10-24 09:20:05 +01001188static struct request *blk_old_get_request(struct request_queue *q, int rw,
1189 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 struct request *rq;
1192
1193 BUG_ON(rw != READ && rw != WRITE);
1194
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001195 /* create ioc upfront */
1196 create_io_context(gfp_mask, q->node);
1197
Nick Piggind6344532005-06-28 20:45:14 -07001198 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001199 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001200 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001201 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001202 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 return rq;
1205}
Jens Axboe320ae512013-10-24 09:20:05 +01001206
1207struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1208{
1209 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001210 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001211 else
1212 return blk_old_get_request(q, rw, gfp_mask);
1213}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214EXPORT_SYMBOL(blk_get_request);
1215
1216/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001217 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001218 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001219 * @bio: The bio describing the memory mappings that will be submitted for IO.
1220 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001221 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001222 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001223 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1224 * type commands. Where the struct request needs to be farther initialized by
1225 * the caller. It is passed a &struct bio, which describes the memory info of
1226 * the I/O transfer.
1227 *
1228 * The caller of blk_make_request must make sure that bi_io_vec
1229 * are set to describe the memory buffers. That bio_data_dir() will return
1230 * the needed direction of the request. (And all bio's in the passed bio-chain
1231 * are properly set accordingly)
1232 *
1233 * If called under none-sleepable conditions, mapped bio buffers must not
1234 * need bouncing, by calling the appropriate masked or flagged allocator,
1235 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1236 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001237 *
1238 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1239 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1240 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1241 * completion of a bio that hasn't been submitted yet, thus resulting in a
1242 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1243 * of bio_alloc(), as that avoids the mempool deadlock.
1244 * If possible a big IO should be split into smaller parts when allocation
1245 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001246 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001247struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1248 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001249{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001250 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1251
Joe Lawrencea492f072014-08-28 08:15:21 -06001252 if (IS_ERR(rq))
1253 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001254
Jens Axboef27b0872014-06-06 07:57:37 -06001255 blk_rq_set_block_pc(rq);
1256
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001257 for_each_bio(bio) {
1258 struct bio *bounce_bio = bio;
1259 int ret;
1260
1261 blk_queue_bounce(q, &bounce_bio);
1262 ret = blk_rq_append_bio(q, rq, bounce_bio);
1263 if (unlikely(ret)) {
1264 blk_put_request(rq);
1265 return ERR_PTR(ret);
1266 }
1267 }
1268
1269 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001270}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001271EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001272
1273/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001274 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001275 * @rq: request to be initialized
1276 *
1277 */
1278void blk_rq_set_block_pc(struct request *rq)
1279{
1280 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1281 rq->__data_len = 0;
1282 rq->__sector = (sector_t) -1;
1283 rq->bio = rq->biotail = NULL;
1284 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001285}
1286EXPORT_SYMBOL(blk_rq_set_block_pc);
1287
1288/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 * blk_requeue_request - put a request back on queue
1290 * @q: request queue where request should be inserted
1291 * @rq: request to be inserted
1292 *
1293 * Description:
1294 * Drivers often keep queueing requests until the hardware cannot accept
1295 * more, when that condition happens we need to put the request back
1296 * on the queue. Must be called with queue lock held.
1297 */
Jens Axboe165125e2007-07-24 09:28:11 +02001298void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001300 blk_delete_timer(rq);
1301 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001302 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001303
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001304 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 blk_queue_end_tag(q, rq);
1306
James Bottomleyba396a62009-05-27 14:17:08 +02001307 BUG_ON(blk_queued_rq(rq));
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 elv_requeue_request(q, rq);
1310}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311EXPORT_SYMBOL(blk_requeue_request);
1312
Jens Axboe73c10102011-03-08 13:19:51 +01001313static void add_acct_request(struct request_queue *q, struct request *rq,
1314 int where)
1315{
Jens Axboe320ae512013-10-24 09:20:05 +01001316 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001317 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001318}
1319
Tejun Heo074a7ac2008-08-25 19:56:14 +09001320static void part_round_stats_single(int cpu, struct hd_struct *part,
1321 unsigned long now)
1322{
Jens Axboe7276d022014-05-09 15:48:23 -06001323 int inflight;
1324
Tejun Heo074a7ac2008-08-25 19:56:14 +09001325 if (now == part->stamp)
1326 return;
1327
Jens Axboe7276d022014-05-09 15:48:23 -06001328 inflight = part_in_flight(part);
1329 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001330 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001331 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001332 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1333 }
1334 part->stamp = now;
1335}
1336
1337/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001338 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1339 * @cpu: cpu number for stats access
1340 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 *
1342 * The average IO queue length and utilisation statistics are maintained
1343 * by observing the current state of the queue length and the amount of
1344 * time it has been in this state for.
1345 *
1346 * Normally, that accounting is done on IO completion, but that can result
1347 * in more than a second's worth of IO being accounted for within any one
1348 * second, leading to >100% utilisation. To deal with that, we call this
1349 * function to do a round-off before returning the results when reading
1350 * /proc/diskstats. This accounts immediately for all queue usage up to
1351 * the current jiffies and restarts the counters again.
1352 */
Tejun Heoc9959052008-08-25 19:47:21 +09001353void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001354{
1355 unsigned long now = jiffies;
1356
Tejun Heo074a7ac2008-08-25 19:56:14 +09001357 if (part->partno)
1358 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1359 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001360}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001361EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001362
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001363#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001364static void blk_pm_put_request(struct request *rq)
1365{
1366 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1367 pm_runtime_mark_last_busy(rq->q->dev);
1368}
1369#else
1370static inline void blk_pm_put_request(struct request *rq) {}
1371#endif
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373/*
1374 * queue lock must be held
1375 */
Jens Axboe165125e2007-07-24 09:28:11 +02001376void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 if (unlikely(!q))
1379 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001381 if (q->mq_ops) {
1382 blk_mq_free_request(req);
1383 return;
1384 }
1385
Lin Mingc8158812013-03-23 11:42:27 +08001386 blk_pm_put_request(req);
1387
Tejun Heo8922e162005-10-20 16:23:44 +02001388 elv_completed_request(q, req);
1389
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001390 /* this is a bio leak */
1391 WARN_ON(req->bio != NULL);
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 /*
1394 * Request may not have originated from ll_rw_blk. if not,
1395 * it didn't come out of our reserved rq pools
1396 */
Jens Axboe49171e52006-08-10 08:59:11 +02001397 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001398 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001399 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001402 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Tejun Heoa0516612012-06-26 15:05:44 -07001404 blk_free_request(rl, req);
1405 freed_request(rl, flags);
1406 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 }
1408}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001409EXPORT_SYMBOL_GPL(__blk_put_request);
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411void blk_put_request(struct request *req)
1412{
Jens Axboe165125e2007-07-24 09:28:11 +02001413 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Jens Axboe320ae512013-10-24 09:20:05 +01001415 if (q->mq_ops)
1416 blk_mq_free_request(req);
1417 else {
1418 unsigned long flags;
1419
1420 spin_lock_irqsave(q->queue_lock, flags);
1421 __blk_put_request(q, req);
1422 spin_unlock_irqrestore(q->queue_lock, flags);
1423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425EXPORT_SYMBOL(blk_put_request);
1426
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001427/**
1428 * blk_add_request_payload - add a payload to a request
1429 * @rq: request to update
1430 * @page: page backing the payload
1431 * @len: length of the payload.
1432 *
1433 * This allows to later add a payload to an already submitted request by
1434 * a block driver. The driver needs to take care of freeing the payload
1435 * itself.
1436 *
1437 * Note that this is a quite horrible hack and nothing but handling of
1438 * discard requests should ever use it.
1439 */
1440void blk_add_request_payload(struct request *rq, struct page *page,
1441 unsigned int len)
1442{
1443 struct bio *bio = rq->bio;
1444
1445 bio->bi_io_vec->bv_page = page;
1446 bio->bi_io_vec->bv_offset = 0;
1447 bio->bi_io_vec->bv_len = len;
1448
Kent Overstreet4f024f32013-10-11 15:44:27 -07001449 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001450 bio->bi_vcnt = 1;
1451 bio->bi_phys_segments = 1;
1452
1453 rq->__data_len = rq->resid_len = len;
1454 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001455}
1456EXPORT_SYMBOL_GPL(blk_add_request_payload);
1457
Jens Axboe320ae512013-10-24 09:20:05 +01001458bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1459 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001460{
1461 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1462
Jens Axboe73c10102011-03-08 13:19:51 +01001463 if (!ll_back_merge_fn(q, req, bio))
1464 return false;
1465
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001466 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001467
1468 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1469 blk_rq_set_mixed_merge(req);
1470
1471 req->biotail->bi_next = bio;
1472 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001473 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001474 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1475
Jens Axboe320ae512013-10-24 09:20:05 +01001476 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001477 return true;
1478}
1479
Jens Axboe320ae512013-10-24 09:20:05 +01001480bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1481 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001482{
1483 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001484
Jens Axboe73c10102011-03-08 13:19:51 +01001485 if (!ll_front_merge_fn(q, req, bio))
1486 return false;
1487
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001488 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001489
1490 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1491 blk_rq_set_mixed_merge(req);
1492
Jens Axboe73c10102011-03-08 13:19:51 +01001493 bio->bi_next = req->bio;
1494 req->bio = bio;
1495
Kent Overstreet4f024f32013-10-11 15:44:27 -07001496 req->__sector = bio->bi_iter.bi_sector;
1497 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001498 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1499
Jens Axboe320ae512013-10-24 09:20:05 +01001500 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001501 return true;
1502}
1503
Tejun Heobd87b582011-10-19 14:33:08 +02001504/**
Jens Axboe320ae512013-10-24 09:20:05 +01001505 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001506 * @q: request_queue new bio is being queued at
1507 * @bio: new bio being queued
1508 * @request_count: out parameter for number of traversed plugged requests
1509 *
1510 * Determine whether @bio being queued on @q can be merged with a request
1511 * on %current's plugged list. Returns %true if merge was successful,
1512 * otherwise %false.
1513 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001514 * Plugging coalesces IOs from the same issuer for the same purpose without
1515 * going through @q->queue_lock. As such it's more of an issuing mechanism
1516 * than scheduling, and the request, while may have elvpriv data, is not
1517 * added on the elevator at this point. In addition, we don't have
1518 * reliable access to the elevator outside queue lock. Only check basic
1519 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001520 *
1521 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001522 */
Jens Axboe320ae512013-10-24 09:20:05 +01001523bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1524 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001525{
1526 struct blk_plug *plug;
1527 struct request *rq;
1528 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001529 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001530
Tejun Heobd87b582011-10-19 14:33:08 +02001531 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001532 if (!plug)
1533 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001534 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001535
Shaohua Li92f399c2013-10-29 12:01:03 -06001536 if (q->mq_ops)
1537 plug_list = &plug->mq_list;
1538 else
1539 plug_list = &plug->list;
1540
1541 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001542 int el_ret;
1543
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001544 if (rq->q == q)
1545 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001546
Tejun Heo07c2bd32012-02-08 09:19:42 +01001547 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001548 continue;
1549
Tejun Heo050c8ea2012-02-08 09:19:38 +01001550 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001551 if (el_ret == ELEVATOR_BACK_MERGE) {
1552 ret = bio_attempt_back_merge(q, rq, bio);
1553 if (ret)
1554 break;
1555 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1556 ret = bio_attempt_front_merge(q, rq, bio);
1557 if (ret)
1558 break;
1559 }
1560 }
1561out:
1562 return ret;
1563}
1564
Jens Axboe86db1e22008-01-29 14:53:40 +01001565void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001566{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001567 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001568
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001569 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1570 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001571 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001572
Tejun Heo52d9e672006-01-06 09:49:58 +01001573 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001574 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001575 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001576 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001577}
1578
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001579void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001581 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001582 struct blk_plug *plug;
1583 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1584 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001585 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 /*
1588 * low level driver can indicate that it wants pages above a
1589 * certain limit bounced to low memory (ie for highmem, or even
1590 * ISA dma in theory)
1591 */
1592 blk_queue_bounce(q, &bio);
1593
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001594 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1595 bio_endio(bio, -EIO);
1596 return;
1597 }
1598
Tejun Heo4fed9472010-09-03 11:56:17 +02001599 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001600 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001601 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001602 goto get_rq;
1603 }
1604
Jens Axboe73c10102011-03-08 13:19:51 +01001605 /*
1606 * Check if we can merge with the plugged list before grabbing
1607 * any locks.
1608 */
Robert Elliottda41a582014-05-20 16:46:26 -05001609 if (!blk_queue_nomerges(q) &&
1610 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001611 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001612
1613 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
1615 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001616 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001617 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001618 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001619 if (!attempt_back_merge(q, req))
1620 elv_merged_request(q, req, el_ret);
1621 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001622 }
Jens Axboe73c10102011-03-08 13:19:51 +01001623 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001624 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001625 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001626 if (!attempt_front_merge(q, req))
1627 elv_merged_request(q, req, el_ret);
1628 goto out_unlock;
1629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 }
1631
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001633 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001634 * This sync check and mask will be re-done in init_request_from_bio(),
1635 * but we need to set it earlier to expose the sync flag to the
1636 * rq allocator and io schedulers.
1637 */
1638 rw_flags = bio_data_dir(bio);
1639 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001640 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001641
1642 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001643 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001644 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001645 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001646 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001647 if (IS_ERR(req)) {
1648 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
Tejun Heoda8303c2011-10-19 14:33:05 +02001649 goto out_unlock;
1650 }
Nick Piggind6344532005-06-28 20:45:14 -07001651
Nick Piggin450991b2005-06-28 20:45:13 -07001652 /*
1653 * After dropping the lock and possibly sleeping here, our request
1654 * may now be mergeable after it had proven unmergeable (above).
1655 * We don't worry about that case for efficiency. It won't happen
1656 * often, and the elevators are able to handle it.
1657 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001658 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Tao Ma9562ad92011-10-24 16:11:30 +02001660 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001661 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001662
Jens Axboe73c10102011-03-08 13:19:51 +01001663 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001664 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001665 /*
1666 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001667 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001668 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001669 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001670 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001671 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001672 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001673 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001674 trace_block_plug(q);
1675 }
Jens Axboe73c10102011-03-08 13:19:51 +01001676 }
Shaohua Lia6327162011-08-24 16:04:32 +02001677 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001678 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001679 } else {
1680 spin_lock_irq(q->queue_lock);
1681 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001682 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001683out_unlock:
1684 spin_unlock_irq(q->queue_lock);
1685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686}
Jens Axboec20e8de2011-09-12 12:03:37 +02001687EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689/*
1690 * If bio->bi_dev is a partition, remap the location
1691 */
1692static inline void blk_partition_remap(struct bio *bio)
1693{
1694 struct block_device *bdev = bio->bi_bdev;
1695
Jens Axboebf2de6f2007-09-27 13:01:25 +02001696 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001698
Kent Overstreet4f024f32013-10-11 15:44:27 -07001699 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001701
Mike Snitzerd07335e2010-11-16 12:52:38 +01001702 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1703 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001704 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
1706}
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708static void handle_bad_sector(struct bio *bio)
1709{
1710 char b[BDEVNAME_SIZE];
1711
1712 printk(KERN_INFO "attempt to access beyond end of device\n");
1713 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1714 bdevname(bio->bi_bdev, b),
1715 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001716 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001717 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719 set_bit(BIO_EOF, &bio->bi_flags);
1720}
1721
Akinobu Mitac17bb492006-12-08 02:39:46 -08001722#ifdef CONFIG_FAIL_MAKE_REQUEST
1723
1724static DECLARE_FAULT_ATTR(fail_make_request);
1725
1726static int __init setup_fail_make_request(char *str)
1727{
1728 return setup_fault_attr(&fail_make_request, str);
1729}
1730__setup("fail_make_request=", setup_fail_make_request);
1731
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001732static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001733{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001734 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001735}
1736
1737static int __init fail_make_request_debugfs(void)
1738{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001739 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1740 NULL, &fail_make_request);
1741
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001742 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001743}
1744
1745late_initcall(fail_make_request_debugfs);
1746
1747#else /* CONFIG_FAIL_MAKE_REQUEST */
1748
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001749static inline bool should_fail_request(struct hd_struct *part,
1750 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001751{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001752 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001753}
1754
1755#endif /* CONFIG_FAIL_MAKE_REQUEST */
1756
Jens Axboec07e2b42007-07-18 13:27:58 +02001757/*
1758 * Check whether this bio extends beyond the end of the device.
1759 */
1760static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1761{
1762 sector_t maxsector;
1763
1764 if (!nr_sectors)
1765 return 0;
1766
1767 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001768 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001769 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001770 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001771
1772 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1773 /*
1774 * This may well happen - the kernel calls bread()
1775 * without checking the size of the device, e.g., when
1776 * mounting a device.
1777 */
1778 handle_bad_sector(bio);
1779 return 1;
1780 }
1781 }
1782
1783 return 0;
1784}
1785
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001786static noinline_for_stack bool
1787generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788{
Jens Axboe165125e2007-07-24 09:28:11 +02001789 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001790 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001791 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001792 char b[BDEVNAME_SIZE];
1793 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
1795 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Jens Axboec07e2b42007-07-18 13:27:58 +02001797 if (bio_check_eod(bio, nr_sectors))
1798 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001800 q = bdev_get_queue(bio->bi_bdev);
1801 if (unlikely(!q)) {
1802 printk(KERN_ERR
1803 "generic_make_request: Trying to access "
1804 "nonexistent block-device %s (%Lu)\n",
1805 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001806 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001807 goto end_io;
1808 }
1809
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001810 if (likely(bio_is_rw(bio) &&
1811 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001812 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1813 bdevname(bio->bi_bdev, b),
1814 bio_sectors(bio),
1815 queue_max_hw_sectors(q));
1816 goto end_io;
1817 }
1818
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001819 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001820 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001821 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001822 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001823 goto end_io;
1824
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001826 * If this device has partitions, remap block n
1827 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001829 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001831 if (bio_check_eod(bio, nr_sectors))
1832 goto end_io;
1833
1834 /*
1835 * Filter flush bio's early so that make_request based
1836 * drivers without flush support don't have to worry
1837 * about them.
1838 */
1839 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1840 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1841 if (!nr_sectors) {
1842 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001843 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001847 if ((bio->bi_rw & REQ_DISCARD) &&
1848 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001849 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001850 err = -EOPNOTSUPP;
1851 goto end_io;
1852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001854 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 err = -EOPNOTSUPP;
1856 goto end_io;
1857 }
1858
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001859 /*
1860 * Various block parts want %current->io_context and lazy ioc
1861 * allocation ends up trading a lot of pain for a small amount of
1862 * memory. Just allocate it upfront. This may fail and block
1863 * layer knows how to live with it.
1864 */
1865 create_io_context(GFP_ATOMIC, q->node);
1866
Tejun Heobc16a4f2011-10-19 14:33:01 +02001867 if (blk_throtl_bio(q, bio))
1868 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001869
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001870 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001871 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001872
1873end_io:
1874 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001875 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876}
1877
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001878/**
1879 * generic_make_request - hand a buffer to its device driver for I/O
1880 * @bio: The bio describing the location in memory and on the device.
1881 *
1882 * generic_make_request() is used to make I/O requests of block
1883 * devices. It is passed a &struct bio, which describes the I/O that needs
1884 * to be done.
1885 *
1886 * generic_make_request() does not return any status. The
1887 * success/failure status of the request, along with notification of
1888 * completion, is delivered asynchronously through the bio->bi_end_io
1889 * function described (one day) else where.
1890 *
1891 * The caller of generic_make_request must make sure that bi_io_vec
1892 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1893 * set to describe the device address, and the
1894 * bi_end_io and optionally bi_private are set to describe how
1895 * completion notification should be signaled.
1896 *
1897 * generic_make_request and the drivers it calls may use bi_next if this
1898 * bio happens to be merged with someone else, and may resubmit the bio to
1899 * a lower device by calling into generic_make_request recursively, which
1900 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001901 */
1902void generic_make_request(struct bio *bio)
1903{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001904 struct bio_list bio_list_on_stack;
1905
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001906 if (!generic_make_request_checks(bio))
1907 return;
1908
1909 /*
1910 * We only want one ->make_request_fn to be active at a time, else
1911 * stack usage with stacked devices could be a problem. So use
1912 * current->bio_list to keep a list of requests submited by a
1913 * make_request_fn function. current->bio_list is also used as a
1914 * flag to say if generic_make_request is currently active in this
1915 * task or not. If it is NULL, then no make_request is active. If
1916 * it is non-NULL, then a make_request is active, and new requests
1917 * should be added at the tail
1918 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001919 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001920 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001921 return;
1922 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001923
Neil Brownd89d8792007-05-01 09:53:42 +02001924 /* following loop may be a bit non-obvious, and so deserves some
1925 * explanation.
1926 * Before entering the loop, bio->bi_next is NULL (as all callers
1927 * ensure that) so we have a list with a single bio.
1928 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001929 * we assign bio_list to a pointer to the bio_list_on_stack,
1930 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001931 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001932 * through a recursive call to generic_make_request. If it
1933 * did, we find a non-NULL value in bio_list and re-enter the loop
1934 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001935 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001936 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001937 */
1938 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001939 bio_list_init(&bio_list_on_stack);
1940 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001941 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001942 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1943
1944 q->make_request_fn(q, bio);
1945
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001946 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001947 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001948 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001949}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950EXPORT_SYMBOL(generic_make_request);
1951
1952/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001953 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1955 * @bio: The &struct bio which describes the I/O
1956 *
1957 * submit_bio() is very similar in purpose to generic_make_request(), and
1958 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001959 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 *
1961 */
1962void submit_bio(int rw, struct bio *bio)
1963{
Jens Axboe22e2c502005-06-27 10:55:12 +02001964 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Jens Axboebf2de6f2007-09-27 13:01:25 +02001966 /*
1967 * If it's a regular read/write or a barrier with data attached,
1968 * go through the normal accounting stuff before submission.
1969 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001970 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001971 unsigned int count;
1972
1973 if (unlikely(rw & REQ_WRITE_SAME))
1974 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1975 else
1976 count = bio_sectors(bio);
1977
Jens Axboebf2de6f2007-09-27 13:01:25 +02001978 if (rw & WRITE) {
1979 count_vm_events(PGPGOUT, count);
1980 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001981 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001982 count_vm_events(PGPGIN, count);
1983 }
1984
1985 if (unlikely(block_dump)) {
1986 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001987 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001988 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001989 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001990 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001991 bdevname(bio->bi_bdev, b),
1992 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
1995
1996 generic_make_request(bio);
1997}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998EXPORT_SYMBOL(submit_bio);
1999
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002000/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002001 * blk_rq_check_limits - Helper function to check a request for the queue limit
2002 * @q: the queue
2003 * @rq: the request being checked
2004 *
2005 * Description:
2006 * @rq may have been made based on weaker limitations of upper-level queues
2007 * in request stacking drivers, and it may violate the limitation of @q.
2008 * Since the block layer and the underlying device driver trust @rq
2009 * after it is inserted to @q, it should be checked against @q before
2010 * the insertion using this generic function.
2011 *
2012 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02002013 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002014 * Request stacking drivers like request-based dm may change the queue
2015 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09002016 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002017 * the new queue limits again when they dispatch those requests,
2018 * although such checkings are also done against the old queue limits
2019 * when submitting requests.
2020 */
2021int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2022{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002023 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04002024 return 0;
2025
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04002026 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002027 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2028 return -EIO;
2029 }
2030
2031 /*
2032 * queue's settings related to segment counting like q->bounce_pfn
2033 * may differ from that of other stacking queues.
2034 * Recalculate it to check the request correctly on this queue's
2035 * limitation.
2036 */
2037 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002038 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002039 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2040 return -EIO;
2041 }
2042
2043 return 0;
2044}
2045EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2046
2047/**
2048 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2049 * @q: the queue to submit the request
2050 * @rq: the request being queued
2051 */
2052int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2053{
2054 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002055 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002056
2057 if (blk_rq_check_limits(q, rq))
2058 return -EIO;
2059
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002060 if (rq->rq_disk &&
2061 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002062 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002063
Keith Busch7fb48982014-10-17 17:46:38 -06002064 if (q->mq_ops) {
2065 if (blk_queue_io_stat(q))
2066 blk_account_io_start(rq, true);
2067 blk_mq_insert_request(rq, false, true, true);
2068 return 0;
2069 }
2070
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002071 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002072 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002073 spin_unlock_irqrestore(q->queue_lock, flags);
2074 return -ENODEV;
2075 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002076
2077 /*
2078 * Submitting request must be dequeued before calling this function
2079 * because it will be linked to another request_queue
2080 */
2081 BUG_ON(blk_queued_rq(rq));
2082
Jeff Moyer4853aba2011-08-15 21:37:25 +02002083 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2084 where = ELEVATOR_INSERT_FLUSH;
2085
2086 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002087 if (where == ELEVATOR_INSERT_FLUSH)
2088 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002089 spin_unlock_irqrestore(q->queue_lock, flags);
2090
2091 return 0;
2092}
2093EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2094
Tejun Heo80a761f2009-07-03 17:48:17 +09002095/**
2096 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2097 * @rq: request to examine
2098 *
2099 * Description:
2100 * A request could be merge of IOs which require different failure
2101 * handling. This function determines the number of bytes which
2102 * can be failed from the beginning of the request without
2103 * crossing into area which need to be retried further.
2104 *
2105 * Return:
2106 * The number of bytes to fail.
2107 *
2108 * Context:
2109 * queue_lock must be held.
2110 */
2111unsigned int blk_rq_err_bytes(const struct request *rq)
2112{
2113 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2114 unsigned int bytes = 0;
2115 struct bio *bio;
2116
2117 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2118 return blk_rq_bytes(rq);
2119
2120 /*
2121 * Currently the only 'mixing' which can happen is between
2122 * different fastfail types. We can safely fail portions
2123 * which have all the failfast bits that the first one has -
2124 * the ones which are at least as eager to fail as the first
2125 * one.
2126 */
2127 for (bio = rq->bio; bio; bio = bio->bi_next) {
2128 if ((bio->bi_rw & ff) != ff)
2129 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002130 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002131 }
2132
2133 /* this could lead to infinite loop */
2134 BUG_ON(blk_rq_bytes(rq) && !bytes);
2135 return bytes;
2136}
2137EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2138
Jens Axboe320ae512013-10-24 09:20:05 +01002139void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002140{
Jens Axboec2553b52009-04-24 08:10:11 +02002141 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002142 const int rw = rq_data_dir(req);
2143 struct hd_struct *part;
2144 int cpu;
2145
2146 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002147 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002148 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2149 part_stat_unlock();
2150 }
2151}
2152
Jens Axboe320ae512013-10-24 09:20:05 +01002153void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002154{
Jens Axboebc58ba92009-01-23 10:54:44 +01002155 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002156 * Account IO completion. flush_rq isn't accounted as a
2157 * normal IO on queueing nor completion. Accounting the
2158 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002159 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002160 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002161 unsigned long duration = jiffies - req->start_time;
2162 const int rw = rq_data_dir(req);
2163 struct hd_struct *part;
2164 int cpu;
2165
2166 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002167 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002168
2169 part_stat_inc(cpu, part, ios[rw]);
2170 part_stat_add(cpu, part, ticks[rw], duration);
2171 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002172 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002173
Jens Axboe6c23a962011-01-07 08:43:37 +01002174 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002175 part_stat_unlock();
2176 }
2177}
2178
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002179#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002180/*
2181 * Don't process normal requests when queue is suspended
2182 * or in the process of suspending/resuming
2183 */
2184static struct request *blk_pm_peek_request(struct request_queue *q,
2185 struct request *rq)
2186{
2187 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2188 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2189 return NULL;
2190 else
2191 return rq;
2192}
2193#else
2194static inline struct request *blk_pm_peek_request(struct request_queue *q,
2195 struct request *rq)
2196{
2197 return rq;
2198}
2199#endif
2200
Jens Axboe320ae512013-10-24 09:20:05 +01002201void blk_account_io_start(struct request *rq, bool new_io)
2202{
2203 struct hd_struct *part;
2204 int rw = rq_data_dir(rq);
2205 int cpu;
2206
2207 if (!blk_do_io_stat(rq))
2208 return;
2209
2210 cpu = part_stat_lock();
2211
2212 if (!new_io) {
2213 part = rq->part;
2214 part_stat_inc(cpu, part, merges[rw]);
2215 } else {
2216 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2217 if (!hd_struct_try_get(part)) {
2218 /*
2219 * The partition is already being removed,
2220 * the request will be accounted on the disk only
2221 *
2222 * We take a reference on disk->part0 although that
2223 * partition will never be deleted, so we can treat
2224 * it as any other partition.
2225 */
2226 part = &rq->rq_disk->part0;
2227 hd_struct_get(part);
2228 }
2229 part_round_stats(cpu, part);
2230 part_inc_in_flight(part, rw);
2231 rq->part = part;
2232 }
2233
2234 part_stat_unlock();
2235}
2236
Tejun Heo53a08802008-12-03 12:41:26 +01002237/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002238 * blk_peek_request - peek at the top of a request queue
2239 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002240 *
2241 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002242 * Return the request at the top of @q. The returned request
2243 * should be started using blk_start_request() before LLD starts
2244 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002245 *
2246 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002247 * Pointer to the request at the top of @q if available. Null
2248 * otherwise.
2249 *
2250 * Context:
2251 * queue_lock must be held.
2252 */
2253struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002254{
2255 struct request *rq;
2256 int ret;
2257
2258 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002259
2260 rq = blk_pm_peek_request(q, rq);
2261 if (!rq)
2262 break;
2263
Tejun Heo158dbda2009-04-23 11:05:18 +09002264 if (!(rq->cmd_flags & REQ_STARTED)) {
2265 /*
2266 * This is the first time the device driver
2267 * sees this request (possibly after
2268 * requeueing). Notify IO scheduler.
2269 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002270 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002271 elv_activate_rq(q, rq);
2272
2273 /*
2274 * just mark as started even if we don't start
2275 * it, a request that has been delayed should
2276 * not be passed by new incoming requests
2277 */
2278 rq->cmd_flags |= REQ_STARTED;
2279 trace_block_rq_issue(q, rq);
2280 }
2281
2282 if (!q->boundary_rq || q->boundary_rq == rq) {
2283 q->end_sector = rq_end_sector(rq);
2284 q->boundary_rq = NULL;
2285 }
2286
2287 if (rq->cmd_flags & REQ_DONTPREP)
2288 break;
2289
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002290 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002291 /*
2292 * make sure space for the drain appears we
2293 * know we can do this because max_hw_segments
2294 * has been adjusted to be one fewer than the
2295 * device can handle
2296 */
2297 rq->nr_phys_segments++;
2298 }
2299
2300 if (!q->prep_rq_fn)
2301 break;
2302
2303 ret = q->prep_rq_fn(q, rq);
2304 if (ret == BLKPREP_OK) {
2305 break;
2306 } else if (ret == BLKPREP_DEFER) {
2307 /*
2308 * the request may have been (partially) prepped.
2309 * we need to keep this request in the front to
2310 * avoid resource deadlock. REQ_STARTED will
2311 * prevent other fs requests from passing this one.
2312 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002313 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002314 !(rq->cmd_flags & REQ_DONTPREP)) {
2315 /*
2316 * remove the space for the drain we added
2317 * so that we don't add it again
2318 */
2319 --rq->nr_phys_segments;
2320 }
2321
2322 rq = NULL;
2323 break;
2324 } else if (ret == BLKPREP_KILL) {
2325 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002326 /*
2327 * Mark this request as started so we don't trigger
2328 * any debug logic in the end I/O path.
2329 */
2330 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002331 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002332 } else {
2333 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2334 break;
2335 }
2336 }
2337
2338 return rq;
2339}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002340EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002341
Tejun Heo9934c8c2009-05-08 11:54:16 +09002342void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002343{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002344 struct request_queue *q = rq->q;
2345
Tejun Heo158dbda2009-04-23 11:05:18 +09002346 BUG_ON(list_empty(&rq->queuelist));
2347 BUG_ON(ELV_ON_HASH(rq));
2348
2349 list_del_init(&rq->queuelist);
2350
2351 /*
2352 * the time frame between a request being removed from the lists
2353 * and to it is freed is accounted as io that is in progress at
2354 * the driver side.
2355 */
Divyesh Shah91952912010-04-01 15:01:41 -07002356 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002357 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002358 set_io_start_time_ns(rq);
2359 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002360}
2361
Tejun Heo5efccd12009-04-23 11:05:18 +09002362/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002363 * blk_start_request - start request processing on the driver
2364 * @req: request to dequeue
2365 *
2366 * Description:
2367 * Dequeue @req and start timeout timer on it. This hands off the
2368 * request to the driver.
2369 *
2370 * Block internal functions which don't want to start timer should
2371 * call blk_dequeue_request().
2372 *
2373 * Context:
2374 * queue_lock must be held.
2375 */
2376void blk_start_request(struct request *req)
2377{
2378 blk_dequeue_request(req);
2379
2380 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002381 * We are now handing the request to the hardware, initialize
2382 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002383 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002384 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002385 if (unlikely(blk_bidi_rq(req)))
2386 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2387
Jeff Moyer4912aa62013-10-08 14:36:41 -04002388 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002389 blk_add_timer(req);
2390}
2391EXPORT_SYMBOL(blk_start_request);
2392
2393/**
2394 * blk_fetch_request - fetch a request from a request queue
2395 * @q: request queue to fetch a request from
2396 *
2397 * Description:
2398 * Return the request at the top of @q. The request is started on
2399 * return and LLD can start processing it immediately.
2400 *
2401 * Return:
2402 * Pointer to the request at the top of @q if available. Null
2403 * otherwise.
2404 *
2405 * Context:
2406 * queue_lock must be held.
2407 */
2408struct request *blk_fetch_request(struct request_queue *q)
2409{
2410 struct request *rq;
2411
2412 rq = blk_peek_request(q);
2413 if (rq)
2414 blk_start_request(rq);
2415 return rq;
2416}
2417EXPORT_SYMBOL(blk_fetch_request);
2418
2419/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002420 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002421 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002422 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002423 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002424 *
2425 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002426 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2427 * the request structure even if @req doesn't have leftover.
2428 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002429 *
2430 * This special helper function is only for request stacking drivers
2431 * (e.g. request-based dm) so that they can handle partial completion.
2432 * Actual device drivers should use blk_end_request instead.
2433 *
2434 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2435 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002436 *
2437 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002438 * %false - this request doesn't have any more data
2439 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002440 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002441bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002443 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002445 trace_block_rq_complete(req->q, req, nr_bytes);
2446
Tejun Heo2e60e022009-04-23 11:05:18 +09002447 if (!req->bio)
2448 return false;
2449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002451 * For fs requests, rq is just carrier of independent bio's
2452 * and each partial completion should be handled separately.
2453 * Reset per-request error on each partial completion.
2454 *
2455 * TODO: tj: This is too subtle. It would be better to let
2456 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002458 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 req->errors = 0;
2460
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002461 if (error && req->cmd_type == REQ_TYPE_FS &&
2462 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002463 char *error_type;
2464
2465 switch (error) {
2466 case -ENOLINK:
2467 error_type = "recoverable transport";
2468 break;
2469 case -EREMOTEIO:
2470 error_type = "critical target";
2471 break;
2472 case -EBADE:
2473 error_type = "critical nexus";
2474 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002475 case -ETIMEDOUT:
2476 error_type = "timeout";
2477 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002478 case -ENOSPC:
2479 error_type = "critical space allocation";
2480 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002481 case -ENODATA:
2482 error_type = "critical medium";
2483 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002484 case -EIO:
2485 default:
2486 error_type = "I/O";
2487 break;
2488 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002489 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2490 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002491 req->rq_disk->disk_name : "?",
2492 (unsigned long long)blk_rq_pos(req));
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 }
2495
Jens Axboebc58ba92009-01-23 10:54:44 +01002496 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002497
Kent Overstreetf79ea412012-09-20 16:38:30 -07002498 total_bytes = 0;
2499 while (req->bio) {
2500 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002501 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Kent Overstreet4f024f32013-10-11 15:44:27 -07002503 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Kent Overstreetf79ea412012-09-20 16:38:30 -07002506 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
Kent Overstreetf79ea412012-09-20 16:38:30 -07002508 total_bytes += bio_bytes;
2509 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Kent Overstreetf79ea412012-09-20 16:38:30 -07002511 if (!nr_bytes)
2512 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 }
2514
2515 /*
2516 * completely done
2517 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002518 if (!req->bio) {
2519 /*
2520 * Reset counters so that the request stacking driver
2521 * can find how many bytes remain in the request
2522 * later.
2523 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002524 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002525 return false;
2526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002528 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002529
2530 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002531 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002532 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002533
Tejun Heo80a761f2009-07-03 17:48:17 +09002534 /* mixed attributes always follow the first bio */
2535 if (req->cmd_flags & REQ_MIXED_MERGE) {
2536 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2537 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2538 }
2539
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002540 /*
2541 * If total number of sectors is less than the first segment
2542 * size, something has gone terribly wrong.
2543 */
2544 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002545 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002546 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002547 }
2548
2549 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002551
Tejun Heo2e60e022009-04-23 11:05:18 +09002552 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553}
Tejun Heo2e60e022009-04-23 11:05:18 +09002554EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555
Tejun Heo2e60e022009-04-23 11:05:18 +09002556static bool blk_update_bidi_request(struct request *rq, int error,
2557 unsigned int nr_bytes,
2558 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002559{
Tejun Heo2e60e022009-04-23 11:05:18 +09002560 if (blk_update_request(rq, error, nr_bytes))
2561 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002562
Tejun Heo2e60e022009-04-23 11:05:18 +09002563 /* Bidi request must be completed as a whole */
2564 if (unlikely(blk_bidi_rq(rq)) &&
2565 blk_update_request(rq->next_rq, error, bidi_bytes))
2566 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002567
Jens Axboee2e1a142010-06-09 10:42:09 +02002568 if (blk_queue_add_random(rq->q))
2569 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002570
2571 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572}
2573
James Bottomley28018c22010-07-01 19:49:17 +09002574/**
2575 * blk_unprep_request - unprepare a request
2576 * @req: the request
2577 *
2578 * This function makes a request ready for complete resubmission (or
2579 * completion). It happens only after all error handling is complete,
2580 * so represents the appropriate moment to deallocate any resources
2581 * that were allocated to the request in the prep_rq_fn. The queue
2582 * lock is held when calling this.
2583 */
2584void blk_unprep_request(struct request *req)
2585{
2586 struct request_queue *q = req->q;
2587
2588 req->cmd_flags &= ~REQ_DONTPREP;
2589 if (q->unprep_rq_fn)
2590 q->unprep_rq_fn(q, req);
2591}
2592EXPORT_SYMBOL_GPL(blk_unprep_request);
2593
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594/*
2595 * queue lock must be held
2596 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002597void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002599 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002600 blk_queue_end_tag(req->q, req);
2601
James Bottomleyba396a62009-05-27 14:17:08 +02002602 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002604 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002605 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606
Mike Andersone78042e2008-10-30 02:16:20 -07002607 blk_delete_timer(req);
2608
James Bottomley28018c22010-07-01 19:49:17 +09002609 if (req->cmd_flags & REQ_DONTPREP)
2610 blk_unprep_request(req);
2611
Jens Axboebc58ba92009-01-23 10:54:44 +01002612 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002613
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002615 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002616 else {
2617 if (blk_bidi_rq(req))
2618 __blk_put_request(req->next_rq->q, req->next_rq);
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 }
2622}
Christoph Hellwig12120072014-04-16 09:44:59 +02002623EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002625/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002626 * blk_end_bidi_request - Complete a bidi request
2627 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002628 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002629 * @nr_bytes: number of bytes to complete @rq
2630 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002631 *
2632 * Description:
2633 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002634 * Drivers that supports bidi can safely call this member for any
2635 * type of request, bidi or uni. In the later case @bidi_bytes is
2636 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002637 *
2638 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002639 * %false - we are done with this request
2640 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002641 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002642static bool blk_end_bidi_request(struct request *rq, int error,
2643 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002644{
2645 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002646 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002647
Tejun Heo2e60e022009-04-23 11:05:18 +09002648 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2649 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002650
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002651 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002652 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002653 spin_unlock_irqrestore(q->queue_lock, flags);
2654
Tejun Heo2e60e022009-04-23 11:05:18 +09002655 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002656}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002657
2658/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002659 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2660 * @rq: the request to complete
2661 * @error: %0 for success, < %0 for error
2662 * @nr_bytes: number of bytes to complete @rq
2663 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002664 *
2665 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002666 * Identical to blk_end_bidi_request() except that queue lock is
2667 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002668 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002669 * Return:
2670 * %false - we are done with this request
2671 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002672 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002673bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002674 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002675{
Tejun Heo2e60e022009-04-23 11:05:18 +09002676 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2677 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002678
Tejun Heo2e60e022009-04-23 11:05:18 +09002679 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002680
Tejun Heo2e60e022009-04-23 11:05:18 +09002681 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002682}
2683
2684/**
2685 * blk_end_request - Helper function for drivers to complete the request.
2686 * @rq: the request being processed
2687 * @error: %0 for success, < %0 for error
2688 * @nr_bytes: number of bytes to complete
2689 *
2690 * Description:
2691 * Ends I/O on a number of bytes attached to @rq.
2692 * If @rq has leftover, sets it up for the next range of segments.
2693 *
2694 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002695 * %false - we are done with this request
2696 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002697 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002698bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002699{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002700 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002701}
Jens Axboe56ad1742009-07-28 22:11:24 +02002702EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002703
2704/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002705 * blk_end_request_all - Helper function for drives to finish the request.
2706 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002707 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002708 *
2709 * Description:
2710 * Completely finish @rq.
2711 */
2712void blk_end_request_all(struct request *rq, int error)
2713{
2714 bool pending;
2715 unsigned int bidi_bytes = 0;
2716
2717 if (unlikely(blk_bidi_rq(rq)))
2718 bidi_bytes = blk_rq_bytes(rq->next_rq);
2719
2720 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2721 BUG_ON(pending);
2722}
Jens Axboe56ad1742009-07-28 22:11:24 +02002723EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002724
2725/**
2726 * blk_end_request_cur - Helper function to finish the current request chunk.
2727 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002728 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002729 *
2730 * Description:
2731 * Complete the current consecutively mapped chunk from @rq.
2732 *
2733 * Return:
2734 * %false - we are done with this request
2735 * %true - still buffers pending for this request
2736 */
2737bool blk_end_request_cur(struct request *rq, int error)
2738{
2739 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2740}
Jens Axboe56ad1742009-07-28 22:11:24 +02002741EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002742
2743/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002744 * blk_end_request_err - Finish a request till the next failure boundary.
2745 * @rq: the request to finish till the next failure boundary for
2746 * @error: must be negative errno
2747 *
2748 * Description:
2749 * Complete @rq till the next failure boundary.
2750 *
2751 * Return:
2752 * %false - we are done with this request
2753 * %true - still buffers pending for this request
2754 */
2755bool blk_end_request_err(struct request *rq, int error)
2756{
2757 WARN_ON(error >= 0);
2758 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2759}
2760EXPORT_SYMBOL_GPL(blk_end_request_err);
2761
2762/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002763 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002764 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002765 * @error: %0 for success, < %0 for error
2766 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002767 *
2768 * Description:
2769 * Must be called with queue lock held unlike blk_end_request().
2770 *
2771 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002772 * %false - we are done with this request
2773 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002774 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002775bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002776{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002777 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002778}
Jens Axboe56ad1742009-07-28 22:11:24 +02002779EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002780
2781/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002782 * __blk_end_request_all - Helper function for drives to finish the request.
2783 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002784 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002785 *
2786 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002787 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002788 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002789void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002790{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002791 bool pending;
2792 unsigned int bidi_bytes = 0;
2793
2794 if (unlikely(blk_bidi_rq(rq)))
2795 bidi_bytes = blk_rq_bytes(rq->next_rq);
2796
2797 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2798 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002799}
Jens Axboe56ad1742009-07-28 22:11:24 +02002800EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002801
2802/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002803 * __blk_end_request_cur - Helper function to finish the current request chunk.
2804 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002805 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002806 *
2807 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002808 * Complete the current consecutively mapped chunk from @rq. Must
2809 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002810 *
2811 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002812 * %false - we are done with this request
2813 * %true - still buffers pending for this request
2814 */
2815bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002816{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002817 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002818}
Jens Axboe56ad1742009-07-28 22:11:24 +02002819EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002820
Tejun Heo80a761f2009-07-03 17:48:17 +09002821/**
2822 * __blk_end_request_err - Finish a request till the next failure boundary.
2823 * @rq: the request to finish till the next failure boundary for
2824 * @error: must be negative errno
2825 *
2826 * Description:
2827 * Complete @rq till the next failure boundary. Must be called
2828 * with queue lock held.
2829 *
2830 * Return:
2831 * %false - we are done with this request
2832 * %true - still buffers pending for this request
2833 */
2834bool __blk_end_request_err(struct request *rq, int error)
2835{
2836 WARN_ON(error >= 0);
2837 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2838}
2839EXPORT_SYMBOL_GPL(__blk_end_request_err);
2840
Jens Axboe86db1e22008-01-29 14:53:40 +01002841void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2842 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002844 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002845 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846
Jens Axboeb4f42e22014-04-10 09:46:28 -06002847 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002848 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002849
Kent Overstreet4f024f32013-10-11 15:44:27 -07002850 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
NeilBrown66846572007-08-16 13:31:28 +02002853 if (bio->bi_bdev)
2854 rq->rq_disk = bio->bi_bdev->bd_disk;
2855}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002857#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2858/**
2859 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2860 * @rq: the request to be flushed
2861 *
2862 * Description:
2863 * Flush all pages in @rq.
2864 */
2865void rq_flush_dcache_pages(struct request *rq)
2866{
2867 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002868 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002869
2870 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002871 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002872}
2873EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2874#endif
2875
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002876/**
2877 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2878 * @q : the queue of the device being checked
2879 *
2880 * Description:
2881 * Check if underlying low-level drivers of a device are busy.
2882 * If the drivers want to export their busy state, they must set own
2883 * exporting function using blk_queue_lld_busy() first.
2884 *
2885 * Basically, this function is used only by request stacking drivers
2886 * to stop dispatching requests to underlying devices when underlying
2887 * devices are busy. This behavior helps more I/O merging on the queue
2888 * of the request stacking driver and prevents I/O throughput regression
2889 * on burst I/O load.
2890 *
2891 * Return:
2892 * 0 - Not busy (The request stacking driver should dispatch request)
2893 * 1 - Busy (The request stacking driver should stop dispatching request)
2894 */
2895int blk_lld_busy(struct request_queue *q)
2896{
2897 if (q->lld_busy_fn)
2898 return q->lld_busy_fn(q);
2899
2900 return 0;
2901}
2902EXPORT_SYMBOL_GPL(blk_lld_busy);
2903
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002904/**
2905 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2906 * @rq: the clone request to be cleaned up
2907 *
2908 * Description:
2909 * Free all bios in @rq for a cloned request.
2910 */
2911void blk_rq_unprep_clone(struct request *rq)
2912{
2913 struct bio *bio;
2914
2915 while ((bio = rq->bio) != NULL) {
2916 rq->bio = bio->bi_next;
2917
2918 bio_put(bio);
2919 }
2920}
2921EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2922
2923/*
2924 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002925 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002926 */
2927static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2928{
2929 dst->cpu = src->cpu;
Keith Busch77a08682015-01-12 15:04:06 -05002930 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002931 dst->cmd_type = src->cmd_type;
2932 dst->__sector = blk_rq_pos(src);
2933 dst->__data_len = blk_rq_bytes(src);
2934 dst->nr_phys_segments = src->nr_phys_segments;
2935 dst->ioprio = src->ioprio;
2936 dst->extra_len = src->extra_len;
2937}
2938
2939/**
2940 * blk_rq_prep_clone - Helper function to setup clone request
2941 * @rq: the request to be setup
2942 * @rq_src: original request to be cloned
2943 * @bs: bio_set that bios for clone are allocated from
2944 * @gfp_mask: memory allocation mask for bio
2945 * @bio_ctr: setup function to be called for each clone bio.
2946 * Returns %0 for success, non %0 for failure.
2947 * @data: private data to be passed to @bio_ctr
2948 *
2949 * Description:
2950 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002951 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002952 * are not copied, and copying such parts is the caller's responsibility.
2953 * Also, pages which the original bios are pointing to are not copied
2954 * and the cloned bios just point same pages.
2955 * So cloned bios must be completed before original bios, which means
2956 * the caller must complete @rq before @rq_src.
2957 */
2958int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2959 struct bio_set *bs, gfp_t gfp_mask,
2960 int (*bio_ctr)(struct bio *, struct bio *, void *),
2961 void *data)
2962{
2963 struct bio *bio, *bio_src;
2964
2965 if (!bs)
2966 bs = fs_bio_set;
2967
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002968 __rq_for_each_bio(bio_src, rq_src) {
Junichi Nomura11dfce52014-10-03 17:27:11 -04002969 bio = bio_clone_fast(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002970 if (!bio)
2971 goto free_and_out;
2972
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002973 if (bio_ctr && bio_ctr(bio, bio_src, data))
2974 goto free_and_out;
2975
2976 if (rq->bio) {
2977 rq->biotail->bi_next = bio;
2978 rq->biotail = bio;
2979 } else
2980 rq->bio = rq->biotail = bio;
2981 }
2982
2983 __blk_rq_prep_clone(rq, rq_src);
2984
2985 return 0;
2986
2987free_and_out:
2988 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002989 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002990 blk_rq_unprep_clone(rq);
2991
2992 return -ENOMEM;
2993}
2994EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2995
Jens Axboe59c3d452014-04-08 09:15:35 -06002996int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997{
2998 return queue_work(kblockd_workqueue, work);
2999}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000EXPORT_SYMBOL(kblockd_schedule_work);
3001
Jens Axboe59c3d452014-04-08 09:15:35 -06003002int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3003 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003004{
3005 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3006}
3007EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3008
Jens Axboe8ab14592014-04-08 09:17:40 -06003009int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3010 unsigned long delay)
3011{
3012 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3013}
3014EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3015
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003016/**
3017 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3018 * @plug: The &struct blk_plug that needs to be initialized
3019 *
3020 * Description:
3021 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3022 * pending I/O should the task end up blocking between blk_start_plug() and
3023 * blk_finish_plug(). This is important from a performance perspective, but
3024 * also ensures that we don't deadlock. For instance, if the task is blocking
3025 * for a memory allocation, memory reclaim could end up wanting to free a
3026 * page belonging to that request that is currently residing in our private
3027 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3028 * this kind of deadlock.
3029 */
Jens Axboe73c10102011-03-08 13:19:51 +01003030void blk_start_plug(struct blk_plug *plug)
3031{
3032 struct task_struct *tsk = current;
3033
Jens Axboe73c10102011-03-08 13:19:51 +01003034 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003035 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003036 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003037
3038 /*
3039 * If this is a nested plug, don't actually assign it. It will be
3040 * flushed on its own.
3041 */
3042 if (!tsk->plug) {
3043 /*
3044 * Store ordering should not be needed here, since a potential
3045 * preempt will imply a full memory barrier
3046 */
3047 tsk->plug = plug;
3048 }
3049}
3050EXPORT_SYMBOL(blk_start_plug);
3051
3052static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3053{
3054 struct request *rqa = container_of(a, struct request, queuelist);
3055 struct request *rqb = container_of(b, struct request, queuelist);
3056
Jianpeng Ma975927b2012-10-25 21:58:17 +02003057 return !(rqa->q < rqb->q ||
3058 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003059}
3060
Jens Axboe49cac012011-04-16 13:51:05 +02003061/*
3062 * If 'from_schedule' is true, then postpone the dispatch of requests
3063 * until a safe kblockd context. We due this to avoid accidental big
3064 * additional stack usage in driver dispatch, in places where the originally
3065 * plugger did not intend it.
3066 */
Jens Axboef6603782011-04-15 15:49:07 +02003067static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003068 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003069 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003070{
Jens Axboe49cac012011-04-16 13:51:05 +02003071 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003072
Bart Van Assche70460572012-11-28 13:45:56 +01003073 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003074 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003075 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003076 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003077 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003078}
3079
NeilBrown74018dc2012-07-31 09:08:15 +02003080static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003081{
3082 LIST_HEAD(callbacks);
3083
Shaohua Li2a7d5552012-07-31 09:08:15 +02003084 while (!list_empty(&plug->cb_list)) {
3085 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003086
Shaohua Li2a7d5552012-07-31 09:08:15 +02003087 while (!list_empty(&callbacks)) {
3088 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003089 struct blk_plug_cb,
3090 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003091 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003092 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003093 }
NeilBrown048c9372011-04-18 09:52:22 +02003094 }
3095}
3096
NeilBrown9cbb1752012-07-31 09:08:14 +02003097struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3098 int size)
3099{
3100 struct blk_plug *plug = current->plug;
3101 struct blk_plug_cb *cb;
3102
3103 if (!plug)
3104 return NULL;
3105
3106 list_for_each_entry(cb, &plug->cb_list, list)
3107 if (cb->callback == unplug && cb->data == data)
3108 return cb;
3109
3110 /* Not currently on the callback list */
3111 BUG_ON(size < sizeof(*cb));
3112 cb = kzalloc(size, GFP_ATOMIC);
3113 if (cb) {
3114 cb->data = data;
3115 cb->callback = unplug;
3116 list_add(&cb->list, &plug->cb_list);
3117 }
3118 return cb;
3119}
3120EXPORT_SYMBOL(blk_check_plugged);
3121
Jens Axboe49cac012011-04-16 13:51:05 +02003122void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003123{
3124 struct request_queue *q;
3125 unsigned long flags;
3126 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003127 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003128 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003129
NeilBrown74018dc2012-07-31 09:08:15 +02003130 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003131
3132 if (!list_empty(&plug->mq_list))
3133 blk_mq_flush_plug_list(plug, from_schedule);
3134
Jens Axboe73c10102011-03-08 13:19:51 +01003135 if (list_empty(&plug->list))
3136 return;
3137
NeilBrown109b8122011-04-11 14:13:10 +02003138 list_splice_init(&plug->list, &list);
3139
Jianpeng Ma422765c2013-01-11 14:46:09 +01003140 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003141
3142 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003143 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003144
3145 /*
3146 * Save and disable interrupts here, to avoid doing it for every
3147 * queue lock we have to take.
3148 */
Jens Axboe73c10102011-03-08 13:19:51 +01003149 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003150 while (!list_empty(&list)) {
3151 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003152 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003153 BUG_ON(!rq->q);
3154 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003155 /*
3156 * This drops the queue lock
3157 */
3158 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003159 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003160 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003161 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003162 spin_lock(q->queue_lock);
3163 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003164
3165 /*
3166 * Short-circuit if @q is dead
3167 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003168 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003169 __blk_end_request_all(rq, -ENODEV);
3170 continue;
3171 }
3172
Jens Axboe73c10102011-03-08 13:19:51 +01003173 /*
3174 * rq is already accounted, so use raw insert
3175 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003176 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3177 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3178 else
3179 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003180
3181 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003182 }
3183
Jens Axboe99e22592011-04-18 09:59:55 +02003184 /*
3185 * This drops the queue lock
3186 */
3187 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003188 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003189
Jens Axboe73c10102011-03-08 13:19:51 +01003190 local_irq_restore(flags);
3191}
Jens Axboe73c10102011-03-08 13:19:51 +01003192
3193void blk_finish_plug(struct blk_plug *plug)
3194{
Jens Axboef6603782011-04-15 15:49:07 +02003195 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003196
3197 if (plug == current->plug)
3198 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003199}
3200EXPORT_SYMBOL(blk_finish_plug);
3201
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003202#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003203/**
3204 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3205 * @q: the queue of the device
3206 * @dev: the device the queue belongs to
3207 *
3208 * Description:
3209 * Initialize runtime-PM-related fields for @q and start auto suspend for
3210 * @dev. Drivers that want to take advantage of request-based runtime PM
3211 * should call this function after @dev has been initialized, and its
3212 * request queue @q has been allocated, and runtime PM for it can not happen
3213 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3214 * cases, driver should call this function before any I/O has taken place.
3215 *
3216 * This function takes care of setting up using auto suspend for the device,
3217 * the autosuspend delay is set to -1 to make runtime suspend impossible
3218 * until an updated value is either set by user or by driver. Drivers do
3219 * not need to touch other autosuspend settings.
3220 *
3221 * The block layer runtime PM is request based, so only works for drivers
3222 * that use request as their IO unit instead of those directly use bio's.
3223 */
3224void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3225{
3226 q->dev = dev;
3227 q->rpm_status = RPM_ACTIVE;
3228 pm_runtime_set_autosuspend_delay(q->dev, -1);
3229 pm_runtime_use_autosuspend(q->dev);
3230}
3231EXPORT_SYMBOL(blk_pm_runtime_init);
3232
3233/**
3234 * blk_pre_runtime_suspend - Pre runtime suspend check
3235 * @q: the queue of the device
3236 *
3237 * Description:
3238 * This function will check if runtime suspend is allowed for the device
3239 * by examining if there are any requests pending in the queue. If there
3240 * are requests pending, the device can not be runtime suspended; otherwise,
3241 * the queue's status will be updated to SUSPENDING and the driver can
3242 * proceed to suspend the device.
3243 *
3244 * For the not allowed case, we mark last busy for the device so that
3245 * runtime PM core will try to autosuspend it some time later.
3246 *
3247 * This function should be called near the start of the device's
3248 * runtime_suspend callback.
3249 *
3250 * Return:
3251 * 0 - OK to runtime suspend the device
3252 * -EBUSY - Device should not be runtime suspended
3253 */
3254int blk_pre_runtime_suspend(struct request_queue *q)
3255{
3256 int ret = 0;
3257
3258 spin_lock_irq(q->queue_lock);
3259 if (q->nr_pending) {
3260 ret = -EBUSY;
3261 pm_runtime_mark_last_busy(q->dev);
3262 } else {
3263 q->rpm_status = RPM_SUSPENDING;
3264 }
3265 spin_unlock_irq(q->queue_lock);
3266 return ret;
3267}
3268EXPORT_SYMBOL(blk_pre_runtime_suspend);
3269
3270/**
3271 * blk_post_runtime_suspend - Post runtime suspend processing
3272 * @q: the queue of the device
3273 * @err: return value of the device's runtime_suspend function
3274 *
3275 * Description:
3276 * Update the queue's runtime status according to the return value of the
3277 * device's runtime suspend function and mark last busy for the device so
3278 * that PM core will try to auto suspend the device at a later time.
3279 *
3280 * This function should be called near the end of the device's
3281 * runtime_suspend callback.
3282 */
3283void blk_post_runtime_suspend(struct request_queue *q, int err)
3284{
3285 spin_lock_irq(q->queue_lock);
3286 if (!err) {
3287 q->rpm_status = RPM_SUSPENDED;
3288 } else {
3289 q->rpm_status = RPM_ACTIVE;
3290 pm_runtime_mark_last_busy(q->dev);
3291 }
3292 spin_unlock_irq(q->queue_lock);
3293}
3294EXPORT_SYMBOL(blk_post_runtime_suspend);
3295
3296/**
3297 * blk_pre_runtime_resume - Pre runtime resume processing
3298 * @q: the queue of the device
3299 *
3300 * Description:
3301 * Update the queue's runtime status to RESUMING in preparation for the
3302 * runtime resume of the device.
3303 *
3304 * This function should be called near the start of the device's
3305 * runtime_resume callback.
3306 */
3307void blk_pre_runtime_resume(struct request_queue *q)
3308{
3309 spin_lock_irq(q->queue_lock);
3310 q->rpm_status = RPM_RESUMING;
3311 spin_unlock_irq(q->queue_lock);
3312}
3313EXPORT_SYMBOL(blk_pre_runtime_resume);
3314
3315/**
3316 * blk_post_runtime_resume - Post runtime resume processing
3317 * @q: the queue of the device
3318 * @err: return value of the device's runtime_resume function
3319 *
3320 * Description:
3321 * Update the queue's runtime status according to the return value of the
3322 * device's runtime_resume function. If it is successfully resumed, process
3323 * the requests that are queued into the device's queue when it is resuming
3324 * and then mark last busy and initiate autosuspend for it.
3325 *
3326 * This function should be called near the end of the device's
3327 * runtime_resume callback.
3328 */
3329void blk_post_runtime_resume(struct request_queue *q, int err)
3330{
3331 spin_lock_irq(q->queue_lock);
3332 if (!err) {
3333 q->rpm_status = RPM_ACTIVE;
3334 __blk_run_queue(q);
3335 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003336 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003337 } else {
3338 q->rpm_status = RPM_SUSPENDED;
3339 }
3340 spin_unlock_irq(q->queue_lock);
3341}
3342EXPORT_SYMBOL(blk_post_runtime_resume);
3343#endif
3344
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345int __init blk_dev_init(void)
3346{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003347 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3348 sizeof(((struct request *)0)->cmd_flags));
3349
Tejun Heo89b90be2011-01-03 15:01:47 +01003350 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3351 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003352 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 if (!kblockd_workqueue)
3354 panic("Failed to create kblockd\n");
3355
3356 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003357 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358
Jens Axboe8324aa92008-01-29 14:51:59 +01003359 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003360 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 return 0;
3363}