blob: 0421b53e6431fe9c2522f0b280058594bc9608cb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
Jens Axboe6728cb02008-01-31 13:03:55 +01006 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
Jens Axboe320ae512013-10-24 09:20:05 +010019#include <linux/blk-mq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/highmem.h>
21#include <linux/mm.h>
22#include <linux/kernel_stat.h>
23#include <linux/string.h>
24#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080029#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080030#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010031#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020032#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070033#include <linux/ratelimit.h>
Lin Ming6c954662013-03-23 11:42:26 +080034#include <linux/pm_runtime.h>
Li Zefan55782132009-06-09 13:43:05 +080035
36#define CREATE_TRACE_POINTS
37#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Jens Axboe8324aa92008-01-29 14:51:59 +010039#include "blk.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080040#include "blk-cgroup.h"
Ming Lei43a5e4e2013-12-26 21:31:35 +080041#include "blk-mq.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010042
Mike Snitzerd07335e2010-11-16 12:52:38 +010043EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020044EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070045EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060046EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010047EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010048
Tejun Heoa73f7302011-12-14 00:33:37 +010049DEFINE_IDA(blk_queue_ida);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * For the allocated request tables
53 */
Jens Axboe320ae512013-10-24 09:20:05 +010054struct kmem_cache *request_cachep = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
57 * For queue allocation
58 */
Jens Axboe6728cb02008-01-31 13:03:55 +010059struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * Controlling structure to kblockd
63 */
Jens Axboeff856ba2006-01-09 16:02:34 +010064static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Jens Axboe8324aa92008-01-29 14:51:59 +010066void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067{
68 int nr;
69
70 nr = q->nr_requests - (q->nr_requests / 8) + 1;
71 if (nr > q->nr_requests)
72 nr = q->nr_requests;
73 q->nr_congestion_on = nr;
74
75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
76 if (nr < 1)
77 nr = 1;
78 q->nr_congestion_off = nr;
79}
80
Linus Torvalds1da177e2005-04-16 15:20:36 -070081/**
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
83 * @bdev: device
84 *
85 * Locates the passed device's request queue and returns the address of its
Tejun Heoff9ea322014-09-08 08:03:56 +090086 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
89struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
90{
Jens Axboe165125e2007-07-24 09:28:11 +020091 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Tejun Heoff9ea322014-09-08 08:03:56 +090093 return &q->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094}
Linus Torvalds1da177e2005-04-16 15:20:36 -070095EXPORT_SYMBOL(blk_get_backing_dev_info);
96
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +020097void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +020099 memset(rq, 0, sizeof(*rq));
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700102 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200103 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100104 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900105 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200106 INIT_HLIST_NODE(&rq->hash);
107 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200108 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800109 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100110 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900111 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700112 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100113 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200115EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
NeilBrown5bb23a62007-09-27 12:46:13 +0200117static void req_bio_endio(struct request *rq, struct bio *bio,
118 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100119{
Tejun Heo143a87f2011-01-25 12:43:52 +0100120 if (error)
121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
123 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100124
Tejun Heo143a87f2011-01-25 12:43:52 +0100125 if (unlikely(rq->cmd_flags & REQ_QUIET))
126 set_bit(BIO_QUIET, &bio->bi_flags);
127
Kent Overstreetf79ea412012-09-20 16:38:30 -0700128 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100129
130 /* don't actually finish bio if it's part of flush sequence */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700131 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Tejun Heo143a87f2011-01-25 12:43:52 +0100132 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100133}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135void blk_dump_rq_flags(struct request *rq, char *msg)
136{
137 int bit;
138
Jens Axboe59533162013-05-23 12:25:08 +0200139 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200140 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200141 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Tejun Heo83096eb2009-05-07 22:24:39 +0900143 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
144 (unsigned long long)blk_rq_pos(rq),
145 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600146 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
147 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200149 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100150 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200151 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 printk("%02x ", rq->cmd[bit]);
153 printk("\n");
154 }
155}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156EXPORT_SYMBOL(blk_dump_rq_flags);
157
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500158static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200159{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500160 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200161
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500162 q = container_of(work, struct request_queue, delay_work.work);
163 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200164 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500165 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500169 * blk_delay_queue - restart queueing after defined interval
170 * @q: The &struct request_queue in question
171 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 *
173 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500174 * Sometimes queueing needs to be postponed for a little while, to allow
175 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100176 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500177 */
178void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
Bart Van Assche70460572012-11-28 13:45:56 +0100180 if (likely(!blk_queue_dead(q)))
181 queue_delayed_work(kblockd_workqueue, &q->delay_work,
182 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500184EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/**
187 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200188 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 *
190 * Description:
191 * blk_start_queue() will clear the stop flag on the queue, and call
192 * the request_fn for the queue if it was in a stopped state when
193 * entered. Also see blk_stop_queue(). Queue lock must be held.
194 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200195void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200197 WARN_ON(!irqs_disabled());
198
Nick Piggin75ad23b2008-04-29 14:48:33 +0200199 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200200 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202EXPORT_SYMBOL(blk_start_queue);
203
204/**
205 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200206 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 *
208 * Description:
209 * The Linux block layer assumes that a block driver will consume all
210 * entries on the request queue when the request_fn strategy is called.
211 * Often this will not happen, because of hardware limitations (queue
212 * depth settings). If a device driver gets a 'queue full' response,
213 * or if it simply chooses not to queue more I/O at one point, it can
214 * call this function to prevent the request_fn from being called until
215 * the driver has signalled it's ready to go again. This happens by calling
216 * blk_start_queue() to restart queue operations. Queue lock must be held.
217 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200218void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
Tejun Heo136b5722012-08-21 13:18:24 -0700220 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200221 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223EXPORT_SYMBOL(blk_stop_queue);
224
225/**
226 * blk_sync_queue - cancel any pending callbacks on a queue
227 * @q: the queue
228 *
229 * Description:
230 * The block layer may perform asynchronous callback activity
231 * on a queue, such as calling the unplug function after a timeout.
232 * A block device may call blk_sync_queue to ensure that any
233 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200234 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * that its ->make_request_fn will not re-add plugging prior to calling
236 * this function.
237 *
Vivek Goyalda527772011-03-02 19:05:33 -0500238 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900239 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800240 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500241 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
243void blk_sync_queue(struct request_queue *q)
244{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100245 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800246
247 if (q->mq_ops) {
248 struct blk_mq_hw_ctx *hctx;
249 int i;
250
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600251 queue_for_each_hw_ctx(q, hctx, i) {
252 cancel_delayed_work_sync(&hctx->run_work);
253 cancel_delayed_work_sync(&hctx->delay_work);
254 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800255 } else {
256 cancel_delayed_work_sync(&q->delay_work);
257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259EXPORT_SYMBOL(blk_sync_queue);
260
261/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100262 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
263 * @q: The queue to run
264 *
265 * Description:
266 * Invoke request handling on a queue if there are any pending requests.
267 * May be used to restart request handling after a request has completed.
268 * This variant runs the queue whether or not the queue has been
269 * stopped. Must be called with the queue lock held and interrupts
270 * disabled. See also @blk_run_queue.
271 */
272inline void __blk_run_queue_uncond(struct request_queue *q)
273{
274 if (unlikely(blk_queue_dead(q)))
275 return;
276
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100277 /*
278 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
279 * the queue lock internally. As a result multiple threads may be
280 * running such a request function concurrently. Keep track of the
281 * number of active request_fn invocations such that blk_drain_queue()
282 * can wait until all these request_fn calls have finished.
283 */
284 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100285 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100286 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100287}
288
289/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200290 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200292 *
293 * Description:
294 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200295 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200297void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Tejun Heoa538cd02009-04-23 11:05:17 +0900299 if (unlikely(blk_queue_stopped(q)))
300 return;
301
Bart Van Asschec246e802012-12-06 14:32:01 +0100302 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200303}
304EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200305
Nick Piggin75ad23b2008-04-29 14:48:33 +0200306/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200307 * blk_run_queue_async - run a single device queue in workqueue context
308 * @q: The queue to run
309 *
310 * Description:
311 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100312 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200313 */
314void blk_run_queue_async(struct request_queue *q)
315{
Bart Van Assche70460572012-11-28 13:45:56 +0100316 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700317 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200318}
Jens Axboec21e6be2011-04-19 13:32:46 +0200319EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200320
321/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200322 * blk_run_queue - run a single device queue
323 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200324 *
325 * Description:
326 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900327 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200328 */
329void blk_run_queue(struct request_queue *q)
330{
331 unsigned long flags;
332
333 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200334 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 spin_unlock_irqrestore(q->queue_lock, flags);
336}
337EXPORT_SYMBOL(blk_run_queue);
338
Jens Axboe165125e2007-07-24 09:28:11 +0200339void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500340{
341 kobject_put(&q->kobj);
342}
Jens Axboed86e0e82011-05-27 07:44:43 +0200343EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500344
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200345/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100346 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200347 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200348 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200349 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200350 * Drain requests from @q. If @drain_all is set, all requests are drained.
351 * If not, only ELVPRIV requests are drained. The caller is responsible
352 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100354static void __blk_drain_queue(struct request_queue *q, bool drain_all)
355 __releases(q->queue_lock)
356 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200357{
Asias He458f27a2012-06-15 08:45:25 +0200358 int i;
359
Bart Van Assche807592a2012-11-28 13:43:38 +0100360 lockdep_assert_held(q->queue_lock);
361
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200362 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100363 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200364
Tejun Heob855b042012-03-06 21:24:55 +0100365 /*
366 * The caller might be trying to drain @q before its
367 * elevator is initialized.
368 */
369 if (q->elevator)
370 elv_drain_elevator(q);
371
Tejun Heo5efd6112012-03-05 13:15:12 -0800372 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200373
Tejun Heo4eabc942011-12-15 20:03:04 +0100374 /*
375 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100376 * driver init after queue creation or is not yet fully
377 * active yet. Some drivers (e.g. fd and loop) get unhappy
378 * in such cases. Kick queue iff dispatch queue has
379 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100380 */
Tejun Heob855b042012-03-06 21:24:55 +0100381 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100382 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200383
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700384 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100385 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100386
387 /*
388 * Unfortunately, requests are queued at and tracked from
389 * multiple places and there's no single counter which can
390 * be drained. Check all the queues and counters.
391 */
392 if (drain_all) {
Ming Leie97c2932014-09-25 23:23:46 +0800393 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
Tejun Heo481a7d62011-12-14 00:33:37 +0100394 drain |= !list_empty(&q->queue_head);
395 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700396 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100397 drain |= q->in_flight[i];
Ming Lei7c94e1c2014-09-25 23:23:43 +0800398 if (fq)
399 drain |= !list_empty(&fq->flush_queue[i]);
Tejun Heo481a7d62011-12-14 00:33:37 +0100400 }
401 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200402
Tejun Heo481a7d62011-12-14 00:33:37 +0100403 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100405
406 spin_unlock_irq(q->queue_lock);
407
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200408 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100409
410 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200411 }
Asias He458f27a2012-06-15 08:45:25 +0200412
413 /*
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
417 */
418 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700419 struct request_list *rl;
420
Tejun Heoa0516612012-06-26 15:05:44 -0700421 blk_queue_for_each_rl(rl, q)
422 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
423 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200424 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200425}
426
Tejun Heoc9a929d2011-10-19 14:42:16 +0200427/**
Tejun Heod7325802012-03-05 13:14:58 -0800428 * blk_queue_bypass_start - enter queue bypass mode
429 * @q: queue of interest
430 *
431 * In bypass mode, only the dispatch FIFO queue of @q is used. This
432 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800433 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700434 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
435 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800436 */
437void blk_queue_bypass_start(struct request_queue *q)
438{
439 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600440 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800441 queue_flag_set(QUEUE_FLAG_BYPASS, q);
442 spin_unlock_irq(q->queue_lock);
443
Tejun Heo776687b2014-07-01 10:29:17 -0600444 /*
445 * Queues start drained. Skip actual draining till init is
446 * complete. This avoids lenghty delays during queue init which
447 * can happen many times during boot.
448 */
449 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100450 spin_lock_irq(q->queue_lock);
451 __blk_drain_queue(q, false);
452 spin_unlock_irq(q->queue_lock);
453
Tejun Heob82d4b12012-04-13 13:11:31 -0700454 /* ensure blk_queue_bypass() is %true inside RCU read lock */
455 synchronize_rcu();
456 }
Tejun Heod7325802012-03-05 13:14:58 -0800457}
458EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
459
460/**
461 * blk_queue_bypass_end - leave queue bypass mode
462 * @q: queue of interest
463 *
464 * Leave bypass mode and restore the normal queueing behavior.
465 */
466void blk_queue_bypass_end(struct request_queue *q)
467{
468 spin_lock_irq(q->queue_lock);
469 if (!--q->bypass_depth)
470 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
471 WARN_ON_ONCE(q->bypass_depth < 0);
472 spin_unlock_irq(q->queue_lock);
473}
474EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
475
476/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200477 * blk_cleanup_queue - shutdown a request queue
478 * @q: request queue to shutdown
479 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100480 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
481 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500482 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100483void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500484{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200485 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700486
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100487 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500488 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100489 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200490 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800491
Tejun Heo80fd9972012-04-13 14:50:53 -0700492 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100493 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700494 * that, unlike blk_queue_bypass_start(), we aren't performing
495 * synchronize_rcu() after entering bypass mode to avoid the delay
496 * as some drivers create and destroy a lot of queues while
497 * probing. This is still safe because blk_release_queue() will be
498 * called only after the queue refcnt drops to zero and nothing,
499 * RCU or not, would be traversing the queue by then.
500 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800501 q->bypass_depth++;
502 queue_flag_set(QUEUE_FLAG_BYPASS, q);
503
Tejun Heoc9a929d2011-10-19 14:42:16 +0200504 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
505 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100506 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200507 spin_unlock_irq(lock);
508 mutex_unlock(&q->sysfs_lock);
509
Bart Van Asschec246e802012-12-06 14:32:01 +0100510 /*
511 * Drain all requests queued before DYING marking. Set DEAD flag to
512 * prevent that q->request_fn() gets invoked after draining finished.
513 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800514 if (q->mq_ops) {
Tejun Heo780db202014-07-01 10:31:13 -0600515 blk_mq_freeze_queue(q);
Ming Lei43a5e4e2013-12-26 21:31:35 +0800516 spin_lock_irq(lock);
517 } else {
518 spin_lock_irq(lock);
519 __blk_drain_queue(q, true);
520 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100521 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100522 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200523
524 /* @q won't process any more request, flush async actions */
525 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
526 blk_sync_queue(q);
527
Asias He5e5cfac2012-05-24 23:28:52 +0800528 spin_lock_irq(lock);
529 if (q->queue_lock != &q->__queue_lock)
530 q->queue_lock = &q->__queue_lock;
531 spin_unlock_irq(lock);
532
Tejun Heoc9a929d2011-10-19 14:42:16 +0200533 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500534 blk_put_queue(q);
535}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536EXPORT_SYMBOL(blk_cleanup_queue);
537
Tejun Heo5b788ce2012-06-04 20:40:59 -0700538int blk_init_rl(struct request_list *rl, struct request_queue *q,
539 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400541 if (unlikely(rl->rq_pool))
542 return 0;
543
Tejun Heo5b788ce2012-06-04 20:40:59 -0700544 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200545 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
546 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200547 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
548 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Christoph Lameter19460892005-06-23 00:08:19 -0700550 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700551 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700552 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 if (!rl->rq_pool)
554 return -ENOMEM;
555
556 return 0;
557}
558
Tejun Heo5b788ce2012-06-04 20:40:59 -0700559void blk_exit_rl(struct request_list *rl)
560{
561 if (rl->rq_pool)
562 mempool_destroy(rl->rq_pool);
563}
564
Jens Axboe165125e2007-07-24 09:28:11 +0200565struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100567 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700568}
569EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jens Axboe165125e2007-07-24 09:28:11 +0200571struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700572{
Jens Axboe165125e2007-07-24 09:28:11 +0200573 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700574 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700575
Jens Axboe8324aa92008-01-29 14:51:59 +0100576 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700577 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (!q)
579 return NULL;
580
Dan Carpenter00380a42012-03-23 09:58:54 +0100581 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100582 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800583 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100584
Jens Axboe0989a022009-06-12 14:42:56 +0200585 q->backing_dev_info.ra_pages =
586 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
587 q->backing_dev_info.state = 0;
588 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200589 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100590 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200591
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700592 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100593 if (err)
594 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700595
Matthew Garrett31373d02010-04-06 14:25:14 +0200596 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
597 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700598 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100599 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700600 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100601 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800602#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800603 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800604#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500605 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500606
Jens Axboe8324aa92008-01-29 14:51:59 +0100607 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Al Viro483f4af2006-03-18 18:34:37 -0500609 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700610 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500611
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500612 /*
613 * By default initialize queue_lock to internal lock and driver can
614 * override it later if need be.
615 */
616 q->queue_lock = &q->__queue_lock;
617
Tejun Heob82d4b12012-04-13 13:11:31 -0700618 /*
619 * A queue starts its life with bypass turned on to avoid
620 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700621 * init. The initial bypass will be finished when the queue is
622 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700623 */
624 q->bypass_depth = 1;
625 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
626
Jens Axboe320ae512013-10-24 09:20:05 +0100627 init_waitqueue_head(&q->mq_freeze_wq);
628
Tejun Heo5efd6112012-03-05 13:15:12 -0800629 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400630 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100633
Mikulas Patockafff49962013-10-14 12:11:36 -0400634fail_bdi:
635 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100636fail_id:
637 ida_simple_remove(&blk_queue_ida, q->id);
638fail_q:
639 kmem_cache_free(blk_requestq_cachep, q);
640 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641}
Christoph Lameter19460892005-06-23 00:08:19 -0700642EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644/**
645 * blk_init_queue - prepare a request queue for use with a block device
646 * @rfn: The function to be called to process requests that have been
647 * placed on the queue.
648 * @lock: Request queue spin lock
649 *
650 * Description:
651 * If a block device wishes to use the standard request handling procedures,
652 * which sorts requests and coalesces adjacent requests, then it must
653 * call blk_init_queue(). The function @rfn will be called when there
654 * are requests on the queue that need to be processed. If the device
655 * supports plugging, then @rfn may not be called immediately when requests
656 * are available on the queue, but may be called at some time later instead.
657 * Plugged queues are generally unplugged when a buffer belonging to one
658 * of the requests on the queue is needed, or due to memory pressure.
659 *
660 * @rfn is not required, or even expected, to remove all requests off the
661 * queue, but only as many as it can handle at a time. If it does leave
662 * requests on the queue, it is responsible for arranging that the requests
663 * get dealt with eventually.
664 *
665 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200666 * request queue; this lock will be taken also from interrupt context, so irq
667 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200669 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 * it didn't succeed.
671 *
672 * Note:
673 * blk_init_queue() must be paired with a blk_cleanup_queue() call
674 * when the block device is deactivated (such as at module unload).
675 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700676
Jens Axboe165125e2007-07-24 09:28:11 +0200677struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100679 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700680}
681EXPORT_SYMBOL(blk_init_queue);
682
Jens Axboe165125e2007-07-24 09:28:11 +0200683struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700684blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
685{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600686 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600688 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
689 if (!uninit_q)
690 return NULL;
691
Mike Snitzer51514122011-11-23 10:59:13 +0100692 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600693 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700694 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700695
Mike Snitzer7982e902014-03-08 17:20:01 -0700696 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200697}
698EXPORT_SYMBOL(blk_init_queue_node);
699
700struct request_queue *
701blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
702 spinlock_t *lock)
703{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 if (!q)
705 return NULL;
706
Ming Leif70ced02014-09-25 23:23:47 +0800707 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800708 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700709 return NULL;
710
Tejun Heoa0516612012-06-26 15:05:44 -0700711 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600712 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900716 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700717 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500718
719 /* Override internal queue lock with supplied lock pointer */
720 if (lock)
721 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Jens Axboef3b144a2009-03-06 08:48:33 +0100723 /*
724 * This also sets hw/phys segments, boundary and size
725 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200726 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Alan Stern44ec9542007-02-20 11:01:57 -0500728 q->sg_reserved_size = INT_MAX;
729
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600730 /* Protect q->elevator from elevator_change */
731 mutex_lock(&q->sysfs_lock);
732
Tejun Heob82d4b12012-04-13 13:11:31 -0700733 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600734 if (elevator_init(q, NULL)) {
735 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600736 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600737 }
738
739 mutex_unlock(&q->sysfs_lock);
740
Tejun Heob82d4b12012-04-13 13:11:31 -0700741 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600742
743fail:
Ming Leiba483382014-09-25 23:23:44 +0800744 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600745 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
Mike Snitzer51514122011-11-23 10:59:13 +0100747EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Tejun Heo09ac46c2011-12-14 00:33:38 +0100749bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100751 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100752 __blk_get_queue(q);
753 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
Tejun Heo09ac46c2011-12-14 00:33:38 +0100756 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
Jens Axboed86e0e82011-05-27 07:44:43 +0200758EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Tejun Heo5b788ce2012-06-04 20:40:59 -0700760static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100762 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700763 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100764 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100765 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100766 }
767
Tejun Heo5b788ce2012-06-04 20:40:59 -0700768 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771/*
772 * ioc_batching returns true if the ioc is a valid batching request and
773 * should be given priority access to a request.
774 */
Jens Axboe165125e2007-07-24 09:28:11 +0200775static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
777 if (!ioc)
778 return 0;
779
780 /*
781 * Make sure the process is able to allocate at least 1 request
782 * even if the batch times out, otherwise we could theoretically
783 * lose wakeups.
784 */
785 return ioc->nr_batch_requests == q->nr_batching ||
786 (ioc->nr_batch_requests > 0
787 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
788}
789
790/*
791 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
792 * will cause the process to be a "batcher" on all queues in the system. This
793 * is the behaviour we want though - once it gets a wakeup it should be given
794 * a nice run.
795 */
Jens Axboe165125e2007-07-24 09:28:11 +0200796static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
798 if (!ioc || ioc_batching(q, ioc))
799 return;
800
801 ioc->nr_batch_requests = q->nr_batching;
802 ioc->last_waited = jiffies;
803}
804
Tejun Heo5b788ce2012-06-04 20:40:59 -0700805static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700807 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Tejun Heoa0516612012-06-26 15:05:44 -0700809 /*
810 * bdi isn't aware of blkcg yet. As all async IOs end up root
811 * blkcg anyway, just use root blkcg state.
812 */
813 if (rl == &q->root_rl &&
814 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200815 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Jens Axboe1faa16d2009-04-06 14:48:01 +0200817 if (rl->count[sync] + 1 <= q->nr_requests) {
818 if (waitqueue_active(&rl->wait[sync]))
819 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Tejun Heo5b788ce2012-06-04 20:40:59 -0700821 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823}
824
825/*
826 * A request has just been released. Account for it, update the full and
827 * congestion status, wake up any waiters. Called under q->queue_lock.
828 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700829static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700831 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200832 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700834 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200835 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200836 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700837 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Tejun Heo5b788ce2012-06-04 20:40:59 -0700839 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Jens Axboe1faa16d2009-04-06 14:48:01 +0200841 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700842 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600845int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
846{
847 struct request_list *rl;
848
849 spin_lock_irq(q->queue_lock);
850 q->nr_requests = nr;
851 blk_queue_congestion_threshold(q);
852
853 /* congestion isn't cgroup aware and follows root blkcg for now */
854 rl = &q->root_rl;
855
856 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
857 blk_set_queue_congested(q, BLK_RW_SYNC);
858 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
859 blk_clear_queue_congested(q, BLK_RW_SYNC);
860
861 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
862 blk_set_queue_congested(q, BLK_RW_ASYNC);
863 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
864 blk_clear_queue_congested(q, BLK_RW_ASYNC);
865
866 blk_queue_for_each_rl(rl, q) {
867 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
868 blk_set_rl_full(rl, BLK_RW_SYNC);
869 } else {
870 blk_clear_rl_full(rl, BLK_RW_SYNC);
871 wake_up(&rl->wait[BLK_RW_SYNC]);
872 }
873
874 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
875 blk_set_rl_full(rl, BLK_RW_ASYNC);
876 } else {
877 blk_clear_rl_full(rl, BLK_RW_ASYNC);
878 wake_up(&rl->wait[BLK_RW_ASYNC]);
879 }
880 }
881
882 spin_unlock_irq(q->queue_lock);
883 return 0;
884}
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100887 * Determine if elevator data should be initialized when allocating the
888 * request associated with @bio.
889 */
890static bool blk_rq_should_init_elevator(struct bio *bio)
891{
892 if (!bio)
893 return true;
894
895 /*
896 * Flush requests do not use the elevator so skip initialization.
897 * This allows a request to share the flush and elevator data.
898 */
899 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
900 return false;
901
902 return true;
903}
904
Tejun Heoda8303c2011-10-19 14:33:05 +0200905/**
Tejun Heo852c7882012-03-05 13:15:27 -0800906 * rq_ioc - determine io_context for request allocation
907 * @bio: request being allocated is for this bio (can be %NULL)
908 *
909 * Determine io_context to use for request allocation for @bio. May return
910 * %NULL if %current->io_context doesn't exist.
911 */
912static struct io_context *rq_ioc(struct bio *bio)
913{
914#ifdef CONFIG_BLK_CGROUP
915 if (bio && bio->bi_ioc)
916 return bio->bi_ioc;
917#endif
918 return current->io_context;
919}
920
921/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700922 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700923 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200924 * @rw_flags: RW and SYNC flags
925 * @bio: bio to allocate request for (can be %NULL)
926 * @gfp_mask: allocation mask
927 *
928 * Get a free request from @q. This function may fail under memory
929 * pressure or if @q is dead.
930 *
Masanari Iidada3dae52014-09-09 01:27:23 +0900931 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -0600932 * Returns ERR_PTR on failure, with @q->queue_lock held.
933 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700935static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700936 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700938 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800939 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700940 struct elevator_type *et = q->elevator->type;
941 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100942 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200943 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200944 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100946 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -0600947 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +0200948
Jens Axboe7749a8d2006-12-13 13:02:26 +0100949 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100950 if (may_queue == ELV_MQUEUE_NO)
951 goto rq_starved;
952
Jens Axboe1faa16d2009-04-06 14:48:01 +0200953 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
954 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100955 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100956 * The queue will fill after this allocation, so set
957 * it as full, and mark this process as "batching".
958 * This process will be allowed to complete a batch of
959 * requests, others will be blocked.
960 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700961 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100962 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700963 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100964 } else {
965 if (may_queue != ELV_MQUEUE_MUST
966 && !ioc_batching(q, ioc)) {
967 /*
968 * The queue is full and the allocating
969 * process is not a "batcher", and not
970 * exempted by the IO scheduler
971 */
Joe Lawrencea492f072014-08-28 08:15:21 -0600972 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100973 }
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 }
Tejun Heoa0516612012-06-26 15:05:44 -0700976 /*
977 * bdi isn't aware of blkcg yet. As all async IOs end up
978 * root blkcg anyway, just use root blkcg state.
979 */
980 if (rl == &q->root_rl)
981 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 }
983
Jens Axboe082cf692005-06-28 16:35:11 +0200984 /*
985 * Only allow batching queuers to allocate up to 50% over the defined
986 * limit of requests, otherwise we could have thousands of requests
987 * allocated with any setting of ->nr_requests
988 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200989 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -0600990 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100991
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700992 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200993 rl->count[is_sync]++;
994 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200995
Tejun Heof1f8cc92011-12-14 00:33:42 +0100996 /*
997 * Decide whether the new request will be managed by elevator. If
998 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
999 * prevent the current elevator from being destroyed until the new
1000 * request is freed. This guarantees icq's won't be destroyed and
1001 * makes creating new ones safe.
1002 *
1003 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1004 * it will be created after releasing queue_lock.
1005 */
Tejun Heod7325802012-03-05 13:14:58 -08001006 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001007 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001008 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001009 if (et->icq_cache && ioc)
1010 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001011 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001012
Jens Axboef253b862010-10-24 22:06:02 +02001013 if (blk_queue_io_stat(q))
1014 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 spin_unlock_irq(q->queue_lock);
1016
Tejun Heo29e2b092012-04-19 16:29:21 -07001017 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001018 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001019 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001020 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Tejun Heo29e2b092012-04-19 16:29:21 -07001022 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001023 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001024 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1025
Tejun Heoaaf7c682012-04-19 16:29:22 -07001026 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001027 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001028 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001029 if (ioc)
1030 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001031 if (!icq)
1032 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001033 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001034
1035 rq->elv.icq = icq;
1036 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1037 goto fail_elvpriv;
1038
1039 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001040 if (icq)
1041 get_io_context(icq->ioc);
1042 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001043out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001044 /*
1045 * ioc may be NULL here, and ioc_batching will be false. That's
1046 * OK, if the queue is under the request limit then requests need
1047 * not count toward the nr_batch_requests limit. There will always
1048 * be some limit enforced by BLK_BATCH_TIME.
1049 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (ioc_batching(q, ioc))
1051 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001052
Jens Axboe1faa16d2009-04-06 14:48:01 +02001053 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001055
Tejun Heoaaf7c682012-04-19 16:29:22 -07001056fail_elvpriv:
1057 /*
1058 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1059 * and may fail indefinitely under memory pressure and thus
1060 * shouldn't stall IO. Treat this request as !elvpriv. This will
1061 * disturb iosched and blkcg but weird is bettern than dead.
1062 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001063 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1064 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001065
1066 rq->cmd_flags &= ~REQ_ELVPRIV;
1067 rq->elv.icq = NULL;
1068
1069 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001070 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001071 spin_unlock_irq(q->queue_lock);
1072 goto out;
1073
Tejun Heob6792812012-03-05 13:15:23 -08001074fail_alloc:
1075 /*
1076 * Allocation failed presumably due to memory. Undo anything we
1077 * might have messed up.
1078 *
1079 * Allocating task should really be put onto the front of the wait
1080 * queue, but this is pretty rare.
1081 */
1082 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001083 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001084
1085 /*
1086 * in the very unlikely event that allocation failed and no
1087 * requests for this direction was pending, mark us starved so that
1088 * freeing of a request in the other direction will notice
1089 * us. another possible fix would be to split the rq mempool into
1090 * READ and WRITE
1091 */
1092rq_starved:
1093 if (unlikely(rl->count[is_sync] == 0))
1094 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001095 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
Tejun Heoda8303c2011-10-19 14:33:05 +02001098/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001099 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001100 * @q: request_queue to allocate request from
1101 * @rw_flags: RW and SYNC flags
1102 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001103 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001104 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001105 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1106 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001107 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001108 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001109 * Returns ERR_PTR on failure, with @q->queue_lock held.
1110 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001112static struct request *get_request(struct request_queue *q, int rw_flags,
1113 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001115 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001116 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001117 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001119
1120 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001121retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001122 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001123 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001124 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001126 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001127 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001128 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Tejun Heoa06e05e2012-06-04 20:40:55 -07001131 /* wait on @rl and retry */
1132 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1133 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001134
Tejun Heoa06e05e2012-06-04 20:40:55 -07001135 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Tejun Heoa06e05e2012-06-04 20:40:55 -07001137 spin_unlock_irq(q->queue_lock);
1138 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Tejun Heoa06e05e2012-06-04 20:40:55 -07001140 /*
1141 * After sleeping, we become a "batching" process and will be able
1142 * to allocate at least one request, and up to a big batch of them
1143 * for a small period time. See ioc_batching, ioc_set_batching
1144 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001145 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Tejun Heoa06e05e2012-06-04 20:40:55 -07001147 spin_lock_irq(q->queue_lock);
1148 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001149
Tejun Heoa06e05e2012-06-04 20:40:55 -07001150 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151}
1152
Jens Axboe320ae512013-10-24 09:20:05 +01001153static struct request *blk_old_get_request(struct request_queue *q, int rw,
1154 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
1156 struct request *rq;
1157
1158 BUG_ON(rw != READ && rw != WRITE);
1159
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001160 /* create ioc upfront */
1161 create_io_context(gfp_mask, q->node);
1162
Nick Piggind6344532005-06-28 20:45:14 -07001163 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001164 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001165 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001166 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001167 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 return rq;
1170}
Jens Axboe320ae512013-10-24 09:20:05 +01001171
1172struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1173{
1174 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001175 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001176 else
1177 return blk_old_get_request(q, rw, gfp_mask);
1178}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179EXPORT_SYMBOL(blk_get_request);
1180
1181/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001182 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001183 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001184 * @bio: The bio describing the memory mappings that will be submitted for IO.
1185 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001186 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001187 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001188 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1189 * type commands. Where the struct request needs to be farther initialized by
1190 * the caller. It is passed a &struct bio, which describes the memory info of
1191 * the I/O transfer.
1192 *
1193 * The caller of blk_make_request must make sure that bi_io_vec
1194 * are set to describe the memory buffers. That bio_data_dir() will return
1195 * the needed direction of the request. (And all bio's in the passed bio-chain
1196 * are properly set accordingly)
1197 *
1198 * If called under none-sleepable conditions, mapped bio buffers must not
1199 * need bouncing, by calling the appropriate masked or flagged allocator,
1200 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1201 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001202 *
1203 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1204 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1205 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1206 * completion of a bio that hasn't been submitted yet, thus resulting in a
1207 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1208 * of bio_alloc(), as that avoids the mempool deadlock.
1209 * If possible a big IO should be split into smaller parts when allocation
1210 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001211 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001212struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1213 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001214{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001215 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1216
Joe Lawrencea492f072014-08-28 08:15:21 -06001217 if (IS_ERR(rq))
1218 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001219
Jens Axboef27b0872014-06-06 07:57:37 -06001220 blk_rq_set_block_pc(rq);
1221
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001222 for_each_bio(bio) {
1223 struct bio *bounce_bio = bio;
1224 int ret;
1225
1226 blk_queue_bounce(q, &bounce_bio);
1227 ret = blk_rq_append_bio(q, rq, bounce_bio);
1228 if (unlikely(ret)) {
1229 blk_put_request(rq);
1230 return ERR_PTR(ret);
1231 }
1232 }
1233
1234 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001235}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001236EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001237
1238/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001239 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001240 * @rq: request to be initialized
1241 *
1242 */
1243void blk_rq_set_block_pc(struct request *rq)
1244{
1245 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1246 rq->__data_len = 0;
1247 rq->__sector = (sector_t) -1;
1248 rq->bio = rq->biotail = NULL;
1249 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001250}
1251EXPORT_SYMBOL(blk_rq_set_block_pc);
1252
1253/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 * blk_requeue_request - put a request back on queue
1255 * @q: request queue where request should be inserted
1256 * @rq: request to be inserted
1257 *
1258 * Description:
1259 * Drivers often keep queueing requests until the hardware cannot accept
1260 * more, when that condition happens we need to put the request back
1261 * on the queue. Must be called with queue lock held.
1262 */
Jens Axboe165125e2007-07-24 09:28:11 +02001263void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001265 blk_delete_timer(rq);
1266 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001267 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (blk_rq_tagged(rq))
1270 blk_queue_end_tag(q, rq);
1271
James Bottomleyba396a62009-05-27 14:17:08 +02001272 BUG_ON(blk_queued_rq(rq));
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 elv_requeue_request(q, rq);
1275}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276EXPORT_SYMBOL(blk_requeue_request);
1277
Jens Axboe73c10102011-03-08 13:19:51 +01001278static void add_acct_request(struct request_queue *q, struct request *rq,
1279 int where)
1280{
Jens Axboe320ae512013-10-24 09:20:05 +01001281 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001282 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001283}
1284
Tejun Heo074a7ac2008-08-25 19:56:14 +09001285static void part_round_stats_single(int cpu, struct hd_struct *part,
1286 unsigned long now)
1287{
Jens Axboe7276d022014-05-09 15:48:23 -06001288 int inflight;
1289
Tejun Heo074a7ac2008-08-25 19:56:14 +09001290 if (now == part->stamp)
1291 return;
1292
Jens Axboe7276d022014-05-09 15:48:23 -06001293 inflight = part_in_flight(part);
1294 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001295 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001296 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001297 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1298 }
1299 part->stamp = now;
1300}
1301
1302/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001303 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1304 * @cpu: cpu number for stats access
1305 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 *
1307 * The average IO queue length and utilisation statistics are maintained
1308 * by observing the current state of the queue length and the amount of
1309 * time it has been in this state for.
1310 *
1311 * Normally, that accounting is done on IO completion, but that can result
1312 * in more than a second's worth of IO being accounted for within any one
1313 * second, leading to >100% utilisation. To deal with that, we call this
1314 * function to do a round-off before returning the results when reading
1315 * /proc/diskstats. This accounts immediately for all queue usage up to
1316 * the current jiffies and restarts the counters again.
1317 */
Tejun Heoc9959052008-08-25 19:47:21 +09001318void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001319{
1320 unsigned long now = jiffies;
1321
Tejun Heo074a7ac2008-08-25 19:56:14 +09001322 if (part->partno)
1323 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1324 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001325}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001326EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001327
Lin Mingc8158812013-03-23 11:42:27 +08001328#ifdef CONFIG_PM_RUNTIME
1329static void blk_pm_put_request(struct request *rq)
1330{
1331 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1332 pm_runtime_mark_last_busy(rq->q->dev);
1333}
1334#else
1335static inline void blk_pm_put_request(struct request *rq) {}
1336#endif
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338/*
1339 * queue lock must be held
1340 */
Jens Axboe165125e2007-07-24 09:28:11 +02001341void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 if (unlikely(!q))
1344 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001346 if (q->mq_ops) {
1347 blk_mq_free_request(req);
1348 return;
1349 }
1350
Lin Mingc8158812013-03-23 11:42:27 +08001351 blk_pm_put_request(req);
1352
Tejun Heo8922e162005-10-20 16:23:44 +02001353 elv_completed_request(q, req);
1354
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001355 /* this is a bio leak */
1356 WARN_ON(req->bio != NULL);
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 /*
1359 * Request may not have originated from ll_rw_blk. if not,
1360 * it didn't come out of our reserved rq pools
1361 */
Jens Axboe49171e52006-08-10 08:59:11 +02001362 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001363 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001364 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001367 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Tejun Heoa0516612012-06-26 15:05:44 -07001369 blk_free_request(rl, req);
1370 freed_request(rl, flags);
1371 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
1373}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001374EXPORT_SYMBOL_GPL(__blk_put_request);
1375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376void blk_put_request(struct request *req)
1377{
Jens Axboe165125e2007-07-24 09:28:11 +02001378 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Jens Axboe320ae512013-10-24 09:20:05 +01001380 if (q->mq_ops)
1381 blk_mq_free_request(req);
1382 else {
1383 unsigned long flags;
1384
1385 spin_lock_irqsave(q->queue_lock, flags);
1386 __blk_put_request(q, req);
1387 spin_unlock_irqrestore(q->queue_lock, flags);
1388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390EXPORT_SYMBOL(blk_put_request);
1391
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001392/**
1393 * blk_add_request_payload - add a payload to a request
1394 * @rq: request to update
1395 * @page: page backing the payload
1396 * @len: length of the payload.
1397 *
1398 * This allows to later add a payload to an already submitted request by
1399 * a block driver. The driver needs to take care of freeing the payload
1400 * itself.
1401 *
1402 * Note that this is a quite horrible hack and nothing but handling of
1403 * discard requests should ever use it.
1404 */
1405void blk_add_request_payload(struct request *rq, struct page *page,
1406 unsigned int len)
1407{
1408 struct bio *bio = rq->bio;
1409
1410 bio->bi_io_vec->bv_page = page;
1411 bio->bi_io_vec->bv_offset = 0;
1412 bio->bi_io_vec->bv_len = len;
1413
Kent Overstreet4f024f32013-10-11 15:44:27 -07001414 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001415 bio->bi_vcnt = 1;
1416 bio->bi_phys_segments = 1;
1417
1418 rq->__data_len = rq->resid_len = len;
1419 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001420}
1421EXPORT_SYMBOL_GPL(blk_add_request_payload);
1422
Jens Axboe320ae512013-10-24 09:20:05 +01001423bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1424 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001425{
1426 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1427
Jens Axboe73c10102011-03-08 13:19:51 +01001428 if (!ll_back_merge_fn(q, req, bio))
1429 return false;
1430
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001431 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001432
1433 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1434 blk_rq_set_mixed_merge(req);
1435
1436 req->biotail->bi_next = bio;
1437 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001438 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001439 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1440
Jens Axboe320ae512013-10-24 09:20:05 +01001441 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001442 return true;
1443}
1444
Jens Axboe320ae512013-10-24 09:20:05 +01001445bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1446 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001447{
1448 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001449
Jens Axboe73c10102011-03-08 13:19:51 +01001450 if (!ll_front_merge_fn(q, req, bio))
1451 return false;
1452
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001453 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001454
1455 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1456 blk_rq_set_mixed_merge(req);
1457
Jens Axboe73c10102011-03-08 13:19:51 +01001458 bio->bi_next = req->bio;
1459 req->bio = bio;
1460
Kent Overstreet4f024f32013-10-11 15:44:27 -07001461 req->__sector = bio->bi_iter.bi_sector;
1462 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001463 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1464
Jens Axboe320ae512013-10-24 09:20:05 +01001465 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001466 return true;
1467}
1468
Tejun Heobd87b582011-10-19 14:33:08 +02001469/**
Jens Axboe320ae512013-10-24 09:20:05 +01001470 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001471 * @q: request_queue new bio is being queued at
1472 * @bio: new bio being queued
1473 * @request_count: out parameter for number of traversed plugged requests
1474 *
1475 * Determine whether @bio being queued on @q can be merged with a request
1476 * on %current's plugged list. Returns %true if merge was successful,
1477 * otherwise %false.
1478 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001479 * Plugging coalesces IOs from the same issuer for the same purpose without
1480 * going through @q->queue_lock. As such it's more of an issuing mechanism
1481 * than scheduling, and the request, while may have elvpriv data, is not
1482 * added on the elevator at this point. In addition, we don't have
1483 * reliable access to the elevator outside queue lock. Only check basic
1484 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001485 *
1486 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001487 */
Jens Axboe320ae512013-10-24 09:20:05 +01001488bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1489 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001490{
1491 struct blk_plug *plug;
1492 struct request *rq;
1493 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001494 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001495
Tejun Heobd87b582011-10-19 14:33:08 +02001496 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001497 if (!plug)
1498 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001499 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001500
Shaohua Li92f399c2013-10-29 12:01:03 -06001501 if (q->mq_ops)
1502 plug_list = &plug->mq_list;
1503 else
1504 plug_list = &plug->list;
1505
1506 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001507 int el_ret;
1508
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001509 if (rq->q == q)
1510 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001511
Tejun Heo07c2bd32012-02-08 09:19:42 +01001512 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001513 continue;
1514
Tejun Heo050c8ea2012-02-08 09:19:38 +01001515 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001516 if (el_ret == ELEVATOR_BACK_MERGE) {
1517 ret = bio_attempt_back_merge(q, rq, bio);
1518 if (ret)
1519 break;
1520 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1521 ret = bio_attempt_front_merge(q, rq, bio);
1522 if (ret)
1523 break;
1524 }
1525 }
1526out:
1527 return ret;
1528}
1529
Jens Axboe86db1e22008-01-29 14:53:40 +01001530void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001531{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001532 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001533
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001534 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1535 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001536 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001537
Tejun Heo52d9e672006-01-06 09:49:58 +01001538 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001539 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001540 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001541 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001542}
1543
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001544void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001546 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001547 struct blk_plug *plug;
1548 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1549 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001550 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 /*
1553 * low level driver can indicate that it wants pages above a
1554 * certain limit bounced to low memory (ie for highmem, or even
1555 * ISA dma in theory)
1556 */
1557 blk_queue_bounce(q, &bio);
1558
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001559 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1560 bio_endio(bio, -EIO);
1561 return;
1562 }
1563
Tejun Heo4fed9472010-09-03 11:56:17 +02001564 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001565 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001566 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001567 goto get_rq;
1568 }
1569
Jens Axboe73c10102011-03-08 13:19:51 +01001570 /*
1571 * Check if we can merge with the plugged list before grabbing
1572 * any locks.
1573 */
Robert Elliottda41a582014-05-20 16:46:26 -05001574 if (!blk_queue_nomerges(q) &&
1575 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001576 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001577
1578 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001581 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001582 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001583 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001584 if (!attempt_back_merge(q, req))
1585 elv_merged_request(q, req, el_ret);
1586 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001587 }
Jens Axboe73c10102011-03-08 13:19:51 +01001588 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001589 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001590 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001591 if (!attempt_front_merge(q, req))
1592 elv_merged_request(q, req, el_ret);
1593 goto out_unlock;
1594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 }
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001598 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001599 * This sync check and mask will be re-done in init_request_from_bio(),
1600 * but we need to set it earlier to expose the sync flag to the
1601 * rq allocator and io schedulers.
1602 */
1603 rw_flags = bio_data_dir(bio);
1604 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001605 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001606
1607 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001608 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001609 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001610 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001611 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001612 if (IS_ERR(req)) {
1613 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
Tejun Heoda8303c2011-10-19 14:33:05 +02001614 goto out_unlock;
1615 }
Nick Piggind6344532005-06-28 20:45:14 -07001616
Nick Piggin450991b2005-06-28 20:45:13 -07001617 /*
1618 * After dropping the lock and possibly sleeping here, our request
1619 * may now be mergeable after it had proven unmergeable (above).
1620 * We don't worry about that case for efficiency. It won't happen
1621 * often, and the elevators are able to handle it.
1622 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001623 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
Tao Ma9562ad92011-10-24 16:11:30 +02001625 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001626 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001627
Jens Axboe73c10102011-03-08 13:19:51 +01001628 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001629 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001630 /*
1631 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001632 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001633 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001634 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001635 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001636 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001637 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001638 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001639 trace_block_plug(q);
1640 }
Jens Axboe73c10102011-03-08 13:19:51 +01001641 }
Shaohua Lia6327162011-08-24 16:04:32 +02001642 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001643 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001644 } else {
1645 spin_lock_irq(q->queue_lock);
1646 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001647 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001648out_unlock:
1649 spin_unlock_irq(q->queue_lock);
1650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651}
Jens Axboec20e8de2011-09-12 12:03:37 +02001652EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654/*
1655 * If bio->bi_dev is a partition, remap the location
1656 */
1657static inline void blk_partition_remap(struct bio *bio)
1658{
1659 struct block_device *bdev = bio->bi_bdev;
1660
Jens Axboebf2de6f2007-09-27 13:01:25 +02001661 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001663
Kent Overstreet4f024f32013-10-11 15:44:27 -07001664 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001666
Mike Snitzerd07335e2010-11-16 12:52:38 +01001667 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1668 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001669 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
1671}
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673static void handle_bad_sector(struct bio *bio)
1674{
1675 char b[BDEVNAME_SIZE];
1676
1677 printk(KERN_INFO "attempt to access beyond end of device\n");
1678 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1679 bdevname(bio->bi_bdev, b),
1680 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001681 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001682 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
1684 set_bit(BIO_EOF, &bio->bi_flags);
1685}
1686
Akinobu Mitac17bb492006-12-08 02:39:46 -08001687#ifdef CONFIG_FAIL_MAKE_REQUEST
1688
1689static DECLARE_FAULT_ATTR(fail_make_request);
1690
1691static int __init setup_fail_make_request(char *str)
1692{
1693 return setup_fault_attr(&fail_make_request, str);
1694}
1695__setup("fail_make_request=", setup_fail_make_request);
1696
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001697static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001698{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001699 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001700}
1701
1702static int __init fail_make_request_debugfs(void)
1703{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001704 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1705 NULL, &fail_make_request);
1706
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001707 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001708}
1709
1710late_initcall(fail_make_request_debugfs);
1711
1712#else /* CONFIG_FAIL_MAKE_REQUEST */
1713
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001714static inline bool should_fail_request(struct hd_struct *part,
1715 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001716{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001717 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001718}
1719
1720#endif /* CONFIG_FAIL_MAKE_REQUEST */
1721
Jens Axboec07e2b42007-07-18 13:27:58 +02001722/*
1723 * Check whether this bio extends beyond the end of the device.
1724 */
1725static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1726{
1727 sector_t maxsector;
1728
1729 if (!nr_sectors)
1730 return 0;
1731
1732 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001733 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001734 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001735 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001736
1737 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1738 /*
1739 * This may well happen - the kernel calls bread()
1740 * without checking the size of the device, e.g., when
1741 * mounting a device.
1742 */
1743 handle_bad_sector(bio);
1744 return 1;
1745 }
1746 }
1747
1748 return 0;
1749}
1750
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001751static noinline_for_stack bool
1752generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Jens Axboe165125e2007-07-24 09:28:11 +02001754 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001755 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001756 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001757 char b[BDEVNAME_SIZE];
1758 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
1760 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
Jens Axboec07e2b42007-07-18 13:27:58 +02001762 if (bio_check_eod(bio, nr_sectors))
1763 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001765 q = bdev_get_queue(bio->bi_bdev);
1766 if (unlikely(!q)) {
1767 printk(KERN_ERR
1768 "generic_make_request: Trying to access "
1769 "nonexistent block-device %s (%Lu)\n",
1770 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001771 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001772 goto end_io;
1773 }
1774
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001775 if (likely(bio_is_rw(bio) &&
1776 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001777 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1778 bdevname(bio->bi_bdev, b),
1779 bio_sectors(bio),
1780 queue_max_hw_sectors(q));
1781 goto end_io;
1782 }
1783
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001784 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001785 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001786 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001787 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001788 goto end_io;
1789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001791 * If this device has partitions, remap block n
1792 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001794 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001796 if (bio_check_eod(bio, nr_sectors))
1797 goto end_io;
1798
1799 /*
1800 * Filter flush bio's early so that make_request based
1801 * drivers without flush support don't have to worry
1802 * about them.
1803 */
1804 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1805 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1806 if (!nr_sectors) {
1807 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001808 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001812 if ((bio->bi_rw & REQ_DISCARD) &&
1813 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001814 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001815 err = -EOPNOTSUPP;
1816 goto end_io;
1817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001819 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 err = -EOPNOTSUPP;
1821 goto end_io;
1822 }
1823
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001824 /*
1825 * Various block parts want %current->io_context and lazy ioc
1826 * allocation ends up trading a lot of pain for a small amount of
1827 * memory. Just allocate it upfront. This may fail and block
1828 * layer knows how to live with it.
1829 */
1830 create_io_context(GFP_ATOMIC, q->node);
1831
Tejun Heobc16a4f2011-10-19 14:33:01 +02001832 if (blk_throtl_bio(q, bio))
1833 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001834
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001835 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001836 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001837
1838end_io:
1839 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001840 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841}
1842
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001843/**
1844 * generic_make_request - hand a buffer to its device driver for I/O
1845 * @bio: The bio describing the location in memory and on the device.
1846 *
1847 * generic_make_request() is used to make I/O requests of block
1848 * devices. It is passed a &struct bio, which describes the I/O that needs
1849 * to be done.
1850 *
1851 * generic_make_request() does not return any status. The
1852 * success/failure status of the request, along with notification of
1853 * completion, is delivered asynchronously through the bio->bi_end_io
1854 * function described (one day) else where.
1855 *
1856 * The caller of generic_make_request must make sure that bi_io_vec
1857 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1858 * set to describe the device address, and the
1859 * bi_end_io and optionally bi_private are set to describe how
1860 * completion notification should be signaled.
1861 *
1862 * generic_make_request and the drivers it calls may use bi_next if this
1863 * bio happens to be merged with someone else, and may resubmit the bio to
1864 * a lower device by calling into generic_make_request recursively, which
1865 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001866 */
1867void generic_make_request(struct bio *bio)
1868{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001869 struct bio_list bio_list_on_stack;
1870
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001871 if (!generic_make_request_checks(bio))
1872 return;
1873
1874 /*
1875 * We only want one ->make_request_fn to be active at a time, else
1876 * stack usage with stacked devices could be a problem. So use
1877 * current->bio_list to keep a list of requests submited by a
1878 * make_request_fn function. current->bio_list is also used as a
1879 * flag to say if generic_make_request is currently active in this
1880 * task or not. If it is NULL, then no make_request is active. If
1881 * it is non-NULL, then a make_request is active, and new requests
1882 * should be added at the tail
1883 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001884 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001885 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001886 return;
1887 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001888
Neil Brownd89d8792007-05-01 09:53:42 +02001889 /* following loop may be a bit non-obvious, and so deserves some
1890 * explanation.
1891 * Before entering the loop, bio->bi_next is NULL (as all callers
1892 * ensure that) so we have a list with a single bio.
1893 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001894 * we assign bio_list to a pointer to the bio_list_on_stack,
1895 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001896 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001897 * through a recursive call to generic_make_request. If it
1898 * did, we find a non-NULL value in bio_list and re-enter the loop
1899 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001900 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001901 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001902 */
1903 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001904 bio_list_init(&bio_list_on_stack);
1905 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001906 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001907 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1908
1909 q->make_request_fn(q, bio);
1910
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001911 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001912 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001913 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001914}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915EXPORT_SYMBOL(generic_make_request);
1916
1917/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001918 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1920 * @bio: The &struct bio which describes the I/O
1921 *
1922 * submit_bio() is very similar in purpose to generic_make_request(), and
1923 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001924 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 *
1926 */
1927void submit_bio(int rw, struct bio *bio)
1928{
Jens Axboe22e2c502005-06-27 10:55:12 +02001929 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Jens Axboebf2de6f2007-09-27 13:01:25 +02001931 /*
1932 * If it's a regular read/write or a barrier with data attached,
1933 * go through the normal accounting stuff before submission.
1934 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001935 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001936 unsigned int count;
1937
1938 if (unlikely(rw & REQ_WRITE_SAME))
1939 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1940 else
1941 count = bio_sectors(bio);
1942
Jens Axboebf2de6f2007-09-27 13:01:25 +02001943 if (rw & WRITE) {
1944 count_vm_events(PGPGOUT, count);
1945 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001946 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001947 count_vm_events(PGPGIN, count);
1948 }
1949
1950 if (unlikely(block_dump)) {
1951 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001952 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001953 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001954 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001955 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001956 bdevname(bio->bi_bdev, b),
1957 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 }
1960
1961 generic_make_request(bio);
1962}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963EXPORT_SYMBOL(submit_bio);
1964
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001965/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001966 * blk_rq_check_limits - Helper function to check a request for the queue limit
1967 * @q: the queue
1968 * @rq: the request being checked
1969 *
1970 * Description:
1971 * @rq may have been made based on weaker limitations of upper-level queues
1972 * in request stacking drivers, and it may violate the limitation of @q.
1973 * Since the block layer and the underlying device driver trust @rq
1974 * after it is inserted to @q, it should be checked against @q before
1975 * the insertion using this generic function.
1976 *
1977 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001978 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001979 * Request stacking drivers like request-based dm may change the queue
1980 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09001981 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001982 * the new queue limits again when they dispatch those requests,
1983 * although such checkings are also done against the old queue limits
1984 * when submitting requests.
1985 */
1986int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1987{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001988 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001989 return 0;
1990
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001991 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001992 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1993 return -EIO;
1994 }
1995
1996 /*
1997 * queue's settings related to segment counting like q->bounce_pfn
1998 * may differ from that of other stacking queues.
1999 * Recalculate it to check the request correctly on this queue's
2000 * limitation.
2001 */
2002 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002003 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002004 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2005 return -EIO;
2006 }
2007
2008 return 0;
2009}
2010EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2011
2012/**
2013 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2014 * @q: the queue to submit the request
2015 * @rq: the request being queued
2016 */
2017int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2018{
2019 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002020 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002021
2022 if (blk_rq_check_limits(q, rq))
2023 return -EIO;
2024
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002025 if (rq->rq_disk &&
2026 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002027 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002028
2029 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002030 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002031 spin_unlock_irqrestore(q->queue_lock, flags);
2032 return -ENODEV;
2033 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002034
2035 /*
2036 * Submitting request must be dequeued before calling this function
2037 * because it will be linked to another request_queue
2038 */
2039 BUG_ON(blk_queued_rq(rq));
2040
Jeff Moyer4853aba2011-08-15 21:37:25 +02002041 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2042 where = ELEVATOR_INSERT_FLUSH;
2043
2044 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002045 if (where == ELEVATOR_INSERT_FLUSH)
2046 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002047 spin_unlock_irqrestore(q->queue_lock, flags);
2048
2049 return 0;
2050}
2051EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2052
Tejun Heo80a761f2009-07-03 17:48:17 +09002053/**
2054 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2055 * @rq: request to examine
2056 *
2057 * Description:
2058 * A request could be merge of IOs which require different failure
2059 * handling. This function determines the number of bytes which
2060 * can be failed from the beginning of the request without
2061 * crossing into area which need to be retried further.
2062 *
2063 * Return:
2064 * The number of bytes to fail.
2065 *
2066 * Context:
2067 * queue_lock must be held.
2068 */
2069unsigned int blk_rq_err_bytes(const struct request *rq)
2070{
2071 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2072 unsigned int bytes = 0;
2073 struct bio *bio;
2074
2075 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2076 return blk_rq_bytes(rq);
2077
2078 /*
2079 * Currently the only 'mixing' which can happen is between
2080 * different fastfail types. We can safely fail portions
2081 * which have all the failfast bits that the first one has -
2082 * the ones which are at least as eager to fail as the first
2083 * one.
2084 */
2085 for (bio = rq->bio; bio; bio = bio->bi_next) {
2086 if ((bio->bi_rw & ff) != ff)
2087 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002088 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002089 }
2090
2091 /* this could lead to infinite loop */
2092 BUG_ON(blk_rq_bytes(rq) && !bytes);
2093 return bytes;
2094}
2095EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2096
Jens Axboe320ae512013-10-24 09:20:05 +01002097void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002098{
Jens Axboec2553b52009-04-24 08:10:11 +02002099 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002100 const int rw = rq_data_dir(req);
2101 struct hd_struct *part;
2102 int cpu;
2103
2104 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002105 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002106 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2107 part_stat_unlock();
2108 }
2109}
2110
Jens Axboe320ae512013-10-24 09:20:05 +01002111void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002112{
Jens Axboebc58ba92009-01-23 10:54:44 +01002113 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002114 * Account IO completion. flush_rq isn't accounted as a
2115 * normal IO on queueing nor completion. Accounting the
2116 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002117 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002118 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002119 unsigned long duration = jiffies - req->start_time;
2120 const int rw = rq_data_dir(req);
2121 struct hd_struct *part;
2122 int cpu;
2123
2124 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002125 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002126
2127 part_stat_inc(cpu, part, ios[rw]);
2128 part_stat_add(cpu, part, ticks[rw], duration);
2129 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002130 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002131
Jens Axboe6c23a962011-01-07 08:43:37 +01002132 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002133 part_stat_unlock();
2134 }
2135}
2136
Lin Mingc8158812013-03-23 11:42:27 +08002137#ifdef CONFIG_PM_RUNTIME
2138/*
2139 * Don't process normal requests when queue is suspended
2140 * or in the process of suspending/resuming
2141 */
2142static struct request *blk_pm_peek_request(struct request_queue *q,
2143 struct request *rq)
2144{
2145 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2146 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2147 return NULL;
2148 else
2149 return rq;
2150}
2151#else
2152static inline struct request *blk_pm_peek_request(struct request_queue *q,
2153 struct request *rq)
2154{
2155 return rq;
2156}
2157#endif
2158
Jens Axboe320ae512013-10-24 09:20:05 +01002159void blk_account_io_start(struct request *rq, bool new_io)
2160{
2161 struct hd_struct *part;
2162 int rw = rq_data_dir(rq);
2163 int cpu;
2164
2165 if (!blk_do_io_stat(rq))
2166 return;
2167
2168 cpu = part_stat_lock();
2169
2170 if (!new_io) {
2171 part = rq->part;
2172 part_stat_inc(cpu, part, merges[rw]);
2173 } else {
2174 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2175 if (!hd_struct_try_get(part)) {
2176 /*
2177 * The partition is already being removed,
2178 * the request will be accounted on the disk only
2179 *
2180 * We take a reference on disk->part0 although that
2181 * partition will never be deleted, so we can treat
2182 * it as any other partition.
2183 */
2184 part = &rq->rq_disk->part0;
2185 hd_struct_get(part);
2186 }
2187 part_round_stats(cpu, part);
2188 part_inc_in_flight(part, rw);
2189 rq->part = part;
2190 }
2191
2192 part_stat_unlock();
2193}
2194
Tejun Heo53a08802008-12-03 12:41:26 +01002195/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002196 * blk_peek_request - peek at the top of a request queue
2197 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002198 *
2199 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002200 * Return the request at the top of @q. The returned request
2201 * should be started using blk_start_request() before LLD starts
2202 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002203 *
2204 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002205 * Pointer to the request at the top of @q if available. Null
2206 * otherwise.
2207 *
2208 * Context:
2209 * queue_lock must be held.
2210 */
2211struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002212{
2213 struct request *rq;
2214 int ret;
2215
2216 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002217
2218 rq = blk_pm_peek_request(q, rq);
2219 if (!rq)
2220 break;
2221
Tejun Heo158dbda2009-04-23 11:05:18 +09002222 if (!(rq->cmd_flags & REQ_STARTED)) {
2223 /*
2224 * This is the first time the device driver
2225 * sees this request (possibly after
2226 * requeueing). Notify IO scheduler.
2227 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002228 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002229 elv_activate_rq(q, rq);
2230
2231 /*
2232 * just mark as started even if we don't start
2233 * it, a request that has been delayed should
2234 * not be passed by new incoming requests
2235 */
2236 rq->cmd_flags |= REQ_STARTED;
2237 trace_block_rq_issue(q, rq);
2238 }
2239
2240 if (!q->boundary_rq || q->boundary_rq == rq) {
2241 q->end_sector = rq_end_sector(rq);
2242 q->boundary_rq = NULL;
2243 }
2244
2245 if (rq->cmd_flags & REQ_DONTPREP)
2246 break;
2247
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002248 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002249 /*
2250 * make sure space for the drain appears we
2251 * know we can do this because max_hw_segments
2252 * has been adjusted to be one fewer than the
2253 * device can handle
2254 */
2255 rq->nr_phys_segments++;
2256 }
2257
2258 if (!q->prep_rq_fn)
2259 break;
2260
2261 ret = q->prep_rq_fn(q, rq);
2262 if (ret == BLKPREP_OK) {
2263 break;
2264 } else if (ret == BLKPREP_DEFER) {
2265 /*
2266 * the request may have been (partially) prepped.
2267 * we need to keep this request in the front to
2268 * avoid resource deadlock. REQ_STARTED will
2269 * prevent other fs requests from passing this one.
2270 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002271 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002272 !(rq->cmd_flags & REQ_DONTPREP)) {
2273 /*
2274 * remove the space for the drain we added
2275 * so that we don't add it again
2276 */
2277 --rq->nr_phys_segments;
2278 }
2279
2280 rq = NULL;
2281 break;
2282 } else if (ret == BLKPREP_KILL) {
2283 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002284 /*
2285 * Mark this request as started so we don't trigger
2286 * any debug logic in the end I/O path.
2287 */
2288 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002289 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002290 } else {
2291 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2292 break;
2293 }
2294 }
2295
2296 return rq;
2297}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002298EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002299
Tejun Heo9934c8c2009-05-08 11:54:16 +09002300void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002301{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002302 struct request_queue *q = rq->q;
2303
Tejun Heo158dbda2009-04-23 11:05:18 +09002304 BUG_ON(list_empty(&rq->queuelist));
2305 BUG_ON(ELV_ON_HASH(rq));
2306
2307 list_del_init(&rq->queuelist);
2308
2309 /*
2310 * the time frame between a request being removed from the lists
2311 * and to it is freed is accounted as io that is in progress at
2312 * the driver side.
2313 */
Divyesh Shah91952912010-04-01 15:01:41 -07002314 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002315 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002316 set_io_start_time_ns(rq);
2317 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002318}
2319
Tejun Heo5efccd12009-04-23 11:05:18 +09002320/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002321 * blk_start_request - start request processing on the driver
2322 * @req: request to dequeue
2323 *
2324 * Description:
2325 * Dequeue @req and start timeout timer on it. This hands off the
2326 * request to the driver.
2327 *
2328 * Block internal functions which don't want to start timer should
2329 * call blk_dequeue_request().
2330 *
2331 * Context:
2332 * queue_lock must be held.
2333 */
2334void blk_start_request(struct request *req)
2335{
2336 blk_dequeue_request(req);
2337
2338 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002339 * We are now handing the request to the hardware, initialize
2340 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002341 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002342 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002343 if (unlikely(blk_bidi_rq(req)))
2344 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2345
Jeff Moyer4912aa62013-10-08 14:36:41 -04002346 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002347 blk_add_timer(req);
2348}
2349EXPORT_SYMBOL(blk_start_request);
2350
2351/**
2352 * blk_fetch_request - fetch a request from a request queue
2353 * @q: request queue to fetch a request from
2354 *
2355 * Description:
2356 * Return the request at the top of @q. The request is started on
2357 * return and LLD can start processing it immediately.
2358 *
2359 * Return:
2360 * Pointer to the request at the top of @q if available. Null
2361 * otherwise.
2362 *
2363 * Context:
2364 * queue_lock must be held.
2365 */
2366struct request *blk_fetch_request(struct request_queue *q)
2367{
2368 struct request *rq;
2369
2370 rq = blk_peek_request(q);
2371 if (rq)
2372 blk_start_request(rq);
2373 return rq;
2374}
2375EXPORT_SYMBOL(blk_fetch_request);
2376
2377/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002378 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002379 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002380 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002381 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002382 *
2383 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002384 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2385 * the request structure even if @req doesn't have leftover.
2386 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002387 *
2388 * This special helper function is only for request stacking drivers
2389 * (e.g. request-based dm) so that they can handle partial completion.
2390 * Actual device drivers should use blk_end_request instead.
2391 *
2392 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2393 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002394 *
2395 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002396 * %false - this request doesn't have any more data
2397 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002398 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002399bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002401 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002403 trace_block_rq_complete(req->q, req, nr_bytes);
2404
Tejun Heo2e60e022009-04-23 11:05:18 +09002405 if (!req->bio)
2406 return false;
2407
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002409 * For fs requests, rq is just carrier of independent bio's
2410 * and each partial completion should be handled separately.
2411 * Reset per-request error on each partial completion.
2412 *
2413 * TODO: tj: This is too subtle. It would be better to let
2414 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002416 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 req->errors = 0;
2418
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002419 if (error && req->cmd_type == REQ_TYPE_FS &&
2420 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002421 char *error_type;
2422
2423 switch (error) {
2424 case -ENOLINK:
2425 error_type = "recoverable transport";
2426 break;
2427 case -EREMOTEIO:
2428 error_type = "critical target";
2429 break;
2430 case -EBADE:
2431 error_type = "critical nexus";
2432 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002433 case -ETIMEDOUT:
2434 error_type = "timeout";
2435 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002436 case -ENOSPC:
2437 error_type = "critical space allocation";
2438 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002439 case -ENODATA:
2440 error_type = "critical medium";
2441 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002442 case -EIO:
2443 default:
2444 error_type = "I/O";
2445 break;
2446 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002447 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2448 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002449 req->rq_disk->disk_name : "?",
2450 (unsigned long long)blk_rq_pos(req));
2451
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 }
2453
Jens Axboebc58ba92009-01-23 10:54:44 +01002454 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002455
Kent Overstreetf79ea412012-09-20 16:38:30 -07002456 total_bytes = 0;
2457 while (req->bio) {
2458 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002459 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Kent Overstreet4f024f32013-10-11 15:44:27 -07002461 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Kent Overstreetf79ea412012-09-20 16:38:30 -07002464 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Kent Overstreetf79ea412012-09-20 16:38:30 -07002466 total_bytes += bio_bytes;
2467 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Kent Overstreetf79ea412012-09-20 16:38:30 -07002469 if (!nr_bytes)
2470 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
2472
2473 /*
2474 * completely done
2475 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002476 if (!req->bio) {
2477 /*
2478 * Reset counters so that the request stacking driver
2479 * can find how many bytes remain in the request
2480 * later.
2481 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002482 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002483 return false;
2484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002486 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002487
2488 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002489 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002490 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002491
Tejun Heo80a761f2009-07-03 17:48:17 +09002492 /* mixed attributes always follow the first bio */
2493 if (req->cmd_flags & REQ_MIXED_MERGE) {
2494 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2495 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2496 }
2497
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002498 /*
2499 * If total number of sectors is less than the first segment
2500 * size, something has gone terribly wrong.
2501 */
2502 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002503 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002504 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002505 }
2506
2507 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002509
Tejun Heo2e60e022009-04-23 11:05:18 +09002510 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511}
Tejun Heo2e60e022009-04-23 11:05:18 +09002512EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Tejun Heo2e60e022009-04-23 11:05:18 +09002514static bool blk_update_bidi_request(struct request *rq, int error,
2515 unsigned int nr_bytes,
2516 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002517{
Tejun Heo2e60e022009-04-23 11:05:18 +09002518 if (blk_update_request(rq, error, nr_bytes))
2519 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002520
Tejun Heo2e60e022009-04-23 11:05:18 +09002521 /* Bidi request must be completed as a whole */
2522 if (unlikely(blk_bidi_rq(rq)) &&
2523 blk_update_request(rq->next_rq, error, bidi_bytes))
2524 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002525
Jens Axboee2e1a142010-06-09 10:42:09 +02002526 if (blk_queue_add_random(rq->q))
2527 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002528
2529 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530}
2531
James Bottomley28018c22010-07-01 19:49:17 +09002532/**
2533 * blk_unprep_request - unprepare a request
2534 * @req: the request
2535 *
2536 * This function makes a request ready for complete resubmission (or
2537 * completion). It happens only after all error handling is complete,
2538 * so represents the appropriate moment to deallocate any resources
2539 * that were allocated to the request in the prep_rq_fn. The queue
2540 * lock is held when calling this.
2541 */
2542void blk_unprep_request(struct request *req)
2543{
2544 struct request_queue *q = req->q;
2545
2546 req->cmd_flags &= ~REQ_DONTPREP;
2547 if (q->unprep_rq_fn)
2548 q->unprep_rq_fn(q, req);
2549}
2550EXPORT_SYMBOL_GPL(blk_unprep_request);
2551
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552/*
2553 * queue lock must be held
2554 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002555void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002557 if (blk_rq_tagged(req))
2558 blk_queue_end_tag(req->q, req);
2559
James Bottomleyba396a62009-05-27 14:17:08 +02002560 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002562 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002563 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
Mike Andersone78042e2008-10-30 02:16:20 -07002565 blk_delete_timer(req);
2566
James Bottomley28018c22010-07-01 19:49:17 +09002567 if (req->cmd_flags & REQ_DONTPREP)
2568 blk_unprep_request(req);
2569
Jens Axboebc58ba92009-01-23 10:54:44 +01002570 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002571
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002573 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002574 else {
2575 if (blk_bidi_rq(req))
2576 __blk_put_request(req->next_rq->q, req->next_rq);
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 }
2580}
Christoph Hellwig12120072014-04-16 09:44:59 +02002581EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002583/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002584 * blk_end_bidi_request - Complete a bidi request
2585 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002586 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002587 * @nr_bytes: number of bytes to complete @rq
2588 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002589 *
2590 * Description:
2591 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002592 * Drivers that supports bidi can safely call this member for any
2593 * type of request, bidi or uni. In the later case @bidi_bytes is
2594 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002595 *
2596 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002597 * %false - we are done with this request
2598 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002599 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002600static bool blk_end_bidi_request(struct request *rq, int error,
2601 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002602{
2603 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002604 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002605
Tejun Heo2e60e022009-04-23 11:05:18 +09002606 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2607 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002608
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002609 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002610 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002611 spin_unlock_irqrestore(q->queue_lock, flags);
2612
Tejun Heo2e60e022009-04-23 11:05:18 +09002613 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002614}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002615
2616/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002617 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2618 * @rq: the request to complete
2619 * @error: %0 for success, < %0 for error
2620 * @nr_bytes: number of bytes to complete @rq
2621 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002622 *
2623 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002624 * Identical to blk_end_bidi_request() except that queue lock is
2625 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002626 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002627 * Return:
2628 * %false - we are done with this request
2629 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002630 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002631bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002632 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002633{
Tejun Heo2e60e022009-04-23 11:05:18 +09002634 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2635 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002636
Tejun Heo2e60e022009-04-23 11:05:18 +09002637 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002638
Tejun Heo2e60e022009-04-23 11:05:18 +09002639 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002640}
2641
2642/**
2643 * blk_end_request - Helper function for drivers to complete the request.
2644 * @rq: the request being processed
2645 * @error: %0 for success, < %0 for error
2646 * @nr_bytes: number of bytes to complete
2647 *
2648 * Description:
2649 * Ends I/O on a number of bytes attached to @rq.
2650 * If @rq has leftover, sets it up for the next range of segments.
2651 *
2652 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002653 * %false - we are done with this request
2654 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002655 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002656bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002657{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002658 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002659}
Jens Axboe56ad1742009-07-28 22:11:24 +02002660EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002661
2662/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002663 * blk_end_request_all - Helper function for drives to finish the request.
2664 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002665 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002666 *
2667 * Description:
2668 * Completely finish @rq.
2669 */
2670void blk_end_request_all(struct request *rq, int error)
2671{
2672 bool pending;
2673 unsigned int bidi_bytes = 0;
2674
2675 if (unlikely(blk_bidi_rq(rq)))
2676 bidi_bytes = blk_rq_bytes(rq->next_rq);
2677
2678 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2679 BUG_ON(pending);
2680}
Jens Axboe56ad1742009-07-28 22:11:24 +02002681EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002682
2683/**
2684 * blk_end_request_cur - Helper function to finish the current request chunk.
2685 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002686 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002687 *
2688 * Description:
2689 * Complete the current consecutively mapped chunk from @rq.
2690 *
2691 * Return:
2692 * %false - we are done with this request
2693 * %true - still buffers pending for this request
2694 */
2695bool blk_end_request_cur(struct request *rq, int error)
2696{
2697 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2698}
Jens Axboe56ad1742009-07-28 22:11:24 +02002699EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002700
2701/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002702 * blk_end_request_err - Finish a request till the next failure boundary.
2703 * @rq: the request to finish till the next failure boundary for
2704 * @error: must be negative errno
2705 *
2706 * Description:
2707 * Complete @rq till the next failure boundary.
2708 *
2709 * Return:
2710 * %false - we are done with this request
2711 * %true - still buffers pending for this request
2712 */
2713bool blk_end_request_err(struct request *rq, int error)
2714{
2715 WARN_ON(error >= 0);
2716 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2717}
2718EXPORT_SYMBOL_GPL(blk_end_request_err);
2719
2720/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002721 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002722 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002723 * @error: %0 for success, < %0 for error
2724 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002725 *
2726 * Description:
2727 * Must be called with queue lock held unlike blk_end_request().
2728 *
2729 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002730 * %false - we are done with this request
2731 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002732 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002733bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002734{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002735 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002736}
Jens Axboe56ad1742009-07-28 22:11:24 +02002737EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002738
2739/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002740 * __blk_end_request_all - Helper function for drives to finish the request.
2741 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002742 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002743 *
2744 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002745 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002746 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002747void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002748{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002749 bool pending;
2750 unsigned int bidi_bytes = 0;
2751
2752 if (unlikely(blk_bidi_rq(rq)))
2753 bidi_bytes = blk_rq_bytes(rq->next_rq);
2754
2755 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2756 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002757}
Jens Axboe56ad1742009-07-28 22:11:24 +02002758EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002759
2760/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002761 * __blk_end_request_cur - Helper function to finish the current request chunk.
2762 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002763 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002764 *
2765 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002766 * Complete the current consecutively mapped chunk from @rq. Must
2767 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002768 *
2769 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002770 * %false - we are done with this request
2771 * %true - still buffers pending for this request
2772 */
2773bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002774{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002775 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002776}
Jens Axboe56ad1742009-07-28 22:11:24 +02002777EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002778
Tejun Heo80a761f2009-07-03 17:48:17 +09002779/**
2780 * __blk_end_request_err - Finish a request till the next failure boundary.
2781 * @rq: the request to finish till the next failure boundary for
2782 * @error: must be negative errno
2783 *
2784 * Description:
2785 * Complete @rq till the next failure boundary. Must be called
2786 * with queue lock held.
2787 *
2788 * Return:
2789 * %false - we are done with this request
2790 * %true - still buffers pending for this request
2791 */
2792bool __blk_end_request_err(struct request *rq, int error)
2793{
2794 WARN_ON(error >= 0);
2795 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2796}
2797EXPORT_SYMBOL_GPL(__blk_end_request_err);
2798
Jens Axboe86db1e22008-01-29 14:53:40 +01002799void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2800 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002802 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002803 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804
Jens Axboeb4f42e22014-04-10 09:46:28 -06002805 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002806 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002807
Kent Overstreet4f024f32013-10-11 15:44:27 -07002808 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
NeilBrown66846572007-08-16 13:31:28 +02002811 if (bio->bi_bdev)
2812 rq->rq_disk = bio->bi_bdev->bd_disk;
2813}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002815#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2816/**
2817 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2818 * @rq: the request to be flushed
2819 *
2820 * Description:
2821 * Flush all pages in @rq.
2822 */
2823void rq_flush_dcache_pages(struct request *rq)
2824{
2825 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002826 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002827
2828 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002829 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002830}
2831EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2832#endif
2833
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002834/**
2835 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2836 * @q : the queue of the device being checked
2837 *
2838 * Description:
2839 * Check if underlying low-level drivers of a device are busy.
2840 * If the drivers want to export their busy state, they must set own
2841 * exporting function using blk_queue_lld_busy() first.
2842 *
2843 * Basically, this function is used only by request stacking drivers
2844 * to stop dispatching requests to underlying devices when underlying
2845 * devices are busy. This behavior helps more I/O merging on the queue
2846 * of the request stacking driver and prevents I/O throughput regression
2847 * on burst I/O load.
2848 *
2849 * Return:
2850 * 0 - Not busy (The request stacking driver should dispatch request)
2851 * 1 - Busy (The request stacking driver should stop dispatching request)
2852 */
2853int blk_lld_busy(struct request_queue *q)
2854{
2855 if (q->lld_busy_fn)
2856 return q->lld_busy_fn(q);
2857
2858 return 0;
2859}
2860EXPORT_SYMBOL_GPL(blk_lld_busy);
2861
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002862/**
2863 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2864 * @rq: the clone request to be cleaned up
2865 *
2866 * Description:
2867 * Free all bios in @rq for a cloned request.
2868 */
2869void blk_rq_unprep_clone(struct request *rq)
2870{
2871 struct bio *bio;
2872
2873 while ((bio = rq->bio) != NULL) {
2874 rq->bio = bio->bi_next;
2875
2876 bio_put(bio);
2877 }
2878}
2879EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2880
2881/*
2882 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002883 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002884 */
2885static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2886{
2887 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002888 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002889 dst->cmd_type = src->cmd_type;
2890 dst->__sector = blk_rq_pos(src);
2891 dst->__data_len = blk_rq_bytes(src);
2892 dst->nr_phys_segments = src->nr_phys_segments;
2893 dst->ioprio = src->ioprio;
2894 dst->extra_len = src->extra_len;
2895}
2896
2897/**
2898 * blk_rq_prep_clone - Helper function to setup clone request
2899 * @rq: the request to be setup
2900 * @rq_src: original request to be cloned
2901 * @bs: bio_set that bios for clone are allocated from
2902 * @gfp_mask: memory allocation mask for bio
2903 * @bio_ctr: setup function to be called for each clone bio.
2904 * Returns %0 for success, non %0 for failure.
2905 * @data: private data to be passed to @bio_ctr
2906 *
2907 * Description:
2908 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002909 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002910 * are not copied, and copying such parts is the caller's responsibility.
2911 * Also, pages which the original bios are pointing to are not copied
2912 * and the cloned bios just point same pages.
2913 * So cloned bios must be completed before original bios, which means
2914 * the caller must complete @rq before @rq_src.
2915 */
2916int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2917 struct bio_set *bs, gfp_t gfp_mask,
2918 int (*bio_ctr)(struct bio *, struct bio *, void *),
2919 void *data)
2920{
2921 struct bio *bio, *bio_src;
2922
2923 if (!bs)
2924 bs = fs_bio_set;
2925
2926 blk_rq_init(NULL, rq);
2927
2928 __rq_for_each_bio(bio_src, rq_src) {
Junichi Nomura11dfce52014-10-03 17:27:11 -04002929 bio = bio_clone_fast(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002930 if (!bio)
2931 goto free_and_out;
2932
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002933 if (bio_ctr && bio_ctr(bio, bio_src, data))
2934 goto free_and_out;
2935
2936 if (rq->bio) {
2937 rq->biotail->bi_next = bio;
2938 rq->biotail = bio;
2939 } else
2940 rq->bio = rq->biotail = bio;
2941 }
2942
2943 __blk_rq_prep_clone(rq, rq_src);
2944
2945 return 0;
2946
2947free_and_out:
2948 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002949 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002950 blk_rq_unprep_clone(rq);
2951
2952 return -ENOMEM;
2953}
2954EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2955
Jens Axboe59c3d452014-04-08 09:15:35 -06002956int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957{
2958 return queue_work(kblockd_workqueue, work);
2959}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960EXPORT_SYMBOL(kblockd_schedule_work);
2961
Jens Axboe59c3d452014-04-08 09:15:35 -06002962int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2963 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04002964{
2965 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2966}
2967EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2968
Jens Axboe8ab14592014-04-08 09:17:40 -06002969int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2970 unsigned long delay)
2971{
2972 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2973}
2974EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2975
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002976/**
2977 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2978 * @plug: The &struct blk_plug that needs to be initialized
2979 *
2980 * Description:
2981 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2982 * pending I/O should the task end up blocking between blk_start_plug() and
2983 * blk_finish_plug(). This is important from a performance perspective, but
2984 * also ensures that we don't deadlock. For instance, if the task is blocking
2985 * for a memory allocation, memory reclaim could end up wanting to free a
2986 * page belonging to that request that is currently residing in our private
2987 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2988 * this kind of deadlock.
2989 */
Jens Axboe73c10102011-03-08 13:19:51 +01002990void blk_start_plug(struct blk_plug *plug)
2991{
2992 struct task_struct *tsk = current;
2993
Jens Axboe73c10102011-03-08 13:19:51 +01002994 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01002995 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02002996 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002997
2998 /*
2999 * If this is a nested plug, don't actually assign it. It will be
3000 * flushed on its own.
3001 */
3002 if (!tsk->plug) {
3003 /*
3004 * Store ordering should not be needed here, since a potential
3005 * preempt will imply a full memory barrier
3006 */
3007 tsk->plug = plug;
3008 }
3009}
3010EXPORT_SYMBOL(blk_start_plug);
3011
3012static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3013{
3014 struct request *rqa = container_of(a, struct request, queuelist);
3015 struct request *rqb = container_of(b, struct request, queuelist);
3016
Jianpeng Ma975927b2012-10-25 21:58:17 +02003017 return !(rqa->q < rqb->q ||
3018 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003019}
3020
Jens Axboe49cac012011-04-16 13:51:05 +02003021/*
3022 * If 'from_schedule' is true, then postpone the dispatch of requests
3023 * until a safe kblockd context. We due this to avoid accidental big
3024 * additional stack usage in driver dispatch, in places where the originally
3025 * plugger did not intend it.
3026 */
Jens Axboef6603782011-04-15 15:49:07 +02003027static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003028 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003029 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003030{
Jens Axboe49cac012011-04-16 13:51:05 +02003031 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003032
Bart Van Assche70460572012-11-28 13:45:56 +01003033 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003034 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003035 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003036 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003037 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003038}
3039
NeilBrown74018dc2012-07-31 09:08:15 +02003040static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003041{
3042 LIST_HEAD(callbacks);
3043
Shaohua Li2a7d5552012-07-31 09:08:15 +02003044 while (!list_empty(&plug->cb_list)) {
3045 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003046
Shaohua Li2a7d5552012-07-31 09:08:15 +02003047 while (!list_empty(&callbacks)) {
3048 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003049 struct blk_plug_cb,
3050 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003051 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003052 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003053 }
NeilBrown048c9372011-04-18 09:52:22 +02003054 }
3055}
3056
NeilBrown9cbb1752012-07-31 09:08:14 +02003057struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3058 int size)
3059{
3060 struct blk_plug *plug = current->plug;
3061 struct blk_plug_cb *cb;
3062
3063 if (!plug)
3064 return NULL;
3065
3066 list_for_each_entry(cb, &plug->cb_list, list)
3067 if (cb->callback == unplug && cb->data == data)
3068 return cb;
3069
3070 /* Not currently on the callback list */
3071 BUG_ON(size < sizeof(*cb));
3072 cb = kzalloc(size, GFP_ATOMIC);
3073 if (cb) {
3074 cb->data = data;
3075 cb->callback = unplug;
3076 list_add(&cb->list, &plug->cb_list);
3077 }
3078 return cb;
3079}
3080EXPORT_SYMBOL(blk_check_plugged);
3081
Jens Axboe49cac012011-04-16 13:51:05 +02003082void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003083{
3084 struct request_queue *q;
3085 unsigned long flags;
3086 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003087 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003088 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003089
NeilBrown74018dc2012-07-31 09:08:15 +02003090 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003091
3092 if (!list_empty(&plug->mq_list))
3093 blk_mq_flush_plug_list(plug, from_schedule);
3094
Jens Axboe73c10102011-03-08 13:19:51 +01003095 if (list_empty(&plug->list))
3096 return;
3097
NeilBrown109b8122011-04-11 14:13:10 +02003098 list_splice_init(&plug->list, &list);
3099
Jianpeng Ma422765c2013-01-11 14:46:09 +01003100 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003101
3102 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003103 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003104
3105 /*
3106 * Save and disable interrupts here, to avoid doing it for every
3107 * queue lock we have to take.
3108 */
Jens Axboe73c10102011-03-08 13:19:51 +01003109 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003110 while (!list_empty(&list)) {
3111 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003112 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003113 BUG_ON(!rq->q);
3114 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003115 /*
3116 * This drops the queue lock
3117 */
3118 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003119 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003120 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003121 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003122 spin_lock(q->queue_lock);
3123 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003124
3125 /*
3126 * Short-circuit if @q is dead
3127 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003128 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003129 __blk_end_request_all(rq, -ENODEV);
3130 continue;
3131 }
3132
Jens Axboe73c10102011-03-08 13:19:51 +01003133 /*
3134 * rq is already accounted, so use raw insert
3135 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003136 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3137 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3138 else
3139 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003140
3141 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003142 }
3143
Jens Axboe99e22592011-04-18 09:59:55 +02003144 /*
3145 * This drops the queue lock
3146 */
3147 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003148 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003149
Jens Axboe73c10102011-03-08 13:19:51 +01003150 local_irq_restore(flags);
3151}
Jens Axboe73c10102011-03-08 13:19:51 +01003152
3153void blk_finish_plug(struct blk_plug *plug)
3154{
Jens Axboef6603782011-04-15 15:49:07 +02003155 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003156
3157 if (plug == current->plug)
3158 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003159}
3160EXPORT_SYMBOL(blk_finish_plug);
3161
Lin Ming6c954662013-03-23 11:42:26 +08003162#ifdef CONFIG_PM_RUNTIME
3163/**
3164 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3165 * @q: the queue of the device
3166 * @dev: the device the queue belongs to
3167 *
3168 * Description:
3169 * Initialize runtime-PM-related fields for @q and start auto suspend for
3170 * @dev. Drivers that want to take advantage of request-based runtime PM
3171 * should call this function after @dev has been initialized, and its
3172 * request queue @q has been allocated, and runtime PM for it can not happen
3173 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3174 * cases, driver should call this function before any I/O has taken place.
3175 *
3176 * This function takes care of setting up using auto suspend for the device,
3177 * the autosuspend delay is set to -1 to make runtime suspend impossible
3178 * until an updated value is either set by user or by driver. Drivers do
3179 * not need to touch other autosuspend settings.
3180 *
3181 * The block layer runtime PM is request based, so only works for drivers
3182 * that use request as their IO unit instead of those directly use bio's.
3183 */
3184void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3185{
3186 q->dev = dev;
3187 q->rpm_status = RPM_ACTIVE;
3188 pm_runtime_set_autosuspend_delay(q->dev, -1);
3189 pm_runtime_use_autosuspend(q->dev);
3190}
3191EXPORT_SYMBOL(blk_pm_runtime_init);
3192
3193/**
3194 * blk_pre_runtime_suspend - Pre runtime suspend check
3195 * @q: the queue of the device
3196 *
3197 * Description:
3198 * This function will check if runtime suspend is allowed for the device
3199 * by examining if there are any requests pending in the queue. If there
3200 * are requests pending, the device can not be runtime suspended; otherwise,
3201 * the queue's status will be updated to SUSPENDING and the driver can
3202 * proceed to suspend the device.
3203 *
3204 * For the not allowed case, we mark last busy for the device so that
3205 * runtime PM core will try to autosuspend it some time later.
3206 *
3207 * This function should be called near the start of the device's
3208 * runtime_suspend callback.
3209 *
3210 * Return:
3211 * 0 - OK to runtime suspend the device
3212 * -EBUSY - Device should not be runtime suspended
3213 */
3214int blk_pre_runtime_suspend(struct request_queue *q)
3215{
3216 int ret = 0;
3217
3218 spin_lock_irq(q->queue_lock);
3219 if (q->nr_pending) {
3220 ret = -EBUSY;
3221 pm_runtime_mark_last_busy(q->dev);
3222 } else {
3223 q->rpm_status = RPM_SUSPENDING;
3224 }
3225 spin_unlock_irq(q->queue_lock);
3226 return ret;
3227}
3228EXPORT_SYMBOL(blk_pre_runtime_suspend);
3229
3230/**
3231 * blk_post_runtime_suspend - Post runtime suspend processing
3232 * @q: the queue of the device
3233 * @err: return value of the device's runtime_suspend function
3234 *
3235 * Description:
3236 * Update the queue's runtime status according to the return value of the
3237 * device's runtime suspend function and mark last busy for the device so
3238 * that PM core will try to auto suspend the device at a later time.
3239 *
3240 * This function should be called near the end of the device's
3241 * runtime_suspend callback.
3242 */
3243void blk_post_runtime_suspend(struct request_queue *q, int err)
3244{
3245 spin_lock_irq(q->queue_lock);
3246 if (!err) {
3247 q->rpm_status = RPM_SUSPENDED;
3248 } else {
3249 q->rpm_status = RPM_ACTIVE;
3250 pm_runtime_mark_last_busy(q->dev);
3251 }
3252 spin_unlock_irq(q->queue_lock);
3253}
3254EXPORT_SYMBOL(blk_post_runtime_suspend);
3255
3256/**
3257 * blk_pre_runtime_resume - Pre runtime resume processing
3258 * @q: the queue of the device
3259 *
3260 * Description:
3261 * Update the queue's runtime status to RESUMING in preparation for the
3262 * runtime resume of the device.
3263 *
3264 * This function should be called near the start of the device's
3265 * runtime_resume callback.
3266 */
3267void blk_pre_runtime_resume(struct request_queue *q)
3268{
3269 spin_lock_irq(q->queue_lock);
3270 q->rpm_status = RPM_RESUMING;
3271 spin_unlock_irq(q->queue_lock);
3272}
3273EXPORT_SYMBOL(blk_pre_runtime_resume);
3274
3275/**
3276 * blk_post_runtime_resume - Post runtime resume processing
3277 * @q: the queue of the device
3278 * @err: return value of the device's runtime_resume function
3279 *
3280 * Description:
3281 * Update the queue's runtime status according to the return value of the
3282 * device's runtime_resume function. If it is successfully resumed, process
3283 * the requests that are queued into the device's queue when it is resuming
3284 * and then mark last busy and initiate autosuspend for it.
3285 *
3286 * This function should be called near the end of the device's
3287 * runtime_resume callback.
3288 */
3289void blk_post_runtime_resume(struct request_queue *q, int err)
3290{
3291 spin_lock_irq(q->queue_lock);
3292 if (!err) {
3293 q->rpm_status = RPM_ACTIVE;
3294 __blk_run_queue(q);
3295 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003296 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003297 } else {
3298 q->rpm_status = RPM_SUSPENDED;
3299 }
3300 spin_unlock_irq(q->queue_lock);
3301}
3302EXPORT_SYMBOL(blk_post_runtime_resume);
3303#endif
3304
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305int __init blk_dev_init(void)
3306{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003307 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3308 sizeof(((struct request *)0)->cmd_flags));
3309
Tejun Heo89b90be2011-01-03 15:01:47 +01003310 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3311 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003312 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 if (!kblockd_workqueue)
3314 panic("Failed to create kblockd\n");
3315
3316 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003317 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318
Jens Axboe8324aa92008-01-29 14:51:59 +01003319 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003320 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 return 0;
3323}