blob: c359d72e9d76f24a44b7c4b6b8c36c6677436191 [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
86 * backing_dev_info
87 *
88 * Will return NULL if the request queue cannot be located.
89 */
90struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91{
92 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +020093 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95 if (q)
96 ret = &q->backing_dev_info;
97 return ret;
98}
Linus Torvalds1da177e2005-04-16 15:20:36 -070099EXPORT_SYMBOL(blk_get_backing_dev_info);
100
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200101void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200103 memset(rq, 0, sizeof(*rq));
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700106 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200107 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100108 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900109 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200110 INIT_HLIST_NODE(&rq->hash);
111 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200112 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800113 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100114 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900115 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700116 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100117 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200119EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
NeilBrown5bb23a62007-09-27 12:46:13 +0200121static void req_bio_endio(struct request *rq, struct bio *bio,
122 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100123{
Tejun Heo143a87f2011-01-25 12:43:52 +0100124 if (error)
125 clear_bit(BIO_UPTODATE, &bio->bi_flags);
126 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
127 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100128
Tejun Heo143a87f2011-01-25 12:43:52 +0100129 if (unlikely(rq->cmd_flags & REQ_QUIET))
130 set_bit(BIO_QUIET, &bio->bi_flags);
131
Kent Overstreetf79ea412012-09-20 16:38:30 -0700132 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100133
134 /* don't actually finish bio if it's part of flush sequence */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700135 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Tejun Heo143a87f2011-01-25 12:43:52 +0100136 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100137}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139void blk_dump_rq_flags(struct request *rq, char *msg)
140{
141 int bit;
142
Jens Axboe59533162013-05-23 12:25:08 +0200143 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200144 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200145 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Tejun Heo83096eb2009-05-07 22:24:39 +0900147 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
148 (unsigned long long)blk_rq_pos(rq),
149 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600150 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
151 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200153 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100154 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200155 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 printk("%02x ", rq->cmd[bit]);
157 printk("\n");
158 }
159}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160EXPORT_SYMBOL(blk_dump_rq_flags);
161
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500162static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200163{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500164 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200165
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500166 q = container_of(work, struct request_queue, delay_work.work);
167 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200168 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500169 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500173 * blk_delay_queue - restart queueing after defined interval
174 * @q: The &struct request_queue in question
175 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 *
177 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500178 * Sometimes queueing needs to be postponed for a little while, to allow
179 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100180 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500181 */
182void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Bart Van Assche70460572012-11-28 13:45:56 +0100184 if (likely(!blk_queue_dead(q)))
185 queue_delayed_work(kblockd_workqueue, &q->delay_work,
186 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500188EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/**
191 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200192 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
194 * Description:
195 * blk_start_queue() will clear the stop flag on the queue, and call
196 * the request_fn for the queue if it was in a stopped state when
197 * entered. Also see blk_stop_queue(). Queue lock must be held.
198 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200199void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200201 WARN_ON(!irqs_disabled());
202
Nick Piggin75ad23b2008-04-29 14:48:33 +0200203 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200204 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206EXPORT_SYMBOL(blk_start_queue);
207
208/**
209 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200210 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 *
212 * Description:
213 * The Linux block layer assumes that a block driver will consume all
214 * entries on the request queue when the request_fn strategy is called.
215 * Often this will not happen, because of hardware limitations (queue
216 * depth settings). If a device driver gets a 'queue full' response,
217 * or if it simply chooses not to queue more I/O at one point, it can
218 * call this function to prevent the request_fn from being called until
219 * the driver has signalled it's ready to go again. This happens by calling
220 * blk_start_queue() to restart queue operations. Queue lock must be held.
221 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200222void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Tejun Heo136b5722012-08-21 13:18:24 -0700224 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200225 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227EXPORT_SYMBOL(blk_stop_queue);
228
229/**
230 * blk_sync_queue - cancel any pending callbacks on a queue
231 * @q: the queue
232 *
233 * Description:
234 * The block layer may perform asynchronous callback activity
235 * on a queue, such as calling the unplug function after a timeout.
236 * A block device may call blk_sync_queue to ensure that any
237 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200238 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 * that its ->make_request_fn will not re-add plugging prior to calling
240 * this function.
241 *
Vivek Goyalda527772011-03-02 19:05:33 -0500242 * This function does not cancel any asynchronous activity arising
243 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800244 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500245 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 */
247void blk_sync_queue(struct request_queue *q)
248{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100249 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800250
251 if (q->mq_ops) {
252 struct blk_mq_hw_ctx *hctx;
253 int i;
254
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600255 queue_for_each_hw_ctx(q, hctx, i) {
256 cancel_delayed_work_sync(&hctx->run_work);
257 cancel_delayed_work_sync(&hctx->delay_work);
258 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800259 } else {
260 cancel_delayed_work_sync(&q->delay_work);
261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263EXPORT_SYMBOL(blk_sync_queue);
264
265/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100266 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
267 * @q: The queue to run
268 *
269 * Description:
270 * Invoke request handling on a queue if there are any pending requests.
271 * May be used to restart request handling after a request has completed.
272 * This variant runs the queue whether or not the queue has been
273 * stopped. Must be called with the queue lock held and interrupts
274 * disabled. See also @blk_run_queue.
275 */
276inline void __blk_run_queue_uncond(struct request_queue *q)
277{
278 if (unlikely(blk_queue_dead(q)))
279 return;
280
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100281 /*
282 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
283 * the queue lock internally. As a result multiple threads may be
284 * running such a request function concurrently. Keep track of the
285 * number of active request_fn invocations such that blk_drain_queue()
286 * can wait until all these request_fn calls have finished.
287 */
288 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100289 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100290 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100291}
292
293/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200294 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200296 *
297 * Description:
298 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200299 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200301void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Tejun Heoa538cd02009-04-23 11:05:17 +0900303 if (unlikely(blk_queue_stopped(q)))
304 return;
305
Bart Van Asschec246e802012-12-06 14:32:01 +0100306 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200307}
308EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200309
Nick Piggin75ad23b2008-04-29 14:48:33 +0200310/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
313 *
314 * Description:
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100316 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200317 */
318void blk_run_queue_async(struct request_queue *q)
319{
Bart Van Assche70460572012-11-28 13:45:56 +0100320 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700321 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200322}
Jens Axboec21e6be2011-04-19 13:32:46 +0200323EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200324
325/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200326 * blk_run_queue - run a single device queue
327 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200328 *
329 * Description:
330 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900331 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200332 */
333void blk_run_queue(struct request_queue *q)
334{
335 unsigned long flags;
336
337 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200338 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 spin_unlock_irqrestore(q->queue_lock, flags);
340}
341EXPORT_SYMBOL(blk_run_queue);
342
Jens Axboe165125e2007-07-24 09:28:11 +0200343void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500344{
345 kobject_put(&q->kobj);
346}
Jens Axboed86e0e82011-05-27 07:44:43 +0200347EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500348
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200349/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100350 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200351 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200352 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200353 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200354 * Drain requests from @q. If @drain_all is set, all requests are drained.
355 * If not, only ELVPRIV requests are drained. The caller is responsible
356 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200357 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100358static void __blk_drain_queue(struct request_queue *q, bool drain_all)
359 __releases(q->queue_lock)
360 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200361{
Asias He458f27a2012-06-15 08:45:25 +0200362 int i;
363
Bart Van Assche807592a2012-11-28 13:43:38 +0100364 lockdep_assert_held(q->queue_lock);
365
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200366 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100367 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200368
Tejun Heob855b042012-03-06 21:24:55 +0100369 /*
370 * The caller might be trying to drain @q before its
371 * elevator is initialized.
372 */
373 if (q->elevator)
374 elv_drain_elevator(q);
375
Tejun Heo5efd6112012-03-05 13:15:12 -0800376 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200377
Tejun Heo4eabc942011-12-15 20:03:04 +0100378 /*
379 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100384 */
Tejun Heob855b042012-03-06 21:24:55 +0100385 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100386 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200387
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700388 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100389 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100390
391 /*
392 * Unfortunately, requests are queued at and tracked from
393 * multiple places and there's no single counter which can
394 * be drained. Check all the queues and counters.
395 */
396 if (drain_all) {
397 drain |= !list_empty(&q->queue_head);
398 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700399 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100400 drain |= q->in_flight[i];
401 drain |= !list_empty(&q->flush_queue[i]);
402 }
403 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404
Tejun Heo481a7d62011-12-14 00:33:37 +0100405 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200406 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100407
408 spin_unlock_irq(q->queue_lock);
409
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200410 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100411
412 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200413 }
Asias He458f27a2012-06-15 08:45:25 +0200414
415 /*
416 * With queue marked dead, any woken up waiter will fail the
417 * allocation path, so the wakeup chaining is lost and we're
418 * left with hung waiters. We need to wake up those waiters.
419 */
420 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700421 struct request_list *rl;
422
Tejun Heoa0516612012-06-26 15:05:44 -0700423 blk_queue_for_each_rl(rl, q)
424 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
425 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200426 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200427}
428
Tejun Heoc9a929d2011-10-19 14:42:16 +0200429/**
Tejun Heod7325802012-03-05 13:14:58 -0800430 * blk_queue_bypass_start - enter queue bypass mode
431 * @q: queue of interest
432 *
433 * In bypass mode, only the dispatch FIFO queue of @q is used. This
434 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800435 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700436 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
437 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800438 */
439void blk_queue_bypass_start(struct request_queue *q)
440{
441 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600442 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800443 queue_flag_set(QUEUE_FLAG_BYPASS, q);
444 spin_unlock_irq(q->queue_lock);
445
Tejun Heo776687b2014-07-01 10:29:17 -0600446 /*
447 * Queues start drained. Skip actual draining till init is
448 * complete. This avoids lenghty delays during queue init which
449 * can happen many times during boot.
450 */
451 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100452 spin_lock_irq(q->queue_lock);
453 __blk_drain_queue(q, false);
454 spin_unlock_irq(q->queue_lock);
455
Tejun Heob82d4b12012-04-13 13:11:31 -0700456 /* ensure blk_queue_bypass() is %true inside RCU read lock */
457 synchronize_rcu();
458 }
Tejun Heod7325802012-03-05 13:14:58 -0800459}
460EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
461
462/**
463 * blk_queue_bypass_end - leave queue bypass mode
464 * @q: queue of interest
465 *
466 * Leave bypass mode and restore the normal queueing behavior.
467 */
468void blk_queue_bypass_end(struct request_queue *q)
469{
470 spin_lock_irq(q->queue_lock);
471 if (!--q->bypass_depth)
472 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
473 WARN_ON_ONCE(q->bypass_depth < 0);
474 spin_unlock_irq(q->queue_lock);
475}
476EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
477
478/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200479 * blk_cleanup_queue - shutdown a request queue
480 * @q: request queue to shutdown
481 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100482 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
483 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500484 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100485void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500486{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200487 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700488
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100489 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500490 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100491 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200492 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800493
Tejun Heo80fd9972012-04-13 14:50:53 -0700494 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100495 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700496 * that, unlike blk_queue_bypass_start(), we aren't performing
497 * synchronize_rcu() after entering bypass mode to avoid the delay
498 * as some drivers create and destroy a lot of queues while
499 * probing. This is still safe because blk_release_queue() will be
500 * called only after the queue refcnt drops to zero and nothing,
501 * RCU or not, would be traversing the queue by then.
502 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800503 q->bypass_depth++;
504 queue_flag_set(QUEUE_FLAG_BYPASS, q);
505
Tejun Heoc9a929d2011-10-19 14:42:16 +0200506 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
507 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100508 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200509 spin_unlock_irq(lock);
510 mutex_unlock(&q->sysfs_lock);
511
Bart Van Asschec246e802012-12-06 14:32:01 +0100512 /*
513 * Drain all requests queued before DYING marking. Set DEAD flag to
514 * prevent that q->request_fn() gets invoked after draining finished.
515 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800516 if (q->mq_ops) {
Tejun Heo780db202014-07-01 10:31:13 -0600517 blk_mq_freeze_queue(q);
Ming Lei43a5e4e2013-12-26 21:31:35 +0800518 spin_lock_irq(lock);
519 } else {
520 spin_lock_irq(lock);
521 __blk_drain_queue(q, true);
522 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100523 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100524 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200525
526 /* @q won't process any more request, flush async actions */
527 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
528 blk_sync_queue(q);
529
Asias He5e5cfac2012-05-24 23:28:52 +0800530 spin_lock_irq(lock);
531 if (q->queue_lock != &q->__queue_lock)
532 q->queue_lock = &q->__queue_lock;
533 spin_unlock_irq(lock);
534
Tejun Heoc9a929d2011-10-19 14:42:16 +0200535 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500536 blk_put_queue(q);
537}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538EXPORT_SYMBOL(blk_cleanup_queue);
539
Tejun Heo5b788ce2012-06-04 20:40:59 -0700540int blk_init_rl(struct request_list *rl, struct request_queue *q,
541 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400543 if (unlikely(rl->rq_pool))
544 return 0;
545
Tejun Heo5b788ce2012-06-04 20:40:59 -0700546 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200547 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
548 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200549 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
550 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Christoph Lameter19460892005-06-23 00:08:19 -0700552 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700553 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700554 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (!rl->rq_pool)
556 return -ENOMEM;
557
558 return 0;
559}
560
Tejun Heo5b788ce2012-06-04 20:40:59 -0700561void blk_exit_rl(struct request_list *rl)
562{
563 if (rl->rq_pool)
564 mempool_destroy(rl->rq_pool);
565}
566
Jens Axboe165125e2007-07-24 09:28:11 +0200567struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100569 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700570}
571EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Jens Axboe165125e2007-07-24 09:28:11 +0200573struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700574{
Jens Axboe165125e2007-07-24 09:28:11 +0200575 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700576 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700577
Jens Axboe8324aa92008-01-29 14:51:59 +0100578 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700579 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (!q)
581 return NULL;
582
Dan Carpenter00380a42012-03-23 09:58:54 +0100583 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100584 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800585 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100586
Jens Axboe0989a022009-06-12 14:42:56 +0200587 q->backing_dev_info.ra_pages =
588 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
589 q->backing_dev_info.state = 0;
590 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200591 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100592 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200593
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700594 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100595 if (err)
596 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700597
Matthew Garrett31373d02010-04-06 14:25:14 +0200598 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
599 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700600 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100601 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700602 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100603 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800604#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800605 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800606#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100607 INIT_LIST_HEAD(&q->flush_queue[0]);
608 INIT_LIST_HEAD(&q->flush_queue[1]);
609 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500610 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500611
Jens Axboe8324aa92008-01-29 14:51:59 +0100612 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Al Viro483f4af2006-03-18 18:34:37 -0500614 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700615 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500616
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500617 /*
618 * By default initialize queue_lock to internal lock and driver can
619 * override it later if need be.
620 */
621 q->queue_lock = &q->__queue_lock;
622
Tejun Heob82d4b12012-04-13 13:11:31 -0700623 /*
624 * A queue starts its life with bypass turned on to avoid
625 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700626 * init. The initial bypass will be finished when the queue is
627 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700628 */
629 q->bypass_depth = 1;
630 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
631
Jens Axboe320ae512013-10-24 09:20:05 +0100632 init_waitqueue_head(&q->mq_freeze_wq);
633
Tejun Heo5efd6112012-03-05 13:15:12 -0800634 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400635 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100638
Mikulas Patockafff49962013-10-14 12:11:36 -0400639fail_bdi:
640 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100641fail_id:
642 ida_simple_remove(&blk_queue_ida, q->id);
643fail_q:
644 kmem_cache_free(blk_requestq_cachep, q);
645 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
Christoph Lameter19460892005-06-23 00:08:19 -0700647EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649/**
650 * blk_init_queue - prepare a request queue for use with a block device
651 * @rfn: The function to be called to process requests that have been
652 * placed on the queue.
653 * @lock: Request queue spin lock
654 *
655 * Description:
656 * If a block device wishes to use the standard request handling procedures,
657 * which sorts requests and coalesces adjacent requests, then it must
658 * call blk_init_queue(). The function @rfn will be called when there
659 * are requests on the queue that need to be processed. If the device
660 * supports plugging, then @rfn may not be called immediately when requests
661 * are available on the queue, but may be called at some time later instead.
662 * Plugged queues are generally unplugged when a buffer belonging to one
663 * of the requests on the queue is needed, or due to memory pressure.
664 *
665 * @rfn is not required, or even expected, to remove all requests off the
666 * queue, but only as many as it can handle at a time. If it does leave
667 * requests on the queue, it is responsible for arranging that the requests
668 * get dealt with eventually.
669 *
670 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200671 * request queue; this lock will be taken also from interrupt context, so irq
672 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200674 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 * it didn't succeed.
676 *
677 * Note:
678 * blk_init_queue() must be paired with a blk_cleanup_queue() call
679 * when the block device is deactivated (such as at module unload).
680 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700681
Jens Axboe165125e2007-07-24 09:28:11 +0200682struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100684 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700685}
686EXPORT_SYMBOL(blk_init_queue);
687
Jens Axboe165125e2007-07-24 09:28:11 +0200688struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700689blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
690{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600691 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600693 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
694 if (!uninit_q)
695 return NULL;
696
Mike Snitzer51514122011-11-23 10:59:13 +0100697 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600698 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700699 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700700
Mike Snitzer7982e902014-03-08 17:20:01 -0700701 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200702}
703EXPORT_SYMBOL(blk_init_queue_node);
704
705struct request_queue *
706blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
707 spinlock_t *lock)
708{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 if (!q)
710 return NULL;
711
Mike Snitzer7982e902014-03-08 17:20:01 -0700712 q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
713 if (!q->flush_rq)
714 return NULL;
715
Tejun Heoa0516612012-06-26 15:05:44 -0700716 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600717 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900721 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700722 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500723
724 /* Override internal queue lock with supplied lock pointer */
725 if (lock)
726 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Jens Axboef3b144a2009-03-06 08:48:33 +0100728 /*
729 * This also sets hw/phys segments, boundary and size
730 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200731 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Alan Stern44ec9542007-02-20 11:01:57 -0500733 q->sg_reserved_size = INT_MAX;
734
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600735 /* Protect q->elevator from elevator_change */
736 mutex_lock(&q->sysfs_lock);
737
Tejun Heob82d4b12012-04-13 13:11:31 -0700738 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600739 if (elevator_init(q, NULL)) {
740 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600741 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600742 }
743
744 mutex_unlock(&q->sysfs_lock);
745
Tejun Heob82d4b12012-04-13 13:11:31 -0700746 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600747
748fail:
749 kfree(q->flush_rq);
750 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
Mike Snitzer51514122011-11-23 10:59:13 +0100752EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Tejun Heo09ac46c2011-12-14 00:33:38 +0100754bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100756 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100757 __blk_get_queue(q);
758 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760
Tejun Heo09ac46c2011-12-14 00:33:38 +0100761 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
Jens Axboed86e0e82011-05-27 07:44:43 +0200763EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Tejun Heo5b788ce2012-06-04 20:40:59 -0700765static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100767 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700768 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100769 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100770 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100771 }
772
Tejun Heo5b788ce2012-06-04 20:40:59 -0700773 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776/*
777 * ioc_batching returns true if the ioc is a valid batching request and
778 * should be given priority access to a request.
779 */
Jens Axboe165125e2007-07-24 09:28:11 +0200780static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
782 if (!ioc)
783 return 0;
784
785 /*
786 * Make sure the process is able to allocate at least 1 request
787 * even if the batch times out, otherwise we could theoretically
788 * lose wakeups.
789 */
790 return ioc->nr_batch_requests == q->nr_batching ||
791 (ioc->nr_batch_requests > 0
792 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
793}
794
795/*
796 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
797 * will cause the process to be a "batcher" on all queues in the system. This
798 * is the behaviour we want though - once it gets a wakeup it should be given
799 * a nice run.
800 */
Jens Axboe165125e2007-07-24 09:28:11 +0200801static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
803 if (!ioc || ioc_batching(q, ioc))
804 return;
805
806 ioc->nr_batch_requests = q->nr_batching;
807 ioc->last_waited = jiffies;
808}
809
Tejun Heo5b788ce2012-06-04 20:40:59 -0700810static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700812 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Tejun Heoa0516612012-06-26 15:05:44 -0700814 /*
815 * bdi isn't aware of blkcg yet. As all async IOs end up root
816 * blkcg anyway, just use root blkcg state.
817 */
818 if (rl == &q->root_rl &&
819 rl->count[sync] < queue_congestion_off_threshold(q))
Jens Axboe1faa16d2009-04-06 14:48:01 +0200820 blk_clear_queue_congested(q, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Jens Axboe1faa16d2009-04-06 14:48:01 +0200822 if (rl->count[sync] + 1 <= q->nr_requests) {
823 if (waitqueue_active(&rl->wait[sync]))
824 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Tejun Heo5b788ce2012-06-04 20:40:59 -0700826 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828}
829
830/*
831 * A request has just been released. Account for it, update the full and
832 * congestion status, wake up any waiters. Called under q->queue_lock.
833 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700834static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700836 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200837 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700839 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200840 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200841 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700842 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Tejun Heo5b788ce2012-06-04 20:40:59 -0700844 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
Jens Axboe1faa16d2009-04-06 14:48:01 +0200846 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700847 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848}
849
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600850int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
851{
852 struct request_list *rl;
853
854 spin_lock_irq(q->queue_lock);
855 q->nr_requests = nr;
856 blk_queue_congestion_threshold(q);
857
858 /* congestion isn't cgroup aware and follows root blkcg for now */
859 rl = &q->root_rl;
860
861 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
862 blk_set_queue_congested(q, BLK_RW_SYNC);
863 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
864 blk_clear_queue_congested(q, BLK_RW_SYNC);
865
866 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
867 blk_set_queue_congested(q, BLK_RW_ASYNC);
868 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
869 blk_clear_queue_congested(q, BLK_RW_ASYNC);
870
871 blk_queue_for_each_rl(rl, q) {
872 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
873 blk_set_rl_full(rl, BLK_RW_SYNC);
874 } else {
875 blk_clear_rl_full(rl, BLK_RW_SYNC);
876 wake_up(&rl->wait[BLK_RW_SYNC]);
877 }
878
879 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
880 blk_set_rl_full(rl, BLK_RW_ASYNC);
881 } else {
882 blk_clear_rl_full(rl, BLK_RW_ASYNC);
883 wake_up(&rl->wait[BLK_RW_ASYNC]);
884 }
885 }
886
887 spin_unlock_irq(q->queue_lock);
888 return 0;
889}
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100892 * Determine if elevator data should be initialized when allocating the
893 * request associated with @bio.
894 */
895static bool blk_rq_should_init_elevator(struct bio *bio)
896{
897 if (!bio)
898 return true;
899
900 /*
901 * Flush requests do not use the elevator so skip initialization.
902 * This allows a request to share the flush and elevator data.
903 */
904 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
905 return false;
906
907 return true;
908}
909
Tejun Heoda8303c2011-10-19 14:33:05 +0200910/**
Tejun Heo852c7882012-03-05 13:15:27 -0800911 * rq_ioc - determine io_context for request allocation
912 * @bio: request being allocated is for this bio (can be %NULL)
913 *
914 * Determine io_context to use for request allocation for @bio. May return
915 * %NULL if %current->io_context doesn't exist.
916 */
917static struct io_context *rq_ioc(struct bio *bio)
918{
919#ifdef CONFIG_BLK_CGROUP
920 if (bio && bio->bi_ioc)
921 return bio->bi_ioc;
922#endif
923 return current->io_context;
924}
925
926/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700927 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700928 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200929 * @rw_flags: RW and SYNC flags
930 * @bio: bio to allocate request for (can be %NULL)
931 * @gfp_mask: allocation mask
932 *
933 * Get a free request from @q. This function may fail under memory
934 * pressure or if @q is dead.
935 *
936 * Must be callled with @q->queue_lock held and,
937 * Returns %NULL on failure, with @q->queue_lock held.
938 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700940static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700941 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700943 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800944 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700945 struct elevator_type *et = q->elevator->type;
946 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100947 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200948 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200949 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100951 if (unlikely(blk_queue_dying(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200952 return NULL;
953
Jens Axboe7749a8d2006-12-13 13:02:26 +0100954 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100955 if (may_queue == ELV_MQUEUE_NO)
956 goto rq_starved;
957
Jens Axboe1faa16d2009-04-06 14:48:01 +0200958 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
959 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100960 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100961 * The queue will fill after this allocation, so set
962 * it as full, and mark this process as "batching".
963 * This process will be allowed to complete a batch of
964 * requests, others will be blocked.
965 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700966 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100967 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700968 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100969 } else {
970 if (may_queue != ELV_MQUEUE_MUST
971 && !ioc_batching(q, ioc)) {
972 /*
973 * The queue is full and the allocating
974 * process is not a "batcher", and not
975 * exempted by the IO scheduler
976 */
Tejun Heob6792812012-03-05 13:15:23 -0800977 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100978 }
979 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 }
Tejun Heoa0516612012-06-26 15:05:44 -0700981 /*
982 * bdi isn't aware of blkcg yet. As all async IOs end up
983 * root blkcg anyway, just use root blkcg state.
984 */
985 if (rl == &q->root_rl)
986 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988
Jens Axboe082cf692005-06-28 16:35:11 +0200989 /*
990 * Only allow batching queuers to allocate up to 50% over the defined
991 * limit of requests, otherwise we could have thousands of requests
992 * allocated with any setting of ->nr_requests
993 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200994 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800995 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100996
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700997 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200998 rl->count[is_sync]++;
999 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001000
Tejun Heof1f8cc92011-12-14 00:33:42 +01001001 /*
1002 * Decide whether the new request will be managed by elevator. If
1003 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1004 * prevent the current elevator from being destroyed until the new
1005 * request is freed. This guarantees icq's won't be destroyed and
1006 * makes creating new ones safe.
1007 *
1008 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1009 * it will be created after releasing queue_lock.
1010 */
Tejun Heod7325802012-03-05 13:14:58 -08001011 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001012 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001013 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001014 if (et->icq_cache && ioc)
1015 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001016 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001017
Jens Axboef253b862010-10-24 22:06:02 +02001018 if (blk_queue_io_stat(q))
1019 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 spin_unlock_irq(q->queue_lock);
1021
Tejun Heo29e2b092012-04-19 16:29:21 -07001022 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001023 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001024 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001025 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Tejun Heo29e2b092012-04-19 16:29:21 -07001027 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001028 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001029 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1030
Tejun Heoaaf7c682012-04-19 16:29:22 -07001031 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001032 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001033 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001034 if (ioc)
1035 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001036 if (!icq)
1037 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001038 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001039
1040 rq->elv.icq = icq;
1041 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1042 goto fail_elvpriv;
1043
1044 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001045 if (icq)
1046 get_io_context(icq->ioc);
1047 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001048out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001049 /*
1050 * ioc may be NULL here, and ioc_batching will be false. That's
1051 * OK, if the queue is under the request limit then requests need
1052 * not count toward the nr_batch_requests limit. There will always
1053 * be some limit enforced by BLK_BATCH_TIME.
1054 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (ioc_batching(q, ioc))
1056 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001057
Jens Axboe1faa16d2009-04-06 14:48:01 +02001058 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001060
Tejun Heoaaf7c682012-04-19 16:29:22 -07001061fail_elvpriv:
1062 /*
1063 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1064 * and may fail indefinitely under memory pressure and thus
1065 * shouldn't stall IO. Treat this request as !elvpriv. This will
1066 * disturb iosched and blkcg but weird is bettern than dead.
1067 */
1068 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1069 dev_name(q->backing_dev_info.dev));
1070
1071 rq->cmd_flags &= ~REQ_ELVPRIV;
1072 rq->elv.icq = NULL;
1073
1074 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001075 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001076 spin_unlock_irq(q->queue_lock);
1077 goto out;
1078
Tejun Heob6792812012-03-05 13:15:23 -08001079fail_alloc:
1080 /*
1081 * Allocation failed presumably due to memory. Undo anything we
1082 * might have messed up.
1083 *
1084 * Allocating task should really be put onto the front of the wait
1085 * queue, but this is pretty rare.
1086 */
1087 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001088 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001089
1090 /*
1091 * in the very unlikely event that allocation failed and no
1092 * requests for this direction was pending, mark us starved so that
1093 * freeing of a request in the other direction will notice
1094 * us. another possible fix would be to split the rq mempool into
1095 * READ and WRITE
1096 */
1097rq_starved:
1098 if (unlikely(rl->count[is_sync] == 0))
1099 rl->starved[is_sync] = 1;
1100 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101}
1102
Tejun Heoda8303c2011-10-19 14:33:05 +02001103/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001104 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001105 * @q: request_queue to allocate request from
1106 * @rw_flags: RW and SYNC flags
1107 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001108 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001109 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001110 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1111 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001112 *
1113 * Must be callled with @q->queue_lock held and,
1114 * Returns %NULL on failure, with @q->queue_lock held.
1115 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001117static struct request *get_request(struct request_queue *q, int rw_flags,
1118 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001120 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001121 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001122 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001124
1125 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001126retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001127 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001128 if (rq)
1129 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001131 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001132 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001133 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Tejun Heoa06e05e2012-06-04 20:40:55 -07001136 /* wait on @rl and retry */
1137 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1138 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001139
Tejun Heoa06e05e2012-06-04 20:40:55 -07001140 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Tejun Heoa06e05e2012-06-04 20:40:55 -07001142 spin_unlock_irq(q->queue_lock);
1143 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Tejun Heoa06e05e2012-06-04 20:40:55 -07001145 /*
1146 * After sleeping, we become a "batching" process and will be able
1147 * to allocate at least one request, and up to a big batch of them
1148 * for a small period time. See ioc_batching, ioc_set_batching
1149 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001150 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Tejun Heoa06e05e2012-06-04 20:40:55 -07001152 spin_lock_irq(q->queue_lock);
1153 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001154
Tejun Heoa06e05e2012-06-04 20:40:55 -07001155 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156}
1157
Jens Axboe320ae512013-10-24 09:20:05 +01001158static struct request *blk_old_get_request(struct request_queue *q, int rw,
1159 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
1161 struct request *rq;
1162
1163 BUG_ON(rw != READ && rw != WRITE);
1164
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001165 /* create ioc upfront */
1166 create_io_context(gfp_mask, q->node);
1167
Nick Piggind6344532005-06-28 20:45:14 -07001168 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001169 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001170 if (!rq)
1171 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001172 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 return rq;
1175}
Jens Axboe320ae512013-10-24 09:20:05 +01001176
1177struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1178{
1179 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001180 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001181 else
1182 return blk_old_get_request(q, rw, gfp_mask);
1183}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184EXPORT_SYMBOL(blk_get_request);
1185
1186/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001187 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001188 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001189 * @bio: The bio describing the memory mappings that will be submitted for IO.
1190 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001191 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001192 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001193 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1194 * type commands. Where the struct request needs to be farther initialized by
1195 * the caller. It is passed a &struct bio, which describes the memory info of
1196 * the I/O transfer.
1197 *
1198 * The caller of blk_make_request must make sure that bi_io_vec
1199 * are set to describe the memory buffers. That bio_data_dir() will return
1200 * the needed direction of the request. (And all bio's in the passed bio-chain
1201 * are properly set accordingly)
1202 *
1203 * If called under none-sleepable conditions, mapped bio buffers must not
1204 * need bouncing, by calling the appropriate masked or flagged allocator,
1205 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1206 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001207 *
1208 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1209 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1210 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1211 * completion of a bio that hasn't been submitted yet, thus resulting in a
1212 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1213 * of bio_alloc(), as that avoids the mempool deadlock.
1214 * If possible a big IO should be split into smaller parts when allocation
1215 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001216 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001217struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1218 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001219{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001220 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1221
1222 if (unlikely(!rq))
1223 return ERR_PTR(-ENOMEM);
1224
Jens Axboef27b0872014-06-06 07:57:37 -06001225 blk_rq_set_block_pc(rq);
1226
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001227 for_each_bio(bio) {
1228 struct bio *bounce_bio = bio;
1229 int ret;
1230
1231 blk_queue_bounce(q, &bounce_bio);
1232 ret = blk_rq_append_bio(q, rq, bounce_bio);
1233 if (unlikely(ret)) {
1234 blk_put_request(rq);
1235 return ERR_PTR(ret);
1236 }
1237 }
1238
1239 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001240}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001241EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001242
1243/**
Jens Axboef27b0872014-06-06 07:57:37 -06001244 * blk_rq_set_block_pc - initialize a requeest to type BLOCK_PC
1245 * @rq: request to be initialized
1246 *
1247 */
1248void blk_rq_set_block_pc(struct request *rq)
1249{
1250 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1251 rq->__data_len = 0;
1252 rq->__sector = (sector_t) -1;
1253 rq->bio = rq->biotail = NULL;
1254 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1255 rq->cmd = rq->__cmd;
1256}
1257EXPORT_SYMBOL(blk_rq_set_block_pc);
1258
1259/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 * blk_requeue_request - put a request back on queue
1261 * @q: request queue where request should be inserted
1262 * @rq: request to be inserted
1263 *
1264 * Description:
1265 * Drivers often keep queueing requests until the hardware cannot accept
1266 * more, when that condition happens we need to put the request back
1267 * on the queue. Must be called with queue lock held.
1268 */
Jens Axboe165125e2007-07-24 09:28:11 +02001269void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001271 blk_delete_timer(rq);
1272 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001273 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 if (blk_rq_tagged(rq))
1276 blk_queue_end_tag(q, rq);
1277
James Bottomleyba396a62009-05-27 14:17:08 +02001278 BUG_ON(blk_queued_rq(rq));
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 elv_requeue_request(q, rq);
1281}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282EXPORT_SYMBOL(blk_requeue_request);
1283
Jens Axboe73c10102011-03-08 13:19:51 +01001284static void add_acct_request(struct request_queue *q, struct request *rq,
1285 int where)
1286{
Jens Axboe320ae512013-10-24 09:20:05 +01001287 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001288 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001289}
1290
Tejun Heo074a7ac2008-08-25 19:56:14 +09001291static void part_round_stats_single(int cpu, struct hd_struct *part,
1292 unsigned long now)
1293{
Jens Axboe7276d022014-05-09 15:48:23 -06001294 int inflight;
1295
Tejun Heo074a7ac2008-08-25 19:56:14 +09001296 if (now == part->stamp)
1297 return;
1298
Jens Axboe7276d022014-05-09 15:48:23 -06001299 inflight = part_in_flight(part);
1300 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001301 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001302 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001303 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1304 }
1305 part->stamp = now;
1306}
1307
1308/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001309 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1310 * @cpu: cpu number for stats access
1311 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 *
1313 * The average IO queue length and utilisation statistics are maintained
1314 * by observing the current state of the queue length and the amount of
1315 * time it has been in this state for.
1316 *
1317 * Normally, that accounting is done on IO completion, but that can result
1318 * in more than a second's worth of IO being accounted for within any one
1319 * second, leading to >100% utilisation. To deal with that, we call this
1320 * function to do a round-off before returning the results when reading
1321 * /proc/diskstats. This accounts immediately for all queue usage up to
1322 * the current jiffies and restarts the counters again.
1323 */
Tejun Heoc9959052008-08-25 19:47:21 +09001324void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001325{
1326 unsigned long now = jiffies;
1327
Tejun Heo074a7ac2008-08-25 19:56:14 +09001328 if (part->partno)
1329 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1330 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001331}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001332EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001333
Lin Mingc8158812013-03-23 11:42:27 +08001334#ifdef CONFIG_PM_RUNTIME
1335static void blk_pm_put_request(struct request *rq)
1336{
1337 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1338 pm_runtime_mark_last_busy(rq->q->dev);
1339}
1340#else
1341static inline void blk_pm_put_request(struct request *rq) {}
1342#endif
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344/*
1345 * queue lock must be held
1346 */
Jens Axboe165125e2007-07-24 09:28:11 +02001347void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 if (unlikely(!q))
1350 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001352 if (q->mq_ops) {
1353 blk_mq_free_request(req);
1354 return;
1355 }
1356
Lin Mingc8158812013-03-23 11:42:27 +08001357 blk_pm_put_request(req);
1358
Tejun Heo8922e162005-10-20 16:23:44 +02001359 elv_completed_request(q, req);
1360
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001361 /* this is a bio leak */
1362 WARN_ON(req->bio != NULL);
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 /*
1365 * Request may not have originated from ll_rw_blk. if not,
1366 * it didn't come out of our reserved rq pools
1367 */
Jens Axboe49171e52006-08-10 08:59:11 +02001368 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001369 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001370 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001373 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Tejun Heoa0516612012-06-26 15:05:44 -07001375 blk_free_request(rl, req);
1376 freed_request(rl, flags);
1377 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
1379}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001380EXPORT_SYMBOL_GPL(__blk_put_request);
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382void blk_put_request(struct request *req)
1383{
Jens Axboe165125e2007-07-24 09:28:11 +02001384 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Jens Axboe320ae512013-10-24 09:20:05 +01001386 if (q->mq_ops)
1387 blk_mq_free_request(req);
1388 else {
1389 unsigned long flags;
1390
1391 spin_lock_irqsave(q->queue_lock, flags);
1392 __blk_put_request(q, req);
1393 spin_unlock_irqrestore(q->queue_lock, flags);
1394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396EXPORT_SYMBOL(blk_put_request);
1397
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001398/**
1399 * blk_add_request_payload - add a payload to a request
1400 * @rq: request to update
1401 * @page: page backing the payload
1402 * @len: length of the payload.
1403 *
1404 * This allows to later add a payload to an already submitted request by
1405 * a block driver. The driver needs to take care of freeing the payload
1406 * itself.
1407 *
1408 * Note that this is a quite horrible hack and nothing but handling of
1409 * discard requests should ever use it.
1410 */
1411void blk_add_request_payload(struct request *rq, struct page *page,
1412 unsigned int len)
1413{
1414 struct bio *bio = rq->bio;
1415
1416 bio->bi_io_vec->bv_page = page;
1417 bio->bi_io_vec->bv_offset = 0;
1418 bio->bi_io_vec->bv_len = len;
1419
Kent Overstreet4f024f32013-10-11 15:44:27 -07001420 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001421 bio->bi_vcnt = 1;
1422 bio->bi_phys_segments = 1;
1423
1424 rq->__data_len = rq->resid_len = len;
1425 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001426}
1427EXPORT_SYMBOL_GPL(blk_add_request_payload);
1428
Jens Axboe320ae512013-10-24 09:20:05 +01001429bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1430 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001431{
1432 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1433
Jens Axboe73c10102011-03-08 13:19:51 +01001434 if (!ll_back_merge_fn(q, req, bio))
1435 return false;
1436
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001437 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001438
1439 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1440 blk_rq_set_mixed_merge(req);
1441
1442 req->biotail->bi_next = bio;
1443 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001444 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001445 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1446
Jens Axboe320ae512013-10-24 09:20:05 +01001447 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001448 return true;
1449}
1450
Jens Axboe320ae512013-10-24 09:20:05 +01001451bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1452 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001453{
1454 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001455
Jens Axboe73c10102011-03-08 13:19:51 +01001456 if (!ll_front_merge_fn(q, req, bio))
1457 return false;
1458
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001459 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001460
1461 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1462 blk_rq_set_mixed_merge(req);
1463
Jens Axboe73c10102011-03-08 13:19:51 +01001464 bio->bi_next = req->bio;
1465 req->bio = bio;
1466
Kent Overstreet4f024f32013-10-11 15:44:27 -07001467 req->__sector = bio->bi_iter.bi_sector;
1468 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001469 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1470
Jens Axboe320ae512013-10-24 09:20:05 +01001471 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001472 return true;
1473}
1474
Tejun Heobd87b582011-10-19 14:33:08 +02001475/**
Jens Axboe320ae512013-10-24 09:20:05 +01001476 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001477 * @q: request_queue new bio is being queued at
1478 * @bio: new bio being queued
1479 * @request_count: out parameter for number of traversed plugged requests
1480 *
1481 * Determine whether @bio being queued on @q can be merged with a request
1482 * on %current's plugged list. Returns %true if merge was successful,
1483 * otherwise %false.
1484 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001485 * Plugging coalesces IOs from the same issuer for the same purpose without
1486 * going through @q->queue_lock. As such it's more of an issuing mechanism
1487 * than scheduling, and the request, while may have elvpriv data, is not
1488 * added on the elevator at this point. In addition, we don't have
1489 * reliable access to the elevator outside queue lock. Only check basic
1490 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001491 *
1492 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001493 */
Jens Axboe320ae512013-10-24 09:20:05 +01001494bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1495 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001496{
1497 struct blk_plug *plug;
1498 struct request *rq;
1499 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001500 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001501
Tejun Heobd87b582011-10-19 14:33:08 +02001502 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001503 if (!plug)
1504 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001505 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001506
Shaohua Li92f399c2013-10-29 12:01:03 -06001507 if (q->mq_ops)
1508 plug_list = &plug->mq_list;
1509 else
1510 plug_list = &plug->list;
1511
1512 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001513 int el_ret;
1514
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001515 if (rq->q == q)
1516 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001517
Tejun Heo07c2bd32012-02-08 09:19:42 +01001518 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001519 continue;
1520
Tejun Heo050c8ea2012-02-08 09:19:38 +01001521 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001522 if (el_ret == ELEVATOR_BACK_MERGE) {
1523 ret = bio_attempt_back_merge(q, rq, bio);
1524 if (ret)
1525 break;
1526 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1527 ret = bio_attempt_front_merge(q, rq, bio);
1528 if (ret)
1529 break;
1530 }
1531 }
1532out:
1533 return ret;
1534}
1535
Jens Axboe86db1e22008-01-29 14:53:40 +01001536void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001537{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001538 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001539
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001540 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1541 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001542 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001543
Tejun Heo52d9e672006-01-06 09:49:58 +01001544 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001545 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001546 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001547 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001548}
1549
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001550void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001552 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001553 struct blk_plug *plug;
1554 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1555 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001556 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /*
1559 * low level driver can indicate that it wants pages above a
1560 * certain limit bounced to low memory (ie for highmem, or even
1561 * ISA dma in theory)
1562 */
1563 blk_queue_bounce(q, &bio);
1564
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001565 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1566 bio_endio(bio, -EIO);
1567 return;
1568 }
1569
Tejun Heo4fed9472010-09-03 11:56:17 +02001570 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001571 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001572 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001573 goto get_rq;
1574 }
1575
Jens Axboe73c10102011-03-08 13:19:51 +01001576 /*
1577 * Check if we can merge with the plugged list before grabbing
1578 * any locks.
1579 */
Robert Elliottda41a582014-05-20 16:46:26 -05001580 if (!blk_queue_nomerges(q) &&
1581 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001582 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001583
1584 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001587 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001588 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001589 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001590 if (!attempt_back_merge(q, req))
1591 elv_merged_request(q, req, el_ret);
1592 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001593 }
Jens Axboe73c10102011-03-08 13:19:51 +01001594 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001595 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001596 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001597 if (!attempt_front_merge(q, req))
1598 elv_merged_request(q, req, el_ret);
1599 goto out_unlock;
1600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 }
1602
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001604 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001605 * This sync check and mask will be re-done in init_request_from_bio(),
1606 * but we need to set it earlier to expose the sync flag to the
1607 * rq allocator and io schedulers.
1608 */
1609 rw_flags = bio_data_dir(bio);
1610 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001611 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001612
1613 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001614 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001615 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001616 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001617 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001618 if (unlikely(!req)) {
1619 bio_endio(bio, -ENODEV); /* @q is dead */
1620 goto out_unlock;
1621 }
Nick Piggind6344532005-06-28 20:45:14 -07001622
Nick Piggin450991b2005-06-28 20:45:13 -07001623 /*
1624 * After dropping the lock and possibly sleeping here, our request
1625 * may now be mergeable after it had proven unmergeable (above).
1626 * We don't worry about that case for efficiency. It won't happen
1627 * often, and the elevators are able to handle it.
1628 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001629 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Tao Ma9562ad92011-10-24 16:11:30 +02001631 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001632 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001633
Jens Axboe73c10102011-03-08 13:19:51 +01001634 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001635 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001636 /*
1637 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001638 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001639 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001640 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001641 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001642 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001643 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001644 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001645 trace_block_plug(q);
1646 }
Jens Axboe73c10102011-03-08 13:19:51 +01001647 }
Shaohua Lia6327162011-08-24 16:04:32 +02001648 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001649 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001650 } else {
1651 spin_lock_irq(q->queue_lock);
1652 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001653 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001654out_unlock:
1655 spin_unlock_irq(q->queue_lock);
1656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657}
Jens Axboec20e8de2011-09-12 12:03:37 +02001658EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
1660/*
1661 * If bio->bi_dev is a partition, remap the location
1662 */
1663static inline void blk_partition_remap(struct bio *bio)
1664{
1665 struct block_device *bdev = bio->bi_bdev;
1666
Jens Axboebf2de6f2007-09-27 13:01:25 +02001667 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001669
Kent Overstreet4f024f32013-10-11 15:44:27 -07001670 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001672
Mike Snitzerd07335e2010-11-16 12:52:38 +01001673 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1674 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001675 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 }
1677}
1678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679static void handle_bad_sector(struct bio *bio)
1680{
1681 char b[BDEVNAME_SIZE];
1682
1683 printk(KERN_INFO "attempt to access beyond end of device\n");
1684 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1685 bdevname(bio->bi_bdev, b),
1686 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001687 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001688 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
1690 set_bit(BIO_EOF, &bio->bi_flags);
1691}
1692
Akinobu Mitac17bb492006-12-08 02:39:46 -08001693#ifdef CONFIG_FAIL_MAKE_REQUEST
1694
1695static DECLARE_FAULT_ATTR(fail_make_request);
1696
1697static int __init setup_fail_make_request(char *str)
1698{
1699 return setup_fault_attr(&fail_make_request, str);
1700}
1701__setup("fail_make_request=", setup_fail_make_request);
1702
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001703static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001704{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001705 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001706}
1707
1708static int __init fail_make_request_debugfs(void)
1709{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001710 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1711 NULL, &fail_make_request);
1712
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001713 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001714}
1715
1716late_initcall(fail_make_request_debugfs);
1717
1718#else /* CONFIG_FAIL_MAKE_REQUEST */
1719
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001720static inline bool should_fail_request(struct hd_struct *part,
1721 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001722{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001723 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001724}
1725
1726#endif /* CONFIG_FAIL_MAKE_REQUEST */
1727
Jens Axboec07e2b42007-07-18 13:27:58 +02001728/*
1729 * Check whether this bio extends beyond the end of the device.
1730 */
1731static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1732{
1733 sector_t maxsector;
1734
1735 if (!nr_sectors)
1736 return 0;
1737
1738 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001739 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001740 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001741 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001742
1743 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1744 /*
1745 * This may well happen - the kernel calls bread()
1746 * without checking the size of the device, e.g., when
1747 * mounting a device.
1748 */
1749 handle_bad_sector(bio);
1750 return 1;
1751 }
1752 }
1753
1754 return 0;
1755}
1756
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001757static noinline_for_stack bool
1758generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759{
Jens Axboe165125e2007-07-24 09:28:11 +02001760 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001761 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001762 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001763 char b[BDEVNAME_SIZE];
1764 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
1766 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
Jens Axboec07e2b42007-07-18 13:27:58 +02001768 if (bio_check_eod(bio, nr_sectors))
1769 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001771 q = bdev_get_queue(bio->bi_bdev);
1772 if (unlikely(!q)) {
1773 printk(KERN_ERR
1774 "generic_make_request: Trying to access "
1775 "nonexistent block-device %s (%Lu)\n",
1776 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001777 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001778 goto end_io;
1779 }
1780
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001781 if (likely(bio_is_rw(bio) &&
1782 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001783 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1784 bdevname(bio->bi_bdev, b),
1785 bio_sectors(bio),
1786 queue_max_hw_sectors(q));
1787 goto end_io;
1788 }
1789
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001790 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001791 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001792 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001793 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001794 goto end_io;
1795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001797 * If this device has partitions, remap block n
1798 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001800 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001802 if (bio_check_eod(bio, nr_sectors))
1803 goto end_io;
1804
1805 /*
1806 * Filter flush bio's early so that make_request based
1807 * drivers without flush support don't have to worry
1808 * about them.
1809 */
1810 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1811 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1812 if (!nr_sectors) {
1813 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001814 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001818 if ((bio->bi_rw & REQ_DISCARD) &&
1819 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001820 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001821 err = -EOPNOTSUPP;
1822 goto end_io;
1823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001825 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 err = -EOPNOTSUPP;
1827 goto end_io;
1828 }
1829
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001830 /*
1831 * Various block parts want %current->io_context and lazy ioc
1832 * allocation ends up trading a lot of pain for a small amount of
1833 * memory. Just allocate it upfront. This may fail and block
1834 * layer knows how to live with it.
1835 */
1836 create_io_context(GFP_ATOMIC, q->node);
1837
Tejun Heobc16a4f2011-10-19 14:33:01 +02001838 if (blk_throtl_bio(q, bio))
1839 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001840
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001841 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001842 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001843
1844end_io:
1845 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001846 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847}
1848
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001849/**
1850 * generic_make_request - hand a buffer to its device driver for I/O
1851 * @bio: The bio describing the location in memory and on the device.
1852 *
1853 * generic_make_request() is used to make I/O requests of block
1854 * devices. It is passed a &struct bio, which describes the I/O that needs
1855 * to be done.
1856 *
1857 * generic_make_request() does not return any status. The
1858 * success/failure status of the request, along with notification of
1859 * completion, is delivered asynchronously through the bio->bi_end_io
1860 * function described (one day) else where.
1861 *
1862 * The caller of generic_make_request must make sure that bi_io_vec
1863 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1864 * set to describe the device address, and the
1865 * bi_end_io and optionally bi_private are set to describe how
1866 * completion notification should be signaled.
1867 *
1868 * generic_make_request and the drivers it calls may use bi_next if this
1869 * bio happens to be merged with someone else, and may resubmit the bio to
1870 * a lower device by calling into generic_make_request recursively, which
1871 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001872 */
1873void generic_make_request(struct bio *bio)
1874{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001875 struct bio_list bio_list_on_stack;
1876
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001877 if (!generic_make_request_checks(bio))
1878 return;
1879
1880 /*
1881 * We only want one ->make_request_fn to be active at a time, else
1882 * stack usage with stacked devices could be a problem. So use
1883 * current->bio_list to keep a list of requests submited by a
1884 * make_request_fn function. current->bio_list is also used as a
1885 * flag to say if generic_make_request is currently active in this
1886 * task or not. If it is NULL, then no make_request is active. If
1887 * it is non-NULL, then a make_request is active, and new requests
1888 * should be added at the tail
1889 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001890 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001891 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001892 return;
1893 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001894
Neil Brownd89d8792007-05-01 09:53:42 +02001895 /* following loop may be a bit non-obvious, and so deserves some
1896 * explanation.
1897 * Before entering the loop, bio->bi_next is NULL (as all callers
1898 * ensure that) so we have a list with a single bio.
1899 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001900 * we assign bio_list to a pointer to the bio_list_on_stack,
1901 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001902 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001903 * through a recursive call to generic_make_request. If it
1904 * did, we find a non-NULL value in bio_list and re-enter the loop
1905 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001906 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001907 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001908 */
1909 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001910 bio_list_init(&bio_list_on_stack);
1911 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001912 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001913 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1914
1915 q->make_request_fn(q, bio);
1916
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001917 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001918 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001919 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001920}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921EXPORT_SYMBOL(generic_make_request);
1922
1923/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001924 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1926 * @bio: The &struct bio which describes the I/O
1927 *
1928 * submit_bio() is very similar in purpose to generic_make_request(), and
1929 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001930 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 *
1932 */
1933void submit_bio(int rw, struct bio *bio)
1934{
Jens Axboe22e2c502005-06-27 10:55:12 +02001935 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Jens Axboebf2de6f2007-09-27 13:01:25 +02001937 /*
1938 * If it's a regular read/write or a barrier with data attached,
1939 * go through the normal accounting stuff before submission.
1940 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001941 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001942 unsigned int count;
1943
1944 if (unlikely(rw & REQ_WRITE_SAME))
1945 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1946 else
1947 count = bio_sectors(bio);
1948
Jens Axboebf2de6f2007-09-27 13:01:25 +02001949 if (rw & WRITE) {
1950 count_vm_events(PGPGOUT, count);
1951 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001952 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001953 count_vm_events(PGPGIN, count);
1954 }
1955
1956 if (unlikely(block_dump)) {
1957 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001958 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001959 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001960 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001961 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001962 bdevname(bio->bi_bdev, b),
1963 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 }
1966
1967 generic_make_request(bio);
1968}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969EXPORT_SYMBOL(submit_bio);
1970
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001971/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001972 * blk_rq_check_limits - Helper function to check a request for the queue limit
1973 * @q: the queue
1974 * @rq: the request being checked
1975 *
1976 * Description:
1977 * @rq may have been made based on weaker limitations of upper-level queues
1978 * in request stacking drivers, and it may violate the limitation of @q.
1979 * Since the block layer and the underlying device driver trust @rq
1980 * after it is inserted to @q, it should be checked against @q before
1981 * the insertion using this generic function.
1982 *
1983 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001984 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001985 * Request stacking drivers like request-based dm may change the queue
1986 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09001987 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001988 * the new queue limits again when they dispatch those requests,
1989 * although such checkings are also done against the old queue limits
1990 * when submitting requests.
1991 */
1992int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1993{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001994 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001995 return 0;
1996
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001997 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001998 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1999 return -EIO;
2000 }
2001
2002 /*
2003 * queue's settings related to segment counting like q->bounce_pfn
2004 * may differ from that of other stacking queues.
2005 * Recalculate it to check the request correctly on this queue's
2006 * limitation.
2007 */
2008 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002009 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002010 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2011 return -EIO;
2012 }
2013
2014 return 0;
2015}
2016EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2017
2018/**
2019 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2020 * @q: the queue to submit the request
2021 * @rq: the request being queued
2022 */
2023int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2024{
2025 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002026 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002027
2028 if (blk_rq_check_limits(q, rq))
2029 return -EIO;
2030
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002031 if (rq->rq_disk &&
2032 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002033 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002034
2035 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002036 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002037 spin_unlock_irqrestore(q->queue_lock, flags);
2038 return -ENODEV;
2039 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002040
2041 /*
2042 * Submitting request must be dequeued before calling this function
2043 * because it will be linked to another request_queue
2044 */
2045 BUG_ON(blk_queued_rq(rq));
2046
Jeff Moyer4853aba2011-08-15 21:37:25 +02002047 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2048 where = ELEVATOR_INSERT_FLUSH;
2049
2050 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002051 if (where == ELEVATOR_INSERT_FLUSH)
2052 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002053 spin_unlock_irqrestore(q->queue_lock, flags);
2054
2055 return 0;
2056}
2057EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2058
Tejun Heo80a761f2009-07-03 17:48:17 +09002059/**
2060 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2061 * @rq: request to examine
2062 *
2063 * Description:
2064 * A request could be merge of IOs which require different failure
2065 * handling. This function determines the number of bytes which
2066 * can be failed from the beginning of the request without
2067 * crossing into area which need to be retried further.
2068 *
2069 * Return:
2070 * The number of bytes to fail.
2071 *
2072 * Context:
2073 * queue_lock must be held.
2074 */
2075unsigned int blk_rq_err_bytes(const struct request *rq)
2076{
2077 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2078 unsigned int bytes = 0;
2079 struct bio *bio;
2080
2081 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2082 return blk_rq_bytes(rq);
2083
2084 /*
2085 * Currently the only 'mixing' which can happen is between
2086 * different fastfail types. We can safely fail portions
2087 * which have all the failfast bits that the first one has -
2088 * the ones which are at least as eager to fail as the first
2089 * one.
2090 */
2091 for (bio = rq->bio; bio; bio = bio->bi_next) {
2092 if ((bio->bi_rw & ff) != ff)
2093 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002094 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002095 }
2096
2097 /* this could lead to infinite loop */
2098 BUG_ON(blk_rq_bytes(rq) && !bytes);
2099 return bytes;
2100}
2101EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2102
Jens Axboe320ae512013-10-24 09:20:05 +01002103void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002104{
Jens Axboec2553b52009-04-24 08:10:11 +02002105 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002106 const int rw = rq_data_dir(req);
2107 struct hd_struct *part;
2108 int cpu;
2109
2110 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002111 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002112 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2113 part_stat_unlock();
2114 }
2115}
2116
Jens Axboe320ae512013-10-24 09:20:05 +01002117void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002118{
Jens Axboebc58ba92009-01-23 10:54:44 +01002119 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002120 * Account IO completion. flush_rq isn't accounted as a
2121 * normal IO on queueing nor completion. Accounting the
2122 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002123 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002124 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002125 unsigned long duration = jiffies - req->start_time;
2126 const int rw = rq_data_dir(req);
2127 struct hd_struct *part;
2128 int cpu;
2129
2130 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002131 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002132
2133 part_stat_inc(cpu, part, ios[rw]);
2134 part_stat_add(cpu, part, ticks[rw], duration);
2135 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002136 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002137
Jens Axboe6c23a962011-01-07 08:43:37 +01002138 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002139 part_stat_unlock();
2140 }
2141}
2142
Lin Mingc8158812013-03-23 11:42:27 +08002143#ifdef CONFIG_PM_RUNTIME
2144/*
2145 * Don't process normal requests when queue is suspended
2146 * or in the process of suspending/resuming
2147 */
2148static struct request *blk_pm_peek_request(struct request_queue *q,
2149 struct request *rq)
2150{
2151 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2152 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2153 return NULL;
2154 else
2155 return rq;
2156}
2157#else
2158static inline struct request *blk_pm_peek_request(struct request_queue *q,
2159 struct request *rq)
2160{
2161 return rq;
2162}
2163#endif
2164
Jens Axboe320ae512013-10-24 09:20:05 +01002165void blk_account_io_start(struct request *rq, bool new_io)
2166{
2167 struct hd_struct *part;
2168 int rw = rq_data_dir(rq);
2169 int cpu;
2170
2171 if (!blk_do_io_stat(rq))
2172 return;
2173
2174 cpu = part_stat_lock();
2175
2176 if (!new_io) {
2177 part = rq->part;
2178 part_stat_inc(cpu, part, merges[rw]);
2179 } else {
2180 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2181 if (!hd_struct_try_get(part)) {
2182 /*
2183 * The partition is already being removed,
2184 * the request will be accounted on the disk only
2185 *
2186 * We take a reference on disk->part0 although that
2187 * partition will never be deleted, so we can treat
2188 * it as any other partition.
2189 */
2190 part = &rq->rq_disk->part0;
2191 hd_struct_get(part);
2192 }
2193 part_round_stats(cpu, part);
2194 part_inc_in_flight(part, rw);
2195 rq->part = part;
2196 }
2197
2198 part_stat_unlock();
2199}
2200
Tejun Heo53a08802008-12-03 12:41:26 +01002201/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002202 * blk_peek_request - peek at the top of a request queue
2203 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002204 *
2205 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002206 * Return the request at the top of @q. The returned request
2207 * should be started using blk_start_request() before LLD starts
2208 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002209 *
2210 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002211 * Pointer to the request at the top of @q if available. Null
2212 * otherwise.
2213 *
2214 * Context:
2215 * queue_lock must be held.
2216 */
2217struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002218{
2219 struct request *rq;
2220 int ret;
2221
2222 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002223
2224 rq = blk_pm_peek_request(q, rq);
2225 if (!rq)
2226 break;
2227
Tejun Heo158dbda2009-04-23 11:05:18 +09002228 if (!(rq->cmd_flags & REQ_STARTED)) {
2229 /*
2230 * This is the first time the device driver
2231 * sees this request (possibly after
2232 * requeueing). Notify IO scheduler.
2233 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002234 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002235 elv_activate_rq(q, rq);
2236
2237 /*
2238 * just mark as started even if we don't start
2239 * it, a request that has been delayed should
2240 * not be passed by new incoming requests
2241 */
2242 rq->cmd_flags |= REQ_STARTED;
2243 trace_block_rq_issue(q, rq);
2244 }
2245
2246 if (!q->boundary_rq || q->boundary_rq == rq) {
2247 q->end_sector = rq_end_sector(rq);
2248 q->boundary_rq = NULL;
2249 }
2250
2251 if (rq->cmd_flags & REQ_DONTPREP)
2252 break;
2253
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002254 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002255 /*
2256 * make sure space for the drain appears we
2257 * know we can do this because max_hw_segments
2258 * has been adjusted to be one fewer than the
2259 * device can handle
2260 */
2261 rq->nr_phys_segments++;
2262 }
2263
2264 if (!q->prep_rq_fn)
2265 break;
2266
2267 ret = q->prep_rq_fn(q, rq);
2268 if (ret == BLKPREP_OK) {
2269 break;
2270 } else if (ret == BLKPREP_DEFER) {
2271 /*
2272 * the request may have been (partially) prepped.
2273 * we need to keep this request in the front to
2274 * avoid resource deadlock. REQ_STARTED will
2275 * prevent other fs requests from passing this one.
2276 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002277 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002278 !(rq->cmd_flags & REQ_DONTPREP)) {
2279 /*
2280 * remove the space for the drain we added
2281 * so that we don't add it again
2282 */
2283 --rq->nr_phys_segments;
2284 }
2285
2286 rq = NULL;
2287 break;
2288 } else if (ret == BLKPREP_KILL) {
2289 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002290 /*
2291 * Mark this request as started so we don't trigger
2292 * any debug logic in the end I/O path.
2293 */
2294 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002295 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002296 } else {
2297 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2298 break;
2299 }
2300 }
2301
2302 return rq;
2303}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002304EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002305
Tejun Heo9934c8c2009-05-08 11:54:16 +09002306void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002307{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002308 struct request_queue *q = rq->q;
2309
Tejun Heo158dbda2009-04-23 11:05:18 +09002310 BUG_ON(list_empty(&rq->queuelist));
2311 BUG_ON(ELV_ON_HASH(rq));
2312
2313 list_del_init(&rq->queuelist);
2314
2315 /*
2316 * the time frame between a request being removed from the lists
2317 * and to it is freed is accounted as io that is in progress at
2318 * the driver side.
2319 */
Divyesh Shah91952912010-04-01 15:01:41 -07002320 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002321 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002322 set_io_start_time_ns(rq);
2323 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002324}
2325
Tejun Heo5efccd12009-04-23 11:05:18 +09002326/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002327 * blk_start_request - start request processing on the driver
2328 * @req: request to dequeue
2329 *
2330 * Description:
2331 * Dequeue @req and start timeout timer on it. This hands off the
2332 * request to the driver.
2333 *
2334 * Block internal functions which don't want to start timer should
2335 * call blk_dequeue_request().
2336 *
2337 * Context:
2338 * queue_lock must be held.
2339 */
2340void blk_start_request(struct request *req)
2341{
2342 blk_dequeue_request(req);
2343
2344 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002345 * We are now handing the request to the hardware, initialize
2346 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002347 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002348 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002349 if (unlikely(blk_bidi_rq(req)))
2350 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2351
Jeff Moyer4912aa62013-10-08 14:36:41 -04002352 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002353 blk_add_timer(req);
2354}
2355EXPORT_SYMBOL(blk_start_request);
2356
2357/**
2358 * blk_fetch_request - fetch a request from a request queue
2359 * @q: request queue to fetch a request from
2360 *
2361 * Description:
2362 * Return the request at the top of @q. The request is started on
2363 * return and LLD can start processing it immediately.
2364 *
2365 * Return:
2366 * Pointer to the request at the top of @q if available. Null
2367 * otherwise.
2368 *
2369 * Context:
2370 * queue_lock must be held.
2371 */
2372struct request *blk_fetch_request(struct request_queue *q)
2373{
2374 struct request *rq;
2375
2376 rq = blk_peek_request(q);
2377 if (rq)
2378 blk_start_request(rq);
2379 return rq;
2380}
2381EXPORT_SYMBOL(blk_fetch_request);
2382
2383/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002384 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002385 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002386 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002387 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002388 *
2389 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002390 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2391 * the request structure even if @req doesn't have leftover.
2392 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002393 *
2394 * This special helper function is only for request stacking drivers
2395 * (e.g. request-based dm) so that they can handle partial completion.
2396 * Actual device drivers should use blk_end_request instead.
2397 *
2398 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2399 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002400 *
2401 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002402 * %false - this request doesn't have any more data
2403 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002404 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002405bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002407 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
Tejun Heo2e60e022009-04-23 11:05:18 +09002409 if (!req->bio)
2410 return false;
2411
Roman Penaf5040d2014-03-04 23:13:10 +09002412 trace_block_rq_complete(req->q, req, nr_bytes);
Jens Axboe2056a782006-03-23 20:00:26 +01002413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002415 * For fs requests, rq is just carrier of independent bio's
2416 * and each partial completion should be handled separately.
2417 * Reset per-request error on each partial completion.
2418 *
2419 * TODO: tj: This is too subtle. It would be better to let
2420 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002422 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 req->errors = 0;
2424
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002425 if (error && req->cmd_type == REQ_TYPE_FS &&
2426 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002427 char *error_type;
2428
2429 switch (error) {
2430 case -ENOLINK:
2431 error_type = "recoverable transport";
2432 break;
2433 case -EREMOTEIO:
2434 error_type = "critical target";
2435 break;
2436 case -EBADE:
2437 error_type = "critical nexus";
2438 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002439 case -ETIMEDOUT:
2440 error_type = "timeout";
2441 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002442 case -ENOSPC:
2443 error_type = "critical space allocation";
2444 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002445 case -ENODATA:
2446 error_type = "critical medium";
2447 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002448 case -EIO:
2449 default:
2450 error_type = "I/O";
2451 break;
2452 }
Yi Zou37d7b342012-08-30 16:26:25 -07002453 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2454 error_type, req->rq_disk ?
2455 req->rq_disk->disk_name : "?",
2456 (unsigned long long)blk_rq_pos(req));
2457
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 }
2459
Jens Axboebc58ba92009-01-23 10:54:44 +01002460 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002461
Kent Overstreetf79ea412012-09-20 16:38:30 -07002462 total_bytes = 0;
2463 while (req->bio) {
2464 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002465 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Kent Overstreet4f024f32013-10-11 15:44:27 -07002467 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
Kent Overstreetf79ea412012-09-20 16:38:30 -07002470 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Kent Overstreetf79ea412012-09-20 16:38:30 -07002472 total_bytes += bio_bytes;
2473 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
Kent Overstreetf79ea412012-09-20 16:38:30 -07002475 if (!nr_bytes)
2476 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 }
2478
2479 /*
2480 * completely done
2481 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002482 if (!req->bio) {
2483 /*
2484 * Reset counters so that the request stacking driver
2485 * can find how many bytes remain in the request
2486 * later.
2487 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002488 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002489 return false;
2490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002492 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002493
2494 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002495 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002496 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002497
Tejun Heo80a761f2009-07-03 17:48:17 +09002498 /* mixed attributes always follow the first bio */
2499 if (req->cmd_flags & REQ_MIXED_MERGE) {
2500 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2501 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2502 }
2503
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002504 /*
2505 * If total number of sectors is less than the first segment
2506 * size, something has gone terribly wrong.
2507 */
2508 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002509 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002510 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002511 }
2512
2513 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002515
Tejun Heo2e60e022009-04-23 11:05:18 +09002516 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517}
Tejun Heo2e60e022009-04-23 11:05:18 +09002518EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Tejun Heo2e60e022009-04-23 11:05:18 +09002520static bool blk_update_bidi_request(struct request *rq, int error,
2521 unsigned int nr_bytes,
2522 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002523{
Tejun Heo2e60e022009-04-23 11:05:18 +09002524 if (blk_update_request(rq, error, nr_bytes))
2525 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002526
Tejun Heo2e60e022009-04-23 11:05:18 +09002527 /* Bidi request must be completed as a whole */
2528 if (unlikely(blk_bidi_rq(rq)) &&
2529 blk_update_request(rq->next_rq, error, bidi_bytes))
2530 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002531
Jens Axboee2e1a142010-06-09 10:42:09 +02002532 if (blk_queue_add_random(rq->q))
2533 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002534
2535 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536}
2537
James Bottomley28018c22010-07-01 19:49:17 +09002538/**
2539 * blk_unprep_request - unprepare a request
2540 * @req: the request
2541 *
2542 * This function makes a request ready for complete resubmission (or
2543 * completion). It happens only after all error handling is complete,
2544 * so represents the appropriate moment to deallocate any resources
2545 * that were allocated to the request in the prep_rq_fn. The queue
2546 * lock is held when calling this.
2547 */
2548void blk_unprep_request(struct request *req)
2549{
2550 struct request_queue *q = req->q;
2551
2552 req->cmd_flags &= ~REQ_DONTPREP;
2553 if (q->unprep_rq_fn)
2554 q->unprep_rq_fn(q, req);
2555}
2556EXPORT_SYMBOL_GPL(blk_unprep_request);
2557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558/*
2559 * queue lock must be held
2560 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002561void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002563 if (blk_rq_tagged(req))
2564 blk_queue_end_tag(req->q, req);
2565
James Bottomleyba396a62009-05-27 14:17:08 +02002566 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002568 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002569 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Mike Andersone78042e2008-10-30 02:16:20 -07002571 blk_delete_timer(req);
2572
James Bottomley28018c22010-07-01 19:49:17 +09002573 if (req->cmd_flags & REQ_DONTPREP)
2574 blk_unprep_request(req);
2575
Jens Axboebc58ba92009-01-23 10:54:44 +01002576 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002579 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002580 else {
2581 if (blk_bidi_rq(req))
2582 __blk_put_request(req->next_rq->q, req->next_rq);
2583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 }
2586}
Christoph Hellwig12120072014-04-16 09:44:59 +02002587EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002589/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002590 * blk_end_bidi_request - Complete a bidi request
2591 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002592 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002593 * @nr_bytes: number of bytes to complete @rq
2594 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002595 *
2596 * Description:
2597 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002598 * Drivers that supports bidi can safely call this member for any
2599 * type of request, bidi or uni. In the later case @bidi_bytes is
2600 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002601 *
2602 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002603 * %false - we are done with this request
2604 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002605 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002606static bool blk_end_bidi_request(struct request *rq, int error,
2607 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002608{
2609 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002610 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002611
Tejun Heo2e60e022009-04-23 11:05:18 +09002612 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2613 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002614
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002615 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002616 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002617 spin_unlock_irqrestore(q->queue_lock, flags);
2618
Tejun Heo2e60e022009-04-23 11:05:18 +09002619 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002620}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002621
2622/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002623 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2624 * @rq: the request to complete
2625 * @error: %0 for success, < %0 for error
2626 * @nr_bytes: number of bytes to complete @rq
2627 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002628 *
2629 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002630 * Identical to blk_end_bidi_request() except that queue lock is
2631 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002632 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002633 * Return:
2634 * %false - we are done with this request
2635 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002636 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002637bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002638 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002639{
Tejun Heo2e60e022009-04-23 11:05:18 +09002640 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2641 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002642
Tejun Heo2e60e022009-04-23 11:05:18 +09002643 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002644
Tejun Heo2e60e022009-04-23 11:05:18 +09002645 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002646}
2647
2648/**
2649 * blk_end_request - Helper function for drivers to complete the request.
2650 * @rq: the request being processed
2651 * @error: %0 for success, < %0 for error
2652 * @nr_bytes: number of bytes to complete
2653 *
2654 * Description:
2655 * Ends I/O on a number of bytes attached to @rq.
2656 * If @rq has leftover, sets it up for the next range of segments.
2657 *
2658 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002659 * %false - we are done with this request
2660 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002661 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002662bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002663{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002664 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002665}
Jens Axboe56ad1742009-07-28 22:11:24 +02002666EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002667
2668/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002669 * blk_end_request_all - Helper function for drives to finish the request.
2670 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002671 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002672 *
2673 * Description:
2674 * Completely finish @rq.
2675 */
2676void blk_end_request_all(struct request *rq, int error)
2677{
2678 bool pending;
2679 unsigned int bidi_bytes = 0;
2680
2681 if (unlikely(blk_bidi_rq(rq)))
2682 bidi_bytes = blk_rq_bytes(rq->next_rq);
2683
2684 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2685 BUG_ON(pending);
2686}
Jens Axboe56ad1742009-07-28 22:11:24 +02002687EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002688
2689/**
2690 * blk_end_request_cur - Helper function to finish the current request chunk.
2691 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002692 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002693 *
2694 * Description:
2695 * Complete the current consecutively mapped chunk from @rq.
2696 *
2697 * Return:
2698 * %false - we are done with this request
2699 * %true - still buffers pending for this request
2700 */
2701bool blk_end_request_cur(struct request *rq, int error)
2702{
2703 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2704}
Jens Axboe56ad1742009-07-28 22:11:24 +02002705EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002706
2707/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002708 * blk_end_request_err - Finish a request till the next failure boundary.
2709 * @rq: the request to finish till the next failure boundary for
2710 * @error: must be negative errno
2711 *
2712 * Description:
2713 * Complete @rq till the next failure boundary.
2714 *
2715 * Return:
2716 * %false - we are done with this request
2717 * %true - still buffers pending for this request
2718 */
2719bool blk_end_request_err(struct request *rq, int error)
2720{
2721 WARN_ON(error >= 0);
2722 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2723}
2724EXPORT_SYMBOL_GPL(blk_end_request_err);
2725
2726/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002727 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002728 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002729 * @error: %0 for success, < %0 for error
2730 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002731 *
2732 * Description:
2733 * Must be called with queue lock held unlike blk_end_request().
2734 *
2735 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002736 * %false - we are done with this request
2737 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002738 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002739bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002740{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002741 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002742}
Jens Axboe56ad1742009-07-28 22:11:24 +02002743EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002744
2745/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002746 * __blk_end_request_all - Helper function for drives to finish the request.
2747 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002748 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002749 *
2750 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002751 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002752 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002753void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002754{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002755 bool pending;
2756 unsigned int bidi_bytes = 0;
2757
2758 if (unlikely(blk_bidi_rq(rq)))
2759 bidi_bytes = blk_rq_bytes(rq->next_rq);
2760
2761 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2762 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002763}
Jens Axboe56ad1742009-07-28 22:11:24 +02002764EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002765
2766/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002767 * __blk_end_request_cur - Helper function to finish the current request chunk.
2768 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002769 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002770 *
2771 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002772 * Complete the current consecutively mapped chunk from @rq. Must
2773 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002774 *
2775 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002776 * %false - we are done with this request
2777 * %true - still buffers pending for this request
2778 */
2779bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002780{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002781 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002782}
Jens Axboe56ad1742009-07-28 22:11:24 +02002783EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002784
Tejun Heo80a761f2009-07-03 17:48:17 +09002785/**
2786 * __blk_end_request_err - Finish a request till the next failure boundary.
2787 * @rq: the request to finish till the next failure boundary for
2788 * @error: must be negative errno
2789 *
2790 * Description:
2791 * Complete @rq till the next failure boundary. Must be called
2792 * with queue lock held.
2793 *
2794 * Return:
2795 * %false - we are done with this request
2796 * %true - still buffers pending for this request
2797 */
2798bool __blk_end_request_err(struct request *rq, int error)
2799{
2800 WARN_ON(error >= 0);
2801 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2802}
2803EXPORT_SYMBOL_GPL(__blk_end_request_err);
2804
Jens Axboe86db1e22008-01-29 14:53:40 +01002805void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2806 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002808 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002809 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
Jens Axboeb4f42e22014-04-10 09:46:28 -06002811 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002812 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002813
Kent Overstreet4f024f32013-10-11 15:44:27 -07002814 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
NeilBrown66846572007-08-16 13:31:28 +02002817 if (bio->bi_bdev)
2818 rq->rq_disk = bio->bi_bdev->bd_disk;
2819}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002821#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2822/**
2823 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2824 * @rq: the request to be flushed
2825 *
2826 * Description:
2827 * Flush all pages in @rq.
2828 */
2829void rq_flush_dcache_pages(struct request *rq)
2830{
2831 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002832 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002833
2834 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002835 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002836}
2837EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2838#endif
2839
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002840/**
2841 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2842 * @q : the queue of the device being checked
2843 *
2844 * Description:
2845 * Check if underlying low-level drivers of a device are busy.
2846 * If the drivers want to export their busy state, they must set own
2847 * exporting function using blk_queue_lld_busy() first.
2848 *
2849 * Basically, this function is used only by request stacking drivers
2850 * to stop dispatching requests to underlying devices when underlying
2851 * devices are busy. This behavior helps more I/O merging on the queue
2852 * of the request stacking driver and prevents I/O throughput regression
2853 * on burst I/O load.
2854 *
2855 * Return:
2856 * 0 - Not busy (The request stacking driver should dispatch request)
2857 * 1 - Busy (The request stacking driver should stop dispatching request)
2858 */
2859int blk_lld_busy(struct request_queue *q)
2860{
2861 if (q->lld_busy_fn)
2862 return q->lld_busy_fn(q);
2863
2864 return 0;
2865}
2866EXPORT_SYMBOL_GPL(blk_lld_busy);
2867
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002868/**
2869 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2870 * @rq: the clone request to be cleaned up
2871 *
2872 * Description:
2873 * Free all bios in @rq for a cloned request.
2874 */
2875void blk_rq_unprep_clone(struct request *rq)
2876{
2877 struct bio *bio;
2878
2879 while ((bio = rq->bio) != NULL) {
2880 rq->bio = bio->bi_next;
2881
2882 bio_put(bio);
2883 }
2884}
2885EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2886
2887/*
2888 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002889 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002890 */
2891static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2892{
2893 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002894 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002895 dst->cmd_type = src->cmd_type;
2896 dst->__sector = blk_rq_pos(src);
2897 dst->__data_len = blk_rq_bytes(src);
2898 dst->nr_phys_segments = src->nr_phys_segments;
2899 dst->ioprio = src->ioprio;
2900 dst->extra_len = src->extra_len;
2901}
2902
2903/**
2904 * blk_rq_prep_clone - Helper function to setup clone request
2905 * @rq: the request to be setup
2906 * @rq_src: original request to be cloned
2907 * @bs: bio_set that bios for clone are allocated from
2908 * @gfp_mask: memory allocation mask for bio
2909 * @bio_ctr: setup function to be called for each clone bio.
2910 * Returns %0 for success, non %0 for failure.
2911 * @data: private data to be passed to @bio_ctr
2912 *
2913 * Description:
2914 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002915 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002916 * are not copied, and copying such parts is the caller's responsibility.
2917 * Also, pages which the original bios are pointing to are not copied
2918 * and the cloned bios just point same pages.
2919 * So cloned bios must be completed before original bios, which means
2920 * the caller must complete @rq before @rq_src.
2921 */
2922int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2923 struct bio_set *bs, gfp_t gfp_mask,
2924 int (*bio_ctr)(struct bio *, struct bio *, void *),
2925 void *data)
2926{
2927 struct bio *bio, *bio_src;
2928
2929 if (!bs)
2930 bs = fs_bio_set;
2931
2932 blk_rq_init(NULL, rq);
2933
2934 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002935 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002936 if (!bio)
2937 goto free_and_out;
2938
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002939 if (bio_ctr && bio_ctr(bio, bio_src, data))
2940 goto free_and_out;
2941
2942 if (rq->bio) {
2943 rq->biotail->bi_next = bio;
2944 rq->biotail = bio;
2945 } else
2946 rq->bio = rq->biotail = bio;
2947 }
2948
2949 __blk_rq_prep_clone(rq, rq_src);
2950
2951 return 0;
2952
2953free_and_out:
2954 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002955 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002956 blk_rq_unprep_clone(rq);
2957
2958 return -ENOMEM;
2959}
2960EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2961
Jens Axboe59c3d452014-04-08 09:15:35 -06002962int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963{
2964 return queue_work(kblockd_workqueue, work);
2965}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966EXPORT_SYMBOL(kblockd_schedule_work);
2967
Jens Axboe59c3d452014-04-08 09:15:35 -06002968int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2969 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04002970{
2971 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2972}
2973EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2974
Jens Axboe8ab14592014-04-08 09:17:40 -06002975int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2976 unsigned long delay)
2977{
2978 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2979}
2980EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2981
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002982/**
2983 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2984 * @plug: The &struct blk_plug that needs to be initialized
2985 *
2986 * Description:
2987 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2988 * pending I/O should the task end up blocking between blk_start_plug() and
2989 * blk_finish_plug(). This is important from a performance perspective, but
2990 * also ensures that we don't deadlock. For instance, if the task is blocking
2991 * for a memory allocation, memory reclaim could end up wanting to free a
2992 * page belonging to that request that is currently residing in our private
2993 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2994 * this kind of deadlock.
2995 */
Jens Axboe73c10102011-03-08 13:19:51 +01002996void blk_start_plug(struct blk_plug *plug)
2997{
2998 struct task_struct *tsk = current;
2999
Jens Axboe73c10102011-03-08 13:19:51 +01003000 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003001 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003002 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003003
3004 /*
3005 * If this is a nested plug, don't actually assign it. It will be
3006 * flushed on its own.
3007 */
3008 if (!tsk->plug) {
3009 /*
3010 * Store ordering should not be needed here, since a potential
3011 * preempt will imply a full memory barrier
3012 */
3013 tsk->plug = plug;
3014 }
3015}
3016EXPORT_SYMBOL(blk_start_plug);
3017
3018static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3019{
3020 struct request *rqa = container_of(a, struct request, queuelist);
3021 struct request *rqb = container_of(b, struct request, queuelist);
3022
Jianpeng Ma975927b2012-10-25 21:58:17 +02003023 return !(rqa->q < rqb->q ||
3024 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003025}
3026
Jens Axboe49cac012011-04-16 13:51:05 +02003027/*
3028 * If 'from_schedule' is true, then postpone the dispatch of requests
3029 * until a safe kblockd context. We due this to avoid accidental big
3030 * additional stack usage in driver dispatch, in places where the originally
3031 * plugger did not intend it.
3032 */
Jens Axboef6603782011-04-15 15:49:07 +02003033static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003034 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003035 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003036{
Jens Axboe49cac012011-04-16 13:51:05 +02003037 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003038
Bart Van Assche70460572012-11-28 13:45:56 +01003039 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003040 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003041 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003042 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003043 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003044}
3045
NeilBrown74018dc2012-07-31 09:08:15 +02003046static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003047{
3048 LIST_HEAD(callbacks);
3049
Shaohua Li2a7d5552012-07-31 09:08:15 +02003050 while (!list_empty(&plug->cb_list)) {
3051 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003052
Shaohua Li2a7d5552012-07-31 09:08:15 +02003053 while (!list_empty(&callbacks)) {
3054 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003055 struct blk_plug_cb,
3056 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003057 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003058 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003059 }
NeilBrown048c9372011-04-18 09:52:22 +02003060 }
3061}
3062
NeilBrown9cbb1752012-07-31 09:08:14 +02003063struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3064 int size)
3065{
3066 struct blk_plug *plug = current->plug;
3067 struct blk_plug_cb *cb;
3068
3069 if (!plug)
3070 return NULL;
3071
3072 list_for_each_entry(cb, &plug->cb_list, list)
3073 if (cb->callback == unplug && cb->data == data)
3074 return cb;
3075
3076 /* Not currently on the callback list */
3077 BUG_ON(size < sizeof(*cb));
3078 cb = kzalloc(size, GFP_ATOMIC);
3079 if (cb) {
3080 cb->data = data;
3081 cb->callback = unplug;
3082 list_add(&cb->list, &plug->cb_list);
3083 }
3084 return cb;
3085}
3086EXPORT_SYMBOL(blk_check_plugged);
3087
Jens Axboe49cac012011-04-16 13:51:05 +02003088void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003089{
3090 struct request_queue *q;
3091 unsigned long flags;
3092 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003093 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003094 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003095
NeilBrown74018dc2012-07-31 09:08:15 +02003096 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003097
3098 if (!list_empty(&plug->mq_list))
3099 blk_mq_flush_plug_list(plug, from_schedule);
3100
Jens Axboe73c10102011-03-08 13:19:51 +01003101 if (list_empty(&plug->list))
3102 return;
3103
NeilBrown109b8122011-04-11 14:13:10 +02003104 list_splice_init(&plug->list, &list);
3105
Jianpeng Ma422765c2013-01-11 14:46:09 +01003106 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003107
3108 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003109 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003110
3111 /*
3112 * Save and disable interrupts here, to avoid doing it for every
3113 * queue lock we have to take.
3114 */
Jens Axboe73c10102011-03-08 13:19:51 +01003115 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003116 while (!list_empty(&list)) {
3117 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003118 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003119 BUG_ON(!rq->q);
3120 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003121 /*
3122 * This drops the queue lock
3123 */
3124 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003125 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003126 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003127 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003128 spin_lock(q->queue_lock);
3129 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003130
3131 /*
3132 * Short-circuit if @q is dead
3133 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003134 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003135 __blk_end_request_all(rq, -ENODEV);
3136 continue;
3137 }
3138
Jens Axboe73c10102011-03-08 13:19:51 +01003139 /*
3140 * rq is already accounted, so use raw insert
3141 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003142 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3143 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3144 else
3145 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003146
3147 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003148 }
3149
Jens Axboe99e22592011-04-18 09:59:55 +02003150 /*
3151 * This drops the queue lock
3152 */
3153 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003154 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003155
Jens Axboe73c10102011-03-08 13:19:51 +01003156 local_irq_restore(flags);
3157}
Jens Axboe73c10102011-03-08 13:19:51 +01003158
3159void blk_finish_plug(struct blk_plug *plug)
3160{
Jens Axboef6603782011-04-15 15:49:07 +02003161 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003162
3163 if (plug == current->plug)
3164 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003165}
3166EXPORT_SYMBOL(blk_finish_plug);
3167
Lin Ming6c954662013-03-23 11:42:26 +08003168#ifdef CONFIG_PM_RUNTIME
3169/**
3170 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3171 * @q: the queue of the device
3172 * @dev: the device the queue belongs to
3173 *
3174 * Description:
3175 * Initialize runtime-PM-related fields for @q and start auto suspend for
3176 * @dev. Drivers that want to take advantage of request-based runtime PM
3177 * should call this function after @dev has been initialized, and its
3178 * request queue @q has been allocated, and runtime PM for it can not happen
3179 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3180 * cases, driver should call this function before any I/O has taken place.
3181 *
3182 * This function takes care of setting up using auto suspend for the device,
3183 * the autosuspend delay is set to -1 to make runtime suspend impossible
3184 * until an updated value is either set by user or by driver. Drivers do
3185 * not need to touch other autosuspend settings.
3186 *
3187 * The block layer runtime PM is request based, so only works for drivers
3188 * that use request as their IO unit instead of those directly use bio's.
3189 */
3190void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3191{
3192 q->dev = dev;
3193 q->rpm_status = RPM_ACTIVE;
3194 pm_runtime_set_autosuspend_delay(q->dev, -1);
3195 pm_runtime_use_autosuspend(q->dev);
3196}
3197EXPORT_SYMBOL(blk_pm_runtime_init);
3198
3199/**
3200 * blk_pre_runtime_suspend - Pre runtime suspend check
3201 * @q: the queue of the device
3202 *
3203 * Description:
3204 * This function will check if runtime suspend is allowed for the device
3205 * by examining if there are any requests pending in the queue. If there
3206 * are requests pending, the device can not be runtime suspended; otherwise,
3207 * the queue's status will be updated to SUSPENDING and the driver can
3208 * proceed to suspend the device.
3209 *
3210 * For the not allowed case, we mark last busy for the device so that
3211 * runtime PM core will try to autosuspend it some time later.
3212 *
3213 * This function should be called near the start of the device's
3214 * runtime_suspend callback.
3215 *
3216 * Return:
3217 * 0 - OK to runtime suspend the device
3218 * -EBUSY - Device should not be runtime suspended
3219 */
3220int blk_pre_runtime_suspend(struct request_queue *q)
3221{
3222 int ret = 0;
3223
3224 spin_lock_irq(q->queue_lock);
3225 if (q->nr_pending) {
3226 ret = -EBUSY;
3227 pm_runtime_mark_last_busy(q->dev);
3228 } else {
3229 q->rpm_status = RPM_SUSPENDING;
3230 }
3231 spin_unlock_irq(q->queue_lock);
3232 return ret;
3233}
3234EXPORT_SYMBOL(blk_pre_runtime_suspend);
3235
3236/**
3237 * blk_post_runtime_suspend - Post runtime suspend processing
3238 * @q: the queue of the device
3239 * @err: return value of the device's runtime_suspend function
3240 *
3241 * Description:
3242 * Update the queue's runtime status according to the return value of the
3243 * device's runtime suspend function and mark last busy for the device so
3244 * that PM core will try to auto suspend the device at a later time.
3245 *
3246 * This function should be called near the end of the device's
3247 * runtime_suspend callback.
3248 */
3249void blk_post_runtime_suspend(struct request_queue *q, int err)
3250{
3251 spin_lock_irq(q->queue_lock);
3252 if (!err) {
3253 q->rpm_status = RPM_SUSPENDED;
3254 } else {
3255 q->rpm_status = RPM_ACTIVE;
3256 pm_runtime_mark_last_busy(q->dev);
3257 }
3258 spin_unlock_irq(q->queue_lock);
3259}
3260EXPORT_SYMBOL(blk_post_runtime_suspend);
3261
3262/**
3263 * blk_pre_runtime_resume - Pre runtime resume processing
3264 * @q: the queue of the device
3265 *
3266 * Description:
3267 * Update the queue's runtime status to RESUMING in preparation for the
3268 * runtime resume of the device.
3269 *
3270 * This function should be called near the start of the device's
3271 * runtime_resume callback.
3272 */
3273void blk_pre_runtime_resume(struct request_queue *q)
3274{
3275 spin_lock_irq(q->queue_lock);
3276 q->rpm_status = RPM_RESUMING;
3277 spin_unlock_irq(q->queue_lock);
3278}
3279EXPORT_SYMBOL(blk_pre_runtime_resume);
3280
3281/**
3282 * blk_post_runtime_resume - Post runtime resume processing
3283 * @q: the queue of the device
3284 * @err: return value of the device's runtime_resume function
3285 *
3286 * Description:
3287 * Update the queue's runtime status according to the return value of the
3288 * device's runtime_resume function. If it is successfully resumed, process
3289 * the requests that are queued into the device's queue when it is resuming
3290 * and then mark last busy and initiate autosuspend for it.
3291 *
3292 * This function should be called near the end of the device's
3293 * runtime_resume callback.
3294 */
3295void blk_post_runtime_resume(struct request_queue *q, int err)
3296{
3297 spin_lock_irq(q->queue_lock);
3298 if (!err) {
3299 q->rpm_status = RPM_ACTIVE;
3300 __blk_run_queue(q);
3301 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003302 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003303 } else {
3304 q->rpm_status = RPM_SUSPENDED;
3305 }
3306 spin_unlock_irq(q->queue_lock);
3307}
3308EXPORT_SYMBOL(blk_post_runtime_resume);
3309#endif
3310
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311int __init blk_dev_init(void)
3312{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003313 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3314 sizeof(((struct request *)0)->cmd_flags));
3315
Tejun Heo89b90be2011-01-03 15:01:47 +01003316 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3317 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003318 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 if (!kblockd_workqueue)
3320 panic("Failed to create kblockd\n");
3321
3322 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003323 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324
Jens Axboe8324aa92008-01-29 14:51:59 +01003325 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003326 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 return 0;
3329}