blob: bf930f481d437ac1b1731d32d453280309773445 [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));
Jens Axboef27b0872014-06-06 07:57:37 -06001255}
1256EXPORT_SYMBOL(blk_rq_set_block_pc);
1257
1258/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 * blk_requeue_request - put a request back on queue
1260 * @q: request queue where request should be inserted
1261 * @rq: request to be inserted
1262 *
1263 * Description:
1264 * Drivers often keep queueing requests until the hardware cannot accept
1265 * more, when that condition happens we need to put the request back
1266 * on the queue. Must be called with queue lock held.
1267 */
Jens Axboe165125e2007-07-24 09:28:11 +02001268void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001270 blk_delete_timer(rq);
1271 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001272 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 if (blk_rq_tagged(rq))
1275 blk_queue_end_tag(q, rq);
1276
James Bottomleyba396a62009-05-27 14:17:08 +02001277 BUG_ON(blk_queued_rq(rq));
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 elv_requeue_request(q, rq);
1280}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281EXPORT_SYMBOL(blk_requeue_request);
1282
Jens Axboe73c10102011-03-08 13:19:51 +01001283static void add_acct_request(struct request_queue *q, struct request *rq,
1284 int where)
1285{
Jens Axboe320ae512013-10-24 09:20:05 +01001286 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001287 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001288}
1289
Tejun Heo074a7ac2008-08-25 19:56:14 +09001290static void part_round_stats_single(int cpu, struct hd_struct *part,
1291 unsigned long now)
1292{
Jens Axboe7276d022014-05-09 15:48:23 -06001293 int inflight;
1294
Tejun Heo074a7ac2008-08-25 19:56:14 +09001295 if (now == part->stamp)
1296 return;
1297
Jens Axboe7276d022014-05-09 15:48:23 -06001298 inflight = part_in_flight(part);
1299 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001300 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001301 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001302 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1303 }
1304 part->stamp = now;
1305}
1306
1307/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001308 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1309 * @cpu: cpu number for stats access
1310 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 *
1312 * The average IO queue length and utilisation statistics are maintained
1313 * by observing the current state of the queue length and the amount of
1314 * time it has been in this state for.
1315 *
1316 * Normally, that accounting is done on IO completion, but that can result
1317 * in more than a second's worth of IO being accounted for within any one
1318 * second, leading to >100% utilisation. To deal with that, we call this
1319 * function to do a round-off before returning the results when reading
1320 * /proc/diskstats. This accounts immediately for all queue usage up to
1321 * the current jiffies and restarts the counters again.
1322 */
Tejun Heoc9959052008-08-25 19:47:21 +09001323void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001324{
1325 unsigned long now = jiffies;
1326
Tejun Heo074a7ac2008-08-25 19:56:14 +09001327 if (part->partno)
1328 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1329 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001330}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001331EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001332
Lin Mingc8158812013-03-23 11:42:27 +08001333#ifdef CONFIG_PM_RUNTIME
1334static void blk_pm_put_request(struct request *rq)
1335{
1336 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1337 pm_runtime_mark_last_busy(rq->q->dev);
1338}
1339#else
1340static inline void blk_pm_put_request(struct request *rq) {}
1341#endif
1342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343/*
1344 * queue lock must be held
1345 */
Jens Axboe165125e2007-07-24 09:28:11 +02001346void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (unlikely(!q))
1349 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001351 if (q->mq_ops) {
1352 blk_mq_free_request(req);
1353 return;
1354 }
1355
Lin Mingc8158812013-03-23 11:42:27 +08001356 blk_pm_put_request(req);
1357
Tejun Heo8922e162005-10-20 16:23:44 +02001358 elv_completed_request(q, req);
1359
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001360 /* this is a bio leak */
1361 WARN_ON(req->bio != NULL);
1362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 /*
1364 * Request may not have originated from ll_rw_blk. if not,
1365 * it didn't come out of our reserved rq pools
1366 */
Jens Axboe49171e52006-08-10 08:59:11 +02001367 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001368 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001369 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001372 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Tejun Heoa0516612012-06-26 15:05:44 -07001374 blk_free_request(rl, req);
1375 freed_request(rl, flags);
1376 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001379EXPORT_SYMBOL_GPL(__blk_put_request);
1380
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381void blk_put_request(struct request *req)
1382{
Jens Axboe165125e2007-07-24 09:28:11 +02001383 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Jens Axboe320ae512013-10-24 09:20:05 +01001385 if (q->mq_ops)
1386 blk_mq_free_request(req);
1387 else {
1388 unsigned long flags;
1389
1390 spin_lock_irqsave(q->queue_lock, flags);
1391 __blk_put_request(q, req);
1392 spin_unlock_irqrestore(q->queue_lock, flags);
1393 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395EXPORT_SYMBOL(blk_put_request);
1396
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001397/**
1398 * blk_add_request_payload - add a payload to a request
1399 * @rq: request to update
1400 * @page: page backing the payload
1401 * @len: length of the payload.
1402 *
1403 * This allows to later add a payload to an already submitted request by
1404 * a block driver. The driver needs to take care of freeing the payload
1405 * itself.
1406 *
1407 * Note that this is a quite horrible hack and nothing but handling of
1408 * discard requests should ever use it.
1409 */
1410void blk_add_request_payload(struct request *rq, struct page *page,
1411 unsigned int len)
1412{
1413 struct bio *bio = rq->bio;
1414
1415 bio->bi_io_vec->bv_page = page;
1416 bio->bi_io_vec->bv_offset = 0;
1417 bio->bi_io_vec->bv_len = len;
1418
Kent Overstreet4f024f32013-10-11 15:44:27 -07001419 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001420 bio->bi_vcnt = 1;
1421 bio->bi_phys_segments = 1;
1422
1423 rq->__data_len = rq->resid_len = len;
1424 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001425}
1426EXPORT_SYMBOL_GPL(blk_add_request_payload);
1427
Jens Axboe320ae512013-10-24 09:20:05 +01001428bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1429 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001430{
1431 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1432
Jens Axboe73c10102011-03-08 13:19:51 +01001433 if (!ll_back_merge_fn(q, req, bio))
1434 return false;
1435
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001436 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001437
1438 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1439 blk_rq_set_mixed_merge(req);
1440
1441 req->biotail->bi_next = bio;
1442 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001443 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001444 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1445
Jens Axboe320ae512013-10-24 09:20:05 +01001446 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001447 return true;
1448}
1449
Jens Axboe320ae512013-10-24 09:20:05 +01001450bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1451 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001452{
1453 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001454
Jens Axboe73c10102011-03-08 13:19:51 +01001455 if (!ll_front_merge_fn(q, req, bio))
1456 return false;
1457
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001458 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001459
1460 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1461 blk_rq_set_mixed_merge(req);
1462
Jens Axboe73c10102011-03-08 13:19:51 +01001463 bio->bi_next = req->bio;
1464 req->bio = bio;
1465
Kent Overstreet4f024f32013-10-11 15:44:27 -07001466 req->__sector = bio->bi_iter.bi_sector;
1467 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001468 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1469
Jens Axboe320ae512013-10-24 09:20:05 +01001470 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001471 return true;
1472}
1473
Tejun Heobd87b582011-10-19 14:33:08 +02001474/**
Jens Axboe320ae512013-10-24 09:20:05 +01001475 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001476 * @q: request_queue new bio is being queued at
1477 * @bio: new bio being queued
1478 * @request_count: out parameter for number of traversed plugged requests
1479 *
1480 * Determine whether @bio being queued on @q can be merged with a request
1481 * on %current's plugged list. Returns %true if merge was successful,
1482 * otherwise %false.
1483 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001484 * Plugging coalesces IOs from the same issuer for the same purpose without
1485 * going through @q->queue_lock. As such it's more of an issuing mechanism
1486 * than scheduling, and the request, while may have elvpriv data, is not
1487 * added on the elevator at this point. In addition, we don't have
1488 * reliable access to the elevator outside queue lock. Only check basic
1489 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001490 *
1491 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001492 */
Jens Axboe320ae512013-10-24 09:20:05 +01001493bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1494 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001495{
1496 struct blk_plug *plug;
1497 struct request *rq;
1498 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001499 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001500
Tejun Heobd87b582011-10-19 14:33:08 +02001501 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001502 if (!plug)
1503 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001504 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001505
Shaohua Li92f399c2013-10-29 12:01:03 -06001506 if (q->mq_ops)
1507 plug_list = &plug->mq_list;
1508 else
1509 plug_list = &plug->list;
1510
1511 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001512 int el_ret;
1513
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001514 if (rq->q == q)
1515 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001516
Tejun Heo07c2bd32012-02-08 09:19:42 +01001517 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001518 continue;
1519
Tejun Heo050c8ea2012-02-08 09:19:38 +01001520 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001521 if (el_ret == ELEVATOR_BACK_MERGE) {
1522 ret = bio_attempt_back_merge(q, rq, bio);
1523 if (ret)
1524 break;
1525 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1526 ret = bio_attempt_front_merge(q, rq, bio);
1527 if (ret)
1528 break;
1529 }
1530 }
1531out:
1532 return ret;
1533}
1534
Jens Axboe86db1e22008-01-29 14:53:40 +01001535void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001536{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001537 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001538
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001539 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1540 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001541 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001542
Tejun Heo52d9e672006-01-06 09:49:58 +01001543 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001544 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001545 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001546 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001547}
1548
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001549void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001551 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001552 struct blk_plug *plug;
1553 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1554 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001555 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 /*
1558 * low level driver can indicate that it wants pages above a
1559 * certain limit bounced to low memory (ie for highmem, or even
1560 * ISA dma in theory)
1561 */
1562 blk_queue_bounce(q, &bio);
1563
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001564 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1565 bio_endio(bio, -EIO);
1566 return;
1567 }
1568
Tejun Heo4fed9472010-09-03 11:56:17 +02001569 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001570 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001571 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001572 goto get_rq;
1573 }
1574
Jens Axboe73c10102011-03-08 13:19:51 +01001575 /*
1576 * Check if we can merge with the plugged list before grabbing
1577 * any locks.
1578 */
Robert Elliottda41a582014-05-20 16:46:26 -05001579 if (!blk_queue_nomerges(q) &&
1580 blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001581 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001582
1583 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001586 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001587 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001588 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001589 if (!attempt_back_merge(q, req))
1590 elv_merged_request(q, req, el_ret);
1591 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001592 }
Jens Axboe73c10102011-03-08 13:19:51 +01001593 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001594 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001595 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001596 if (!attempt_front_merge(q, req))
1597 elv_merged_request(q, req, el_ret);
1598 goto out_unlock;
1599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 }
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001603 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001604 * This sync check and mask will be re-done in init_request_from_bio(),
1605 * but we need to set it earlier to expose the sync flag to the
1606 * rq allocator and io schedulers.
1607 */
1608 rw_flags = bio_data_dir(bio);
1609 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001610 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001611
1612 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001613 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001614 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001615 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001616 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001617 if (unlikely(!req)) {
1618 bio_endio(bio, -ENODEV); /* @q is dead */
1619 goto out_unlock;
1620 }
Nick Piggind6344532005-06-28 20:45:14 -07001621
Nick Piggin450991b2005-06-28 20:45:13 -07001622 /*
1623 * After dropping the lock and possibly sleeping here, our request
1624 * may now be mergeable after it had proven unmergeable (above).
1625 * We don't worry about that case for efficiency. It won't happen
1626 * often, and the elevators are able to handle it.
1627 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001628 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
Tao Ma9562ad92011-10-24 16:11:30 +02001630 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001631 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001632
Jens Axboe73c10102011-03-08 13:19:51 +01001633 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001634 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001635 /*
1636 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001637 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001638 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001639 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001640 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001641 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001642 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001643 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001644 trace_block_plug(q);
1645 }
Jens Axboe73c10102011-03-08 13:19:51 +01001646 }
Shaohua Lia6327162011-08-24 16:04:32 +02001647 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001648 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001649 } else {
1650 spin_lock_irq(q->queue_lock);
1651 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001652 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001653out_unlock:
1654 spin_unlock_irq(q->queue_lock);
1655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656}
Jens Axboec20e8de2011-09-12 12:03:37 +02001657EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659/*
1660 * If bio->bi_dev is a partition, remap the location
1661 */
1662static inline void blk_partition_remap(struct bio *bio)
1663{
1664 struct block_device *bdev = bio->bi_bdev;
1665
Jens Axboebf2de6f2007-09-27 13:01:25 +02001666 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001668
Kent Overstreet4f024f32013-10-11 15:44:27 -07001669 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001671
Mike Snitzerd07335e2010-11-16 12:52:38 +01001672 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1673 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001674 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
1676}
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678static void handle_bad_sector(struct bio *bio)
1679{
1680 char b[BDEVNAME_SIZE];
1681
1682 printk(KERN_INFO "attempt to access beyond end of device\n");
1683 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1684 bdevname(bio->bi_bdev, b),
1685 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001686 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001687 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 set_bit(BIO_EOF, &bio->bi_flags);
1690}
1691
Akinobu Mitac17bb492006-12-08 02:39:46 -08001692#ifdef CONFIG_FAIL_MAKE_REQUEST
1693
1694static DECLARE_FAULT_ATTR(fail_make_request);
1695
1696static int __init setup_fail_make_request(char *str)
1697{
1698 return setup_fault_attr(&fail_make_request, str);
1699}
1700__setup("fail_make_request=", setup_fail_make_request);
1701
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001702static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001703{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001704 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001705}
1706
1707static int __init fail_make_request_debugfs(void)
1708{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001709 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1710 NULL, &fail_make_request);
1711
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001712 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001713}
1714
1715late_initcall(fail_make_request_debugfs);
1716
1717#else /* CONFIG_FAIL_MAKE_REQUEST */
1718
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001719static inline bool should_fail_request(struct hd_struct *part,
1720 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001721{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001722 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001723}
1724
1725#endif /* CONFIG_FAIL_MAKE_REQUEST */
1726
Jens Axboec07e2b42007-07-18 13:27:58 +02001727/*
1728 * Check whether this bio extends beyond the end of the device.
1729 */
1730static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1731{
1732 sector_t maxsector;
1733
1734 if (!nr_sectors)
1735 return 0;
1736
1737 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001738 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001739 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001740 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001741
1742 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1743 /*
1744 * This may well happen - the kernel calls bread()
1745 * without checking the size of the device, e.g., when
1746 * mounting a device.
1747 */
1748 handle_bad_sector(bio);
1749 return 1;
1750 }
1751 }
1752
1753 return 0;
1754}
1755
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001756static noinline_for_stack bool
1757generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758{
Jens Axboe165125e2007-07-24 09:28:11 +02001759 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001760 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001761 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001762 char b[BDEVNAME_SIZE];
1763 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
1765 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766
Jens Axboec07e2b42007-07-18 13:27:58 +02001767 if (bio_check_eod(bio, nr_sectors))
1768 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001770 q = bdev_get_queue(bio->bi_bdev);
1771 if (unlikely(!q)) {
1772 printk(KERN_ERR
1773 "generic_make_request: Trying to access "
1774 "nonexistent block-device %s (%Lu)\n",
1775 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001776 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001777 goto end_io;
1778 }
1779
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001780 if (likely(bio_is_rw(bio) &&
1781 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001782 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1783 bdevname(bio->bi_bdev, b),
1784 bio_sectors(bio),
1785 queue_max_hw_sectors(q));
1786 goto end_io;
1787 }
1788
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001789 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001790 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001791 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001792 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001793 goto end_io;
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001796 * If this device has partitions, remap block n
1797 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001799 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001801 if (bio_check_eod(bio, nr_sectors))
1802 goto end_io;
1803
1804 /*
1805 * Filter flush bio's early so that make_request based
1806 * drivers without flush support don't have to worry
1807 * about them.
1808 */
1809 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1810 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1811 if (!nr_sectors) {
1812 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001813 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001817 if ((bio->bi_rw & REQ_DISCARD) &&
1818 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001819 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001820 err = -EOPNOTSUPP;
1821 goto end_io;
1822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001824 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 err = -EOPNOTSUPP;
1826 goto end_io;
1827 }
1828
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001829 /*
1830 * Various block parts want %current->io_context and lazy ioc
1831 * allocation ends up trading a lot of pain for a small amount of
1832 * memory. Just allocate it upfront. This may fail and block
1833 * layer knows how to live with it.
1834 */
1835 create_io_context(GFP_ATOMIC, q->node);
1836
Tejun Heobc16a4f2011-10-19 14:33:01 +02001837 if (blk_throtl_bio(q, bio))
1838 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001839
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001840 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001841 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001842
1843end_io:
1844 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001845 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
1847
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001848/**
1849 * generic_make_request - hand a buffer to its device driver for I/O
1850 * @bio: The bio describing the location in memory and on the device.
1851 *
1852 * generic_make_request() is used to make I/O requests of block
1853 * devices. It is passed a &struct bio, which describes the I/O that needs
1854 * to be done.
1855 *
1856 * generic_make_request() does not return any status. The
1857 * success/failure status of the request, along with notification of
1858 * completion, is delivered asynchronously through the bio->bi_end_io
1859 * function described (one day) else where.
1860 *
1861 * The caller of generic_make_request must make sure that bi_io_vec
1862 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1863 * set to describe the device address, and the
1864 * bi_end_io and optionally bi_private are set to describe how
1865 * completion notification should be signaled.
1866 *
1867 * generic_make_request and the drivers it calls may use bi_next if this
1868 * bio happens to be merged with someone else, and may resubmit the bio to
1869 * a lower device by calling into generic_make_request recursively, which
1870 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001871 */
1872void generic_make_request(struct bio *bio)
1873{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001874 struct bio_list bio_list_on_stack;
1875
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001876 if (!generic_make_request_checks(bio))
1877 return;
1878
1879 /*
1880 * We only want one ->make_request_fn to be active at a time, else
1881 * stack usage with stacked devices could be a problem. So use
1882 * current->bio_list to keep a list of requests submited by a
1883 * make_request_fn function. current->bio_list is also used as a
1884 * flag to say if generic_make_request is currently active in this
1885 * task or not. If it is NULL, then no make_request is active. If
1886 * it is non-NULL, then a make_request is active, and new requests
1887 * should be added at the tail
1888 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001889 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001890 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001891 return;
1892 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001893
Neil Brownd89d8792007-05-01 09:53:42 +02001894 /* following loop may be a bit non-obvious, and so deserves some
1895 * explanation.
1896 * Before entering the loop, bio->bi_next is NULL (as all callers
1897 * ensure that) so we have a list with a single bio.
1898 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001899 * we assign bio_list to a pointer to the bio_list_on_stack,
1900 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001901 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001902 * through a recursive call to generic_make_request. If it
1903 * did, we find a non-NULL value in bio_list and re-enter the loop
1904 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001905 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001906 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001907 */
1908 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001909 bio_list_init(&bio_list_on_stack);
1910 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001911 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001912 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1913
1914 q->make_request_fn(q, bio);
1915
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001916 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001917 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001918 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001919}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920EXPORT_SYMBOL(generic_make_request);
1921
1922/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001923 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1925 * @bio: The &struct bio which describes the I/O
1926 *
1927 * submit_bio() is very similar in purpose to generic_make_request(), and
1928 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001929 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 *
1931 */
1932void submit_bio(int rw, struct bio *bio)
1933{
Jens Axboe22e2c502005-06-27 10:55:12 +02001934 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Jens Axboebf2de6f2007-09-27 13:01:25 +02001936 /*
1937 * If it's a regular read/write or a barrier with data attached,
1938 * go through the normal accounting stuff before submission.
1939 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001940 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001941 unsigned int count;
1942
1943 if (unlikely(rw & REQ_WRITE_SAME))
1944 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1945 else
1946 count = bio_sectors(bio);
1947
Jens Axboebf2de6f2007-09-27 13:01:25 +02001948 if (rw & WRITE) {
1949 count_vm_events(PGPGOUT, count);
1950 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001951 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001952 count_vm_events(PGPGIN, count);
1953 }
1954
1955 if (unlikely(block_dump)) {
1956 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001957 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001958 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001959 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001960 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001961 bdevname(bio->bi_bdev, b),
1962 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 }
1965
1966 generic_make_request(bio);
1967}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968EXPORT_SYMBOL(submit_bio);
1969
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001970/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001971 * blk_rq_check_limits - Helper function to check a request for the queue limit
1972 * @q: the queue
1973 * @rq: the request being checked
1974 *
1975 * Description:
1976 * @rq may have been made based on weaker limitations of upper-level queues
1977 * in request stacking drivers, and it may violate the limitation of @q.
1978 * Since the block layer and the underlying device driver trust @rq
1979 * after it is inserted to @q, it should be checked against @q before
1980 * the insertion using this generic function.
1981 *
1982 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001983 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001984 * Request stacking drivers like request-based dm may change the queue
1985 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09001986 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001987 * the new queue limits again when they dispatch those requests,
1988 * although such checkings are also done against the old queue limits
1989 * when submitting requests.
1990 */
1991int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1992{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001993 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001994 return 0;
1995
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001996 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001997 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1998 return -EIO;
1999 }
2000
2001 /*
2002 * queue's settings related to segment counting like q->bounce_pfn
2003 * may differ from that of other stacking queues.
2004 * Recalculate it to check the request correctly on this queue's
2005 * limitation.
2006 */
2007 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002008 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002009 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2010 return -EIO;
2011 }
2012
2013 return 0;
2014}
2015EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2016
2017/**
2018 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2019 * @q: the queue to submit the request
2020 * @rq: the request being queued
2021 */
2022int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2023{
2024 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002025 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002026
2027 if (blk_rq_check_limits(q, rq))
2028 return -EIO;
2029
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002030 if (rq->rq_disk &&
2031 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002032 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002033
2034 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002035 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002036 spin_unlock_irqrestore(q->queue_lock, flags);
2037 return -ENODEV;
2038 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002039
2040 /*
2041 * Submitting request must be dequeued before calling this function
2042 * because it will be linked to another request_queue
2043 */
2044 BUG_ON(blk_queued_rq(rq));
2045
Jeff Moyer4853aba2011-08-15 21:37:25 +02002046 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2047 where = ELEVATOR_INSERT_FLUSH;
2048
2049 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002050 if (where == ELEVATOR_INSERT_FLUSH)
2051 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002052 spin_unlock_irqrestore(q->queue_lock, flags);
2053
2054 return 0;
2055}
2056EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2057
Tejun Heo80a761f2009-07-03 17:48:17 +09002058/**
2059 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2060 * @rq: request to examine
2061 *
2062 * Description:
2063 * A request could be merge of IOs which require different failure
2064 * handling. This function determines the number of bytes which
2065 * can be failed from the beginning of the request without
2066 * crossing into area which need to be retried further.
2067 *
2068 * Return:
2069 * The number of bytes to fail.
2070 *
2071 * Context:
2072 * queue_lock must be held.
2073 */
2074unsigned int blk_rq_err_bytes(const struct request *rq)
2075{
2076 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2077 unsigned int bytes = 0;
2078 struct bio *bio;
2079
2080 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2081 return blk_rq_bytes(rq);
2082
2083 /*
2084 * Currently the only 'mixing' which can happen is between
2085 * different fastfail types. We can safely fail portions
2086 * which have all the failfast bits that the first one has -
2087 * the ones which are at least as eager to fail as the first
2088 * one.
2089 */
2090 for (bio = rq->bio; bio; bio = bio->bi_next) {
2091 if ((bio->bi_rw & ff) != ff)
2092 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002093 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002094 }
2095
2096 /* this could lead to infinite loop */
2097 BUG_ON(blk_rq_bytes(rq) && !bytes);
2098 return bytes;
2099}
2100EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2101
Jens Axboe320ae512013-10-24 09:20:05 +01002102void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002103{
Jens Axboec2553b52009-04-24 08:10:11 +02002104 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002105 const int rw = rq_data_dir(req);
2106 struct hd_struct *part;
2107 int cpu;
2108
2109 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002110 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002111 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2112 part_stat_unlock();
2113 }
2114}
2115
Jens Axboe320ae512013-10-24 09:20:05 +01002116void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002117{
Jens Axboebc58ba92009-01-23 10:54:44 +01002118 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002119 * Account IO completion. flush_rq isn't accounted as a
2120 * normal IO on queueing nor completion. Accounting the
2121 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002122 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002123 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002124 unsigned long duration = jiffies - req->start_time;
2125 const int rw = rq_data_dir(req);
2126 struct hd_struct *part;
2127 int cpu;
2128
2129 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002130 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002131
2132 part_stat_inc(cpu, part, ios[rw]);
2133 part_stat_add(cpu, part, ticks[rw], duration);
2134 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002135 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002136
Jens Axboe6c23a962011-01-07 08:43:37 +01002137 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002138 part_stat_unlock();
2139 }
2140}
2141
Lin Mingc8158812013-03-23 11:42:27 +08002142#ifdef CONFIG_PM_RUNTIME
2143/*
2144 * Don't process normal requests when queue is suspended
2145 * or in the process of suspending/resuming
2146 */
2147static struct request *blk_pm_peek_request(struct request_queue *q,
2148 struct request *rq)
2149{
2150 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2151 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2152 return NULL;
2153 else
2154 return rq;
2155}
2156#else
2157static inline struct request *blk_pm_peek_request(struct request_queue *q,
2158 struct request *rq)
2159{
2160 return rq;
2161}
2162#endif
2163
Jens Axboe320ae512013-10-24 09:20:05 +01002164void blk_account_io_start(struct request *rq, bool new_io)
2165{
2166 struct hd_struct *part;
2167 int rw = rq_data_dir(rq);
2168 int cpu;
2169
2170 if (!blk_do_io_stat(rq))
2171 return;
2172
2173 cpu = part_stat_lock();
2174
2175 if (!new_io) {
2176 part = rq->part;
2177 part_stat_inc(cpu, part, merges[rw]);
2178 } else {
2179 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2180 if (!hd_struct_try_get(part)) {
2181 /*
2182 * The partition is already being removed,
2183 * the request will be accounted on the disk only
2184 *
2185 * We take a reference on disk->part0 although that
2186 * partition will never be deleted, so we can treat
2187 * it as any other partition.
2188 */
2189 part = &rq->rq_disk->part0;
2190 hd_struct_get(part);
2191 }
2192 part_round_stats(cpu, part);
2193 part_inc_in_flight(part, rw);
2194 rq->part = part;
2195 }
2196
2197 part_stat_unlock();
2198}
2199
Tejun Heo53a08802008-12-03 12:41:26 +01002200/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002201 * blk_peek_request - peek at the top of a request queue
2202 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002203 *
2204 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002205 * Return the request at the top of @q. The returned request
2206 * should be started using blk_start_request() before LLD starts
2207 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002208 *
2209 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002210 * Pointer to the request at the top of @q if available. Null
2211 * otherwise.
2212 *
2213 * Context:
2214 * queue_lock must be held.
2215 */
2216struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002217{
2218 struct request *rq;
2219 int ret;
2220
2221 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002222
2223 rq = blk_pm_peek_request(q, rq);
2224 if (!rq)
2225 break;
2226
Tejun Heo158dbda2009-04-23 11:05:18 +09002227 if (!(rq->cmd_flags & REQ_STARTED)) {
2228 /*
2229 * This is the first time the device driver
2230 * sees this request (possibly after
2231 * requeueing). Notify IO scheduler.
2232 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002233 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002234 elv_activate_rq(q, rq);
2235
2236 /*
2237 * just mark as started even if we don't start
2238 * it, a request that has been delayed should
2239 * not be passed by new incoming requests
2240 */
2241 rq->cmd_flags |= REQ_STARTED;
2242 trace_block_rq_issue(q, rq);
2243 }
2244
2245 if (!q->boundary_rq || q->boundary_rq == rq) {
2246 q->end_sector = rq_end_sector(rq);
2247 q->boundary_rq = NULL;
2248 }
2249
2250 if (rq->cmd_flags & REQ_DONTPREP)
2251 break;
2252
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002253 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002254 /*
2255 * make sure space for the drain appears we
2256 * know we can do this because max_hw_segments
2257 * has been adjusted to be one fewer than the
2258 * device can handle
2259 */
2260 rq->nr_phys_segments++;
2261 }
2262
2263 if (!q->prep_rq_fn)
2264 break;
2265
2266 ret = q->prep_rq_fn(q, rq);
2267 if (ret == BLKPREP_OK) {
2268 break;
2269 } else if (ret == BLKPREP_DEFER) {
2270 /*
2271 * the request may have been (partially) prepped.
2272 * we need to keep this request in the front to
2273 * avoid resource deadlock. REQ_STARTED will
2274 * prevent other fs requests from passing this one.
2275 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002276 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002277 !(rq->cmd_flags & REQ_DONTPREP)) {
2278 /*
2279 * remove the space for the drain we added
2280 * so that we don't add it again
2281 */
2282 --rq->nr_phys_segments;
2283 }
2284
2285 rq = NULL;
2286 break;
2287 } else if (ret == BLKPREP_KILL) {
2288 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002289 /*
2290 * Mark this request as started so we don't trigger
2291 * any debug logic in the end I/O path.
2292 */
2293 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002294 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002295 } else {
2296 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2297 break;
2298 }
2299 }
2300
2301 return rq;
2302}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002303EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002304
Tejun Heo9934c8c2009-05-08 11:54:16 +09002305void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002306{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002307 struct request_queue *q = rq->q;
2308
Tejun Heo158dbda2009-04-23 11:05:18 +09002309 BUG_ON(list_empty(&rq->queuelist));
2310 BUG_ON(ELV_ON_HASH(rq));
2311
2312 list_del_init(&rq->queuelist);
2313
2314 /*
2315 * the time frame between a request being removed from the lists
2316 * and to it is freed is accounted as io that is in progress at
2317 * the driver side.
2318 */
Divyesh Shah91952912010-04-01 15:01:41 -07002319 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002320 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002321 set_io_start_time_ns(rq);
2322 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002323}
2324
Tejun Heo5efccd12009-04-23 11:05:18 +09002325/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002326 * blk_start_request - start request processing on the driver
2327 * @req: request to dequeue
2328 *
2329 * Description:
2330 * Dequeue @req and start timeout timer on it. This hands off the
2331 * request to the driver.
2332 *
2333 * Block internal functions which don't want to start timer should
2334 * call blk_dequeue_request().
2335 *
2336 * Context:
2337 * queue_lock must be held.
2338 */
2339void blk_start_request(struct request *req)
2340{
2341 blk_dequeue_request(req);
2342
2343 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002344 * We are now handing the request to the hardware, initialize
2345 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002346 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002347 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002348 if (unlikely(blk_bidi_rq(req)))
2349 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2350
Jeff Moyer4912aa62013-10-08 14:36:41 -04002351 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002352 blk_add_timer(req);
2353}
2354EXPORT_SYMBOL(blk_start_request);
2355
2356/**
2357 * blk_fetch_request - fetch a request from a request queue
2358 * @q: request queue to fetch a request from
2359 *
2360 * Description:
2361 * Return the request at the top of @q. The request is started on
2362 * return and LLD can start processing it immediately.
2363 *
2364 * Return:
2365 * Pointer to the request at the top of @q if available. Null
2366 * otherwise.
2367 *
2368 * Context:
2369 * queue_lock must be held.
2370 */
2371struct request *blk_fetch_request(struct request_queue *q)
2372{
2373 struct request *rq;
2374
2375 rq = blk_peek_request(q);
2376 if (rq)
2377 blk_start_request(rq);
2378 return rq;
2379}
2380EXPORT_SYMBOL(blk_fetch_request);
2381
2382/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002383 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002384 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002385 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002386 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002387 *
2388 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002389 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2390 * the request structure even if @req doesn't have leftover.
2391 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002392 *
2393 * This special helper function is only for request stacking drivers
2394 * (e.g. request-based dm) so that they can handle partial completion.
2395 * Actual device drivers should use blk_end_request instead.
2396 *
2397 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2398 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002399 *
2400 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002401 * %false - this request doesn't have any more data
2402 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002403 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002404bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002406 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Tejun Heo2e60e022009-04-23 11:05:18 +09002408 if (!req->bio)
2409 return false;
2410
Roman Penaf5040d2014-03-04 23:13:10 +09002411 trace_block_rq_complete(req->q, req, nr_bytes);
Jens Axboe2056a782006-03-23 20:00:26 +01002412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002414 * For fs requests, rq is just carrier of independent bio's
2415 * and each partial completion should be handled separately.
2416 * Reset per-request error on each partial completion.
2417 *
2418 * TODO: tj: This is too subtle. It would be better to let
2419 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002421 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 req->errors = 0;
2423
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002424 if (error && req->cmd_type == REQ_TYPE_FS &&
2425 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002426 char *error_type;
2427
2428 switch (error) {
2429 case -ENOLINK:
2430 error_type = "recoverable transport";
2431 break;
2432 case -EREMOTEIO:
2433 error_type = "critical target";
2434 break;
2435 case -EBADE:
2436 error_type = "critical nexus";
2437 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002438 case -ETIMEDOUT:
2439 error_type = "timeout";
2440 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002441 case -ENOSPC:
2442 error_type = "critical space allocation";
2443 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002444 case -ENODATA:
2445 error_type = "critical medium";
2446 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002447 case -EIO:
2448 default:
2449 error_type = "I/O";
2450 break;
2451 }
Yi Zou37d7b342012-08-30 16:26:25 -07002452 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2453 error_type, req->rq_disk ?
2454 req->rq_disk->disk_name : "?",
2455 (unsigned long long)blk_rq_pos(req));
2456
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
Jens Axboebc58ba92009-01-23 10:54:44 +01002459 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002460
Kent Overstreetf79ea412012-09-20 16:38:30 -07002461 total_bytes = 0;
2462 while (req->bio) {
2463 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002464 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Kent Overstreet4f024f32013-10-11 15:44:27 -07002466 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Kent Overstreetf79ea412012-09-20 16:38:30 -07002469 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Kent Overstreetf79ea412012-09-20 16:38:30 -07002471 total_bytes += bio_bytes;
2472 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473
Kent Overstreetf79ea412012-09-20 16:38:30 -07002474 if (!nr_bytes)
2475 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 }
2477
2478 /*
2479 * completely done
2480 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002481 if (!req->bio) {
2482 /*
2483 * Reset counters so that the request stacking driver
2484 * can find how many bytes remain in the request
2485 * later.
2486 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002487 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002488 return false;
2489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002491 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002492
2493 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002494 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002495 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002496
Tejun Heo80a761f2009-07-03 17:48:17 +09002497 /* mixed attributes always follow the first bio */
2498 if (req->cmd_flags & REQ_MIXED_MERGE) {
2499 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2500 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2501 }
2502
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002503 /*
2504 * If total number of sectors is less than the first segment
2505 * size, something has gone terribly wrong.
2506 */
2507 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002508 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002509 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002510 }
2511
2512 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002514
Tejun Heo2e60e022009-04-23 11:05:18 +09002515 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516}
Tejun Heo2e60e022009-04-23 11:05:18 +09002517EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Tejun Heo2e60e022009-04-23 11:05:18 +09002519static bool blk_update_bidi_request(struct request *rq, int error,
2520 unsigned int nr_bytes,
2521 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002522{
Tejun Heo2e60e022009-04-23 11:05:18 +09002523 if (blk_update_request(rq, error, nr_bytes))
2524 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002525
Tejun Heo2e60e022009-04-23 11:05:18 +09002526 /* Bidi request must be completed as a whole */
2527 if (unlikely(blk_bidi_rq(rq)) &&
2528 blk_update_request(rq->next_rq, error, bidi_bytes))
2529 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002530
Jens Axboee2e1a142010-06-09 10:42:09 +02002531 if (blk_queue_add_random(rq->q))
2532 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002533
2534 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535}
2536
James Bottomley28018c22010-07-01 19:49:17 +09002537/**
2538 * blk_unprep_request - unprepare a request
2539 * @req: the request
2540 *
2541 * This function makes a request ready for complete resubmission (or
2542 * completion). It happens only after all error handling is complete,
2543 * so represents the appropriate moment to deallocate any resources
2544 * that were allocated to the request in the prep_rq_fn. The queue
2545 * lock is held when calling this.
2546 */
2547void blk_unprep_request(struct request *req)
2548{
2549 struct request_queue *q = req->q;
2550
2551 req->cmd_flags &= ~REQ_DONTPREP;
2552 if (q->unprep_rq_fn)
2553 q->unprep_rq_fn(q, req);
2554}
2555EXPORT_SYMBOL_GPL(blk_unprep_request);
2556
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557/*
2558 * queue lock must be held
2559 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002560void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002562 if (blk_rq_tagged(req))
2563 blk_queue_end_tag(req->q, req);
2564
James Bottomleyba396a62009-05-27 14:17:08 +02002565 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002567 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002568 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Mike Andersone78042e2008-10-30 02:16:20 -07002570 blk_delete_timer(req);
2571
James Bottomley28018c22010-07-01 19:49:17 +09002572 if (req->cmd_flags & REQ_DONTPREP)
2573 blk_unprep_request(req);
2574
Jens Axboebc58ba92009-01-23 10:54:44 +01002575 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002578 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002579 else {
2580 if (blk_bidi_rq(req))
2581 __blk_put_request(req->next_rq->q, req->next_rq);
2582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 }
2585}
Christoph Hellwig12120072014-04-16 09:44:59 +02002586EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002588/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002589 * blk_end_bidi_request - Complete a bidi request
2590 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002591 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002592 * @nr_bytes: number of bytes to complete @rq
2593 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002594 *
2595 * Description:
2596 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002597 * Drivers that supports bidi can safely call this member for any
2598 * type of request, bidi or uni. In the later case @bidi_bytes is
2599 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002600 *
2601 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002602 * %false - we are done with this request
2603 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002604 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002605static bool blk_end_bidi_request(struct request *rq, int error,
2606 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002607{
2608 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002609 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002610
Tejun Heo2e60e022009-04-23 11:05:18 +09002611 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2612 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002613
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002614 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002615 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002616 spin_unlock_irqrestore(q->queue_lock, flags);
2617
Tejun Heo2e60e022009-04-23 11:05:18 +09002618 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002619}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002620
2621/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002622 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2623 * @rq: the request to complete
2624 * @error: %0 for success, < %0 for error
2625 * @nr_bytes: number of bytes to complete @rq
2626 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002627 *
2628 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002629 * Identical to blk_end_bidi_request() except that queue lock is
2630 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002631 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002632 * Return:
2633 * %false - we are done with this request
2634 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002635 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002636bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002637 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002638{
Tejun Heo2e60e022009-04-23 11:05:18 +09002639 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2640 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002641
Tejun Heo2e60e022009-04-23 11:05:18 +09002642 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002643
Tejun Heo2e60e022009-04-23 11:05:18 +09002644 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002645}
2646
2647/**
2648 * blk_end_request - Helper function for drivers to complete the request.
2649 * @rq: the request being processed
2650 * @error: %0 for success, < %0 for error
2651 * @nr_bytes: number of bytes to complete
2652 *
2653 * Description:
2654 * Ends I/O on a number of bytes attached to @rq.
2655 * If @rq has leftover, sets it up for the next range of segments.
2656 *
2657 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002658 * %false - we are done with this request
2659 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002660 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002661bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002662{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002663 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002664}
Jens Axboe56ad1742009-07-28 22:11:24 +02002665EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002666
2667/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002668 * blk_end_request_all - Helper function for drives to finish the request.
2669 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002670 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002671 *
2672 * Description:
2673 * Completely finish @rq.
2674 */
2675void blk_end_request_all(struct request *rq, int error)
2676{
2677 bool pending;
2678 unsigned int bidi_bytes = 0;
2679
2680 if (unlikely(blk_bidi_rq(rq)))
2681 bidi_bytes = blk_rq_bytes(rq->next_rq);
2682
2683 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2684 BUG_ON(pending);
2685}
Jens Axboe56ad1742009-07-28 22:11:24 +02002686EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002687
2688/**
2689 * blk_end_request_cur - Helper function to finish the current request chunk.
2690 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002691 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002692 *
2693 * Description:
2694 * Complete the current consecutively mapped chunk from @rq.
2695 *
2696 * Return:
2697 * %false - we are done with this request
2698 * %true - still buffers pending for this request
2699 */
2700bool blk_end_request_cur(struct request *rq, int error)
2701{
2702 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2703}
Jens Axboe56ad1742009-07-28 22:11:24 +02002704EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002705
2706/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002707 * blk_end_request_err - Finish a request till the next failure boundary.
2708 * @rq: the request to finish till the next failure boundary for
2709 * @error: must be negative errno
2710 *
2711 * Description:
2712 * Complete @rq till the next failure boundary.
2713 *
2714 * Return:
2715 * %false - we are done with this request
2716 * %true - still buffers pending for this request
2717 */
2718bool blk_end_request_err(struct request *rq, int error)
2719{
2720 WARN_ON(error >= 0);
2721 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2722}
2723EXPORT_SYMBOL_GPL(blk_end_request_err);
2724
2725/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002726 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002727 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002728 * @error: %0 for success, < %0 for error
2729 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002730 *
2731 * Description:
2732 * Must be called with queue lock held unlike blk_end_request().
2733 *
2734 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002735 * %false - we are done with this request
2736 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002737 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002738bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002739{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002740 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002741}
Jens Axboe56ad1742009-07-28 22:11:24 +02002742EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002743
2744/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002745 * __blk_end_request_all - Helper function for drives to finish the request.
2746 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002747 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002748 *
2749 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002750 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002751 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002752void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002753{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002754 bool pending;
2755 unsigned int bidi_bytes = 0;
2756
2757 if (unlikely(blk_bidi_rq(rq)))
2758 bidi_bytes = blk_rq_bytes(rq->next_rq);
2759
2760 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2761 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002762}
Jens Axboe56ad1742009-07-28 22:11:24 +02002763EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002764
2765/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002766 * __blk_end_request_cur - Helper function to finish the current request chunk.
2767 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002768 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002769 *
2770 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002771 * Complete the current consecutively mapped chunk from @rq. Must
2772 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002773 *
2774 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
2777 */
2778bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002779{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002780 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002781}
Jens Axboe56ad1742009-07-28 22:11:24 +02002782EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002783
Tejun Heo80a761f2009-07-03 17:48:17 +09002784/**
2785 * __blk_end_request_err - Finish a request till the next failure boundary.
2786 * @rq: the request to finish till the next failure boundary for
2787 * @error: must be negative errno
2788 *
2789 * Description:
2790 * Complete @rq till the next failure boundary. Must be called
2791 * with queue lock held.
2792 *
2793 * Return:
2794 * %false - we are done with this request
2795 * %true - still buffers pending for this request
2796 */
2797bool __blk_end_request_err(struct request *rq, int error)
2798{
2799 WARN_ON(error >= 0);
2800 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2801}
2802EXPORT_SYMBOL_GPL(__blk_end_request_err);
2803
Jens Axboe86db1e22008-01-29 14:53:40 +01002804void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2805 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002807 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002808 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
Jens Axboeb4f42e22014-04-10 09:46:28 -06002810 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002811 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002812
Kent Overstreet4f024f32013-10-11 15:44:27 -07002813 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
NeilBrown66846572007-08-16 13:31:28 +02002816 if (bio->bi_bdev)
2817 rq->rq_disk = bio->bi_bdev->bd_disk;
2818}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002820#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2821/**
2822 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2823 * @rq: the request to be flushed
2824 *
2825 * Description:
2826 * Flush all pages in @rq.
2827 */
2828void rq_flush_dcache_pages(struct request *rq)
2829{
2830 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002831 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002832
2833 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002834 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002835}
2836EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2837#endif
2838
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002839/**
2840 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2841 * @q : the queue of the device being checked
2842 *
2843 * Description:
2844 * Check if underlying low-level drivers of a device are busy.
2845 * If the drivers want to export their busy state, they must set own
2846 * exporting function using blk_queue_lld_busy() first.
2847 *
2848 * Basically, this function is used only by request stacking drivers
2849 * to stop dispatching requests to underlying devices when underlying
2850 * devices are busy. This behavior helps more I/O merging on the queue
2851 * of the request stacking driver and prevents I/O throughput regression
2852 * on burst I/O load.
2853 *
2854 * Return:
2855 * 0 - Not busy (The request stacking driver should dispatch request)
2856 * 1 - Busy (The request stacking driver should stop dispatching request)
2857 */
2858int blk_lld_busy(struct request_queue *q)
2859{
2860 if (q->lld_busy_fn)
2861 return q->lld_busy_fn(q);
2862
2863 return 0;
2864}
2865EXPORT_SYMBOL_GPL(blk_lld_busy);
2866
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002867/**
2868 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2869 * @rq: the clone request to be cleaned up
2870 *
2871 * Description:
2872 * Free all bios in @rq for a cloned request.
2873 */
2874void blk_rq_unprep_clone(struct request *rq)
2875{
2876 struct bio *bio;
2877
2878 while ((bio = rq->bio) != NULL) {
2879 rq->bio = bio->bi_next;
2880
2881 bio_put(bio);
2882 }
2883}
2884EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2885
2886/*
2887 * Copy attributes of the original request to the clone request.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002888 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002889 */
2890static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2891{
2892 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002893 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002894 dst->cmd_type = src->cmd_type;
2895 dst->__sector = blk_rq_pos(src);
2896 dst->__data_len = blk_rq_bytes(src);
2897 dst->nr_phys_segments = src->nr_phys_segments;
2898 dst->ioprio = src->ioprio;
2899 dst->extra_len = src->extra_len;
2900}
2901
2902/**
2903 * blk_rq_prep_clone - Helper function to setup clone request
2904 * @rq: the request to be setup
2905 * @rq_src: original request to be cloned
2906 * @bs: bio_set that bios for clone are allocated from
2907 * @gfp_mask: memory allocation mask for bio
2908 * @bio_ctr: setup function to be called for each clone bio.
2909 * Returns %0 for success, non %0 for failure.
2910 * @data: private data to be passed to @bio_ctr
2911 *
2912 * Description:
2913 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
Jens Axboeb4f42e22014-04-10 09:46:28 -06002914 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002915 * are not copied, and copying such parts is the caller's responsibility.
2916 * Also, pages which the original bios are pointing to are not copied
2917 * and the cloned bios just point same pages.
2918 * So cloned bios must be completed before original bios, which means
2919 * the caller must complete @rq before @rq_src.
2920 */
2921int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2922 struct bio_set *bs, gfp_t gfp_mask,
2923 int (*bio_ctr)(struct bio *, struct bio *, void *),
2924 void *data)
2925{
2926 struct bio *bio, *bio_src;
2927
2928 if (!bs)
2929 bs = fs_bio_set;
2930
2931 blk_rq_init(NULL, rq);
2932
2933 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002934 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002935 if (!bio)
2936 goto free_and_out;
2937
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002938 if (bio_ctr && bio_ctr(bio, bio_src, data))
2939 goto free_and_out;
2940
2941 if (rq->bio) {
2942 rq->biotail->bi_next = bio;
2943 rq->biotail = bio;
2944 } else
2945 rq->bio = rq->biotail = bio;
2946 }
2947
2948 __blk_rq_prep_clone(rq, rq_src);
2949
2950 return 0;
2951
2952free_and_out:
2953 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002954 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002955 blk_rq_unprep_clone(rq);
2956
2957 return -ENOMEM;
2958}
2959EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2960
Jens Axboe59c3d452014-04-08 09:15:35 -06002961int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962{
2963 return queue_work(kblockd_workqueue, work);
2964}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965EXPORT_SYMBOL(kblockd_schedule_work);
2966
Jens Axboe59c3d452014-04-08 09:15:35 -06002967int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2968 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04002969{
2970 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2971}
2972EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2973
Jens Axboe8ab14592014-04-08 09:17:40 -06002974int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2975 unsigned long delay)
2976{
2977 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2978}
2979EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2980
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002981/**
2982 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2983 * @plug: The &struct blk_plug that needs to be initialized
2984 *
2985 * Description:
2986 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2987 * pending I/O should the task end up blocking between blk_start_plug() and
2988 * blk_finish_plug(). This is important from a performance perspective, but
2989 * also ensures that we don't deadlock. For instance, if the task is blocking
2990 * for a memory allocation, memory reclaim could end up wanting to free a
2991 * page belonging to that request that is currently residing in our private
2992 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2993 * this kind of deadlock.
2994 */
Jens Axboe73c10102011-03-08 13:19:51 +01002995void blk_start_plug(struct blk_plug *plug)
2996{
2997 struct task_struct *tsk = current;
2998
Jens Axboe73c10102011-03-08 13:19:51 +01002999 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003000 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003001 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003002
3003 /*
3004 * If this is a nested plug, don't actually assign it. It will be
3005 * flushed on its own.
3006 */
3007 if (!tsk->plug) {
3008 /*
3009 * Store ordering should not be needed here, since a potential
3010 * preempt will imply a full memory barrier
3011 */
3012 tsk->plug = plug;
3013 }
3014}
3015EXPORT_SYMBOL(blk_start_plug);
3016
3017static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3018{
3019 struct request *rqa = container_of(a, struct request, queuelist);
3020 struct request *rqb = container_of(b, struct request, queuelist);
3021
Jianpeng Ma975927b2012-10-25 21:58:17 +02003022 return !(rqa->q < rqb->q ||
3023 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003024}
3025
Jens Axboe49cac012011-04-16 13:51:05 +02003026/*
3027 * If 'from_schedule' is true, then postpone the dispatch of requests
3028 * until a safe kblockd context. We due this to avoid accidental big
3029 * additional stack usage in driver dispatch, in places where the originally
3030 * plugger did not intend it.
3031 */
Jens Axboef6603782011-04-15 15:49:07 +02003032static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003033 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003034 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003035{
Jens Axboe49cac012011-04-16 13:51:05 +02003036 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003037
Bart Van Assche70460572012-11-28 13:45:56 +01003038 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003039 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003040 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003041 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003042 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003043}
3044
NeilBrown74018dc2012-07-31 09:08:15 +02003045static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003046{
3047 LIST_HEAD(callbacks);
3048
Shaohua Li2a7d5552012-07-31 09:08:15 +02003049 while (!list_empty(&plug->cb_list)) {
3050 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003051
Shaohua Li2a7d5552012-07-31 09:08:15 +02003052 while (!list_empty(&callbacks)) {
3053 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003054 struct blk_plug_cb,
3055 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003056 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003057 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003058 }
NeilBrown048c9372011-04-18 09:52:22 +02003059 }
3060}
3061
NeilBrown9cbb1752012-07-31 09:08:14 +02003062struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3063 int size)
3064{
3065 struct blk_plug *plug = current->plug;
3066 struct blk_plug_cb *cb;
3067
3068 if (!plug)
3069 return NULL;
3070
3071 list_for_each_entry(cb, &plug->cb_list, list)
3072 if (cb->callback == unplug && cb->data == data)
3073 return cb;
3074
3075 /* Not currently on the callback list */
3076 BUG_ON(size < sizeof(*cb));
3077 cb = kzalloc(size, GFP_ATOMIC);
3078 if (cb) {
3079 cb->data = data;
3080 cb->callback = unplug;
3081 list_add(&cb->list, &plug->cb_list);
3082 }
3083 return cb;
3084}
3085EXPORT_SYMBOL(blk_check_plugged);
3086
Jens Axboe49cac012011-04-16 13:51:05 +02003087void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003088{
3089 struct request_queue *q;
3090 unsigned long flags;
3091 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003092 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003093 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003094
NeilBrown74018dc2012-07-31 09:08:15 +02003095 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003096
3097 if (!list_empty(&plug->mq_list))
3098 blk_mq_flush_plug_list(plug, from_schedule);
3099
Jens Axboe73c10102011-03-08 13:19:51 +01003100 if (list_empty(&plug->list))
3101 return;
3102
NeilBrown109b8122011-04-11 14:13:10 +02003103 list_splice_init(&plug->list, &list);
3104
Jianpeng Ma422765c2013-01-11 14:46:09 +01003105 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003106
3107 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003108 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003109
3110 /*
3111 * Save and disable interrupts here, to avoid doing it for every
3112 * queue lock we have to take.
3113 */
Jens Axboe73c10102011-03-08 13:19:51 +01003114 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003115 while (!list_empty(&list)) {
3116 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003117 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003118 BUG_ON(!rq->q);
3119 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003120 /*
3121 * This drops the queue lock
3122 */
3123 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003124 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003125 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003126 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003127 spin_lock(q->queue_lock);
3128 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003129
3130 /*
3131 * Short-circuit if @q is dead
3132 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003133 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003134 __blk_end_request_all(rq, -ENODEV);
3135 continue;
3136 }
3137
Jens Axboe73c10102011-03-08 13:19:51 +01003138 /*
3139 * rq is already accounted, so use raw insert
3140 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003141 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3142 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3143 else
3144 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003145
3146 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003147 }
3148
Jens Axboe99e22592011-04-18 09:59:55 +02003149 /*
3150 * This drops the queue lock
3151 */
3152 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003153 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003154
Jens Axboe73c10102011-03-08 13:19:51 +01003155 local_irq_restore(flags);
3156}
Jens Axboe73c10102011-03-08 13:19:51 +01003157
3158void blk_finish_plug(struct blk_plug *plug)
3159{
Jens Axboef6603782011-04-15 15:49:07 +02003160 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003161
3162 if (plug == current->plug)
3163 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003164}
3165EXPORT_SYMBOL(blk_finish_plug);
3166
Lin Ming6c954662013-03-23 11:42:26 +08003167#ifdef CONFIG_PM_RUNTIME
3168/**
3169 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3170 * @q: the queue of the device
3171 * @dev: the device the queue belongs to
3172 *
3173 * Description:
3174 * Initialize runtime-PM-related fields for @q and start auto suspend for
3175 * @dev. Drivers that want to take advantage of request-based runtime PM
3176 * should call this function after @dev has been initialized, and its
3177 * request queue @q has been allocated, and runtime PM for it can not happen
3178 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3179 * cases, driver should call this function before any I/O has taken place.
3180 *
3181 * This function takes care of setting up using auto suspend for the device,
3182 * the autosuspend delay is set to -1 to make runtime suspend impossible
3183 * until an updated value is either set by user or by driver. Drivers do
3184 * not need to touch other autosuspend settings.
3185 *
3186 * The block layer runtime PM is request based, so only works for drivers
3187 * that use request as their IO unit instead of those directly use bio's.
3188 */
3189void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3190{
3191 q->dev = dev;
3192 q->rpm_status = RPM_ACTIVE;
3193 pm_runtime_set_autosuspend_delay(q->dev, -1);
3194 pm_runtime_use_autosuspend(q->dev);
3195}
3196EXPORT_SYMBOL(blk_pm_runtime_init);
3197
3198/**
3199 * blk_pre_runtime_suspend - Pre runtime suspend check
3200 * @q: the queue of the device
3201 *
3202 * Description:
3203 * This function will check if runtime suspend is allowed for the device
3204 * by examining if there are any requests pending in the queue. If there
3205 * are requests pending, the device can not be runtime suspended; otherwise,
3206 * the queue's status will be updated to SUSPENDING and the driver can
3207 * proceed to suspend the device.
3208 *
3209 * For the not allowed case, we mark last busy for the device so that
3210 * runtime PM core will try to autosuspend it some time later.
3211 *
3212 * This function should be called near the start of the device's
3213 * runtime_suspend callback.
3214 *
3215 * Return:
3216 * 0 - OK to runtime suspend the device
3217 * -EBUSY - Device should not be runtime suspended
3218 */
3219int blk_pre_runtime_suspend(struct request_queue *q)
3220{
3221 int ret = 0;
3222
3223 spin_lock_irq(q->queue_lock);
3224 if (q->nr_pending) {
3225 ret = -EBUSY;
3226 pm_runtime_mark_last_busy(q->dev);
3227 } else {
3228 q->rpm_status = RPM_SUSPENDING;
3229 }
3230 spin_unlock_irq(q->queue_lock);
3231 return ret;
3232}
3233EXPORT_SYMBOL(blk_pre_runtime_suspend);
3234
3235/**
3236 * blk_post_runtime_suspend - Post runtime suspend processing
3237 * @q: the queue of the device
3238 * @err: return value of the device's runtime_suspend function
3239 *
3240 * Description:
3241 * Update the queue's runtime status according to the return value of the
3242 * device's runtime suspend function and mark last busy for the device so
3243 * that PM core will try to auto suspend the device at a later time.
3244 *
3245 * This function should be called near the end of the device's
3246 * runtime_suspend callback.
3247 */
3248void blk_post_runtime_suspend(struct request_queue *q, int err)
3249{
3250 spin_lock_irq(q->queue_lock);
3251 if (!err) {
3252 q->rpm_status = RPM_SUSPENDED;
3253 } else {
3254 q->rpm_status = RPM_ACTIVE;
3255 pm_runtime_mark_last_busy(q->dev);
3256 }
3257 spin_unlock_irq(q->queue_lock);
3258}
3259EXPORT_SYMBOL(blk_post_runtime_suspend);
3260
3261/**
3262 * blk_pre_runtime_resume - Pre runtime resume processing
3263 * @q: the queue of the device
3264 *
3265 * Description:
3266 * Update the queue's runtime status to RESUMING in preparation for the
3267 * runtime resume of the device.
3268 *
3269 * This function should be called near the start of the device's
3270 * runtime_resume callback.
3271 */
3272void blk_pre_runtime_resume(struct request_queue *q)
3273{
3274 spin_lock_irq(q->queue_lock);
3275 q->rpm_status = RPM_RESUMING;
3276 spin_unlock_irq(q->queue_lock);
3277}
3278EXPORT_SYMBOL(blk_pre_runtime_resume);
3279
3280/**
3281 * blk_post_runtime_resume - Post runtime resume processing
3282 * @q: the queue of the device
3283 * @err: return value of the device's runtime_resume function
3284 *
3285 * Description:
3286 * Update the queue's runtime status according to the return value of the
3287 * device's runtime_resume function. If it is successfully resumed, process
3288 * the requests that are queued into the device's queue when it is resuming
3289 * and then mark last busy and initiate autosuspend for it.
3290 *
3291 * This function should be called near the end of the device's
3292 * runtime_resume callback.
3293 */
3294void blk_post_runtime_resume(struct request_queue *q, int err)
3295{
3296 spin_lock_irq(q->queue_lock);
3297 if (!err) {
3298 q->rpm_status = RPM_ACTIVE;
3299 __blk_run_queue(q);
3300 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003301 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003302 } else {
3303 q->rpm_status = RPM_SUSPENDED;
3304 }
3305 spin_unlock_irq(q->queue_lock);
3306}
3307EXPORT_SYMBOL(blk_post_runtime_resume);
3308#endif
3309
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310int __init blk_dev_init(void)
3311{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003312 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3313 sizeof(((struct request *)0)->cmd_flags));
3314
Tejun Heo89b90be2011-01-03 15:01:47 +01003315 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3316 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003317 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 if (!kblockd_workqueue)
3319 panic("Failed to create kblockd\n");
3320
3321 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003322 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323
Jens Axboe8324aa92008-01-29 14:51:59 +01003324 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003325 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 return 0;
3328}