blob: 853f92749202cbfe5b252d557f667b7f970ac872 [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);
NeilBrowncbae8d42012-12-14 20:49:27 +010046EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010047
Tejun Heoa73f7302011-12-14 00:33:37 +010048DEFINE_IDA(blk_queue_ida);
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * For the allocated request tables
52 */
Jens Axboe320ae512013-10-24 09:20:05 +010053struct kmem_cache *request_cachep = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * For queue allocation
57 */
Jens Axboe6728cb02008-01-31 13:03:55 +010058struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * Controlling structure to kblockd
62 */
Jens Axboeff856ba2006-01-09 16:02:34 +010063static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jens Axboe8324aa92008-01-29 14:51:59 +010065void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
67 int nr;
68
69 nr = q->nr_requests - (q->nr_requests / 8) + 1;
70 if (nr > q->nr_requests)
71 nr = q->nr_requests;
72 q->nr_congestion_on = nr;
73
74 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
75 if (nr < 1)
76 nr = 1;
77 q->nr_congestion_off = nr;
78}
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/**
81 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
82 * @bdev: device
83 *
84 * Locates the passed device's request queue and returns the address of its
85 * backing_dev_info
86 *
87 * Will return NULL if the request queue cannot be located.
88 */
89struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
90{
91 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +020092 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94 if (q)
95 ret = &q->backing_dev_info;
96 return ret;
97}
Linus Torvalds1da177e2005-04-16 15:20:36 -070098EXPORT_SYMBOL(blk_get_backing_dev_info);
99
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200100void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200102 memset(rq, 0, sizeof(*rq));
103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700105 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200106 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100107 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900108 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200109 INIT_HLIST_NODE(&rq->hash);
110 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200111 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800112 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100113 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900114 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700115 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100116 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200118EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
NeilBrown5bb23a62007-09-27 12:46:13 +0200120static void req_bio_endio(struct request *rq, struct bio *bio,
121 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100122{
Tejun Heo143a87f2011-01-25 12:43:52 +0100123 if (error)
124 clear_bit(BIO_UPTODATE, &bio->bi_flags);
125 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
126 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100127
Tejun Heo143a87f2011-01-25 12:43:52 +0100128 if (unlikely(rq->cmd_flags & REQ_QUIET))
129 set_bit(BIO_QUIET, &bio->bi_flags);
130
Kent Overstreetf79ea412012-09-20 16:38:30 -0700131 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100132
133 /* don't actually finish bio if it's part of flush sequence */
Kent Overstreet4f024f32013-10-11 15:44:27 -0700134 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Tejun Heo143a87f2011-01-25 12:43:52 +0100135 bio_endio(bio, error);
Tejun Heo797e7db2006-01-06 09:51:03 +0100136}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138void blk_dump_rq_flags(struct request *rq, char *msg)
139{
140 int bit;
141
Jens Axboe59533162013-05-23 12:25:08 +0200142 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200143 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200144 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Tejun Heo83096eb2009-05-07 22:24:39 +0900146 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
147 (unsigned long long)blk_rq_pos(rq),
148 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Tejun Heo731ec492009-04-23 11:05:20 +0900149 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900150 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200152 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100153 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200154 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 printk("%02x ", rq->cmd[bit]);
156 printk("\n");
157 }
158}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159EXPORT_SYMBOL(blk_dump_rq_flags);
160
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500161static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200162{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500163 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200164
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500165 q = container_of(work, struct request_queue, delay_work.work);
166 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200167 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500168 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500172 * blk_delay_queue - restart queueing after defined interval
173 * @q: The &struct request_queue in question
174 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 *
176 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500177 * Sometimes queueing needs to be postponed for a little while, to allow
178 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100179 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500180 */
181void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
Bart Van Assche70460572012-11-28 13:45:56 +0100183 if (likely(!blk_queue_dead(q)))
184 queue_delayed_work(kblockd_workqueue, &q->delay_work,
185 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500187EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/**
190 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200191 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 *
193 * Description:
194 * blk_start_queue() will clear the stop flag on the queue, and call
195 * the request_fn for the queue if it was in a stopped state when
196 * entered. Also see blk_stop_queue(). Queue lock must be held.
197 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200198void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200200 WARN_ON(!irqs_disabled());
201
Nick Piggin75ad23b2008-04-29 14:48:33 +0200202 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200203 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205EXPORT_SYMBOL(blk_start_queue);
206
207/**
208 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200209 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 *
211 * Description:
212 * The Linux block layer assumes that a block driver will consume all
213 * entries on the request queue when the request_fn strategy is called.
214 * Often this will not happen, because of hardware limitations (queue
215 * depth settings). If a device driver gets a 'queue full' response,
216 * or if it simply chooses not to queue more I/O at one point, it can
217 * call this function to prevent the request_fn from being called until
218 * the driver has signalled it's ready to go again. This happens by calling
219 * blk_start_queue() to restart queue operations. Queue lock must be held.
220 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200221void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Tejun Heo136b5722012-08-21 13:18:24 -0700223 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200224 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226EXPORT_SYMBOL(blk_stop_queue);
227
228/**
229 * blk_sync_queue - cancel any pending callbacks on a queue
230 * @q: the queue
231 *
232 * Description:
233 * The block layer may perform asynchronous callback activity
234 * on a queue, such as calling the unplug function after a timeout.
235 * A block device may call blk_sync_queue to ensure that any
236 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200237 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 * that its ->make_request_fn will not re-add plugging prior to calling
239 * this function.
240 *
Vivek Goyalda527772011-03-02 19:05:33 -0500241 * This function does not cancel any asynchronous activity arising
242 * out of elevator or throttling code. That would require elevaotor_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800243 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500244 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 */
246void blk_sync_queue(struct request_queue *q)
247{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100248 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800249
250 if (q->mq_ops) {
251 struct blk_mq_hw_ctx *hctx;
252 int i;
253
254 queue_for_each_hw_ctx(q, hctx, i)
255 cancel_delayed_work_sync(&hctx->delayed_work);
256 } else {
257 cancel_delayed_work_sync(&q->delay_work);
258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260EXPORT_SYMBOL(blk_sync_queue);
261
262/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100263 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
264 * @q: The queue to run
265 *
266 * Description:
267 * Invoke request handling on a queue if there are any pending requests.
268 * May be used to restart request handling after a request has completed.
269 * This variant runs the queue whether or not the queue has been
270 * stopped. Must be called with the queue lock held and interrupts
271 * disabled. See also @blk_run_queue.
272 */
273inline void __blk_run_queue_uncond(struct request_queue *q)
274{
275 if (unlikely(blk_queue_dead(q)))
276 return;
277
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100278 /*
279 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
280 * the queue lock internally. As a result multiple threads may be
281 * running such a request function concurrently. Keep track of the
282 * number of active request_fn invocations such that blk_drain_queue()
283 * can wait until all these request_fn calls have finished.
284 */
285 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100286 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100287 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100288}
289
290/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200291 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200293 *
294 * Description:
295 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200296 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200298void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Tejun Heoa538cd02009-04-23 11:05:17 +0900300 if (unlikely(blk_queue_stopped(q)))
301 return;
302
Bart Van Asschec246e802012-12-06 14:32:01 +0100303 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200304}
305EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200306
Nick Piggin75ad23b2008-04-29 14:48:33 +0200307/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
310 *
311 * Description:
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100313 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200314 */
315void blk_run_queue_async(struct request_queue *q)
316{
Bart Van Assche70460572012-11-28 13:45:56 +0100317 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700318 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200319}
Jens Axboec21e6be2011-04-19 13:32:46 +0200320EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200321
322/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200323 * blk_run_queue - run a single device queue
324 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200325 *
326 * Description:
327 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900328 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200329 */
330void blk_run_queue(struct request_queue *q)
331{
332 unsigned long flags;
333
334 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200335 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 spin_unlock_irqrestore(q->queue_lock, flags);
337}
338EXPORT_SYMBOL(blk_run_queue);
339
Jens Axboe165125e2007-07-24 09:28:11 +0200340void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500341{
342 kobject_put(&q->kobj);
343}
Jens Axboed86e0e82011-05-27 07:44:43 +0200344EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500345
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200346/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100347 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200348 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200349 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200350 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200351 * Drain requests from @q. If @drain_all is set, all requests are drained.
352 * If not, only ELVPRIV requests are drained. The caller is responsible
353 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200354 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100355static void __blk_drain_queue(struct request_queue *q, bool drain_all)
356 __releases(q->queue_lock)
357 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200358{
Asias He458f27a2012-06-15 08:45:25 +0200359 int i;
360
Bart Van Assche807592a2012-11-28 13:43:38 +0100361 lockdep_assert_held(q->queue_lock);
362
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200363 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100364 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200365
Tejun Heob855b042012-03-06 21:24:55 +0100366 /*
367 * The caller might be trying to drain @q before its
368 * elevator is initialized.
369 */
370 if (q->elevator)
371 elv_drain_elevator(q);
372
Tejun Heo5efd6112012-03-05 13:15:12 -0800373 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200374
Tejun Heo4eabc942011-12-15 20:03:04 +0100375 /*
376 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100377 * driver init after queue creation or is not yet fully
378 * active yet. Some drivers (e.g. fd and loop) get unhappy
379 * in such cases. Kick queue iff dispatch queue has
380 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100381 */
Tejun Heob855b042012-03-06 21:24:55 +0100382 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100383 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200384
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700385 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100386 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100387
388 /*
389 * Unfortunately, requests are queued at and tracked from
390 * multiple places and there's no single counter which can
391 * be drained. Check all the queues and counters.
392 */
393 if (drain_all) {
394 drain |= !list_empty(&q->queue_head);
395 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700396 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100397 drain |= q->in_flight[i];
398 drain |= !list_empty(&q->flush_queue[i]);
399 }
400 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200401
Tejun Heo481a7d62011-12-14 00:33:37 +0100402 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200403 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100404
405 spin_unlock_irq(q->queue_lock);
406
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200407 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100408
409 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200410 }
Asias He458f27a2012-06-15 08:45:25 +0200411
412 /*
413 * With queue marked dead, any woken up waiter will fail the
414 * allocation path, so the wakeup chaining is lost and we're
415 * left with hung waiters. We need to wake up those waiters.
416 */
417 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700418 struct request_list *rl;
419
Tejun Heoa0516612012-06-26 15:05:44 -0700420 blk_queue_for_each_rl(rl, q)
421 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
422 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200423 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200424}
425
Tejun Heoc9a929d2011-10-19 14:42:16 +0200426/**
Tejun Heod7325802012-03-05 13:14:58 -0800427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
429 *
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800432 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800435 */
436void blk_queue_bypass_start(struct request_queue *q)
437{
Tejun Heob82d4b12012-04-13 13:11:31 -0700438 bool drain;
439
Tejun Heod7325802012-03-05 13:14:58 -0800440 spin_lock_irq(q->queue_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700441 drain = !q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800442 queue_flag_set(QUEUE_FLAG_BYPASS, q);
443 spin_unlock_irq(q->queue_lock);
444
Tejun Heob82d4b12012-04-13 13:11:31 -0700445 if (drain) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100446 spin_lock_irq(q->queue_lock);
447 __blk_drain_queue(q, false);
448 spin_unlock_irq(q->queue_lock);
449
Tejun Heob82d4b12012-04-13 13:11:31 -0700450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
451 synchronize_rcu();
452 }
Tejun Heod7325802012-03-05 13:14:58 -0800453}
454EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
455
456/**
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
459 *
460 * Leave bypass mode and restore the normal queueing behavior.
461 */
462void blk_queue_bypass_end(struct request_queue *q)
463{
464 spin_lock_irq(q->queue_lock);
465 if (!--q->bypass_depth)
466 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
467 WARN_ON_ONCE(q->bypass_depth < 0);
468 spin_unlock_irq(q->queue_lock);
469}
470EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
471
472/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
475 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100476 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
477 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500478 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100479void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500480{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200481 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700482
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100483 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500484 mutex_lock(&q->sysfs_lock);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100485 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200486 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800487
Tejun Heo80fd9972012-04-13 14:50:53 -0700488 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100489 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
496 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800497 q->bypass_depth++;
498 queue_flag_set(QUEUE_FLAG_BYPASS, q);
499
Tejun Heoc9a929d2011-10-19 14:42:16 +0200500 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100502 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200503 spin_unlock_irq(lock);
504 mutex_unlock(&q->sysfs_lock);
505
Bart Van Asschec246e802012-12-06 14:32:01 +0100506 /*
507 * Drain all requests queued before DYING marking. Set DEAD flag to
508 * prevent that q->request_fn() gets invoked after draining finished.
509 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800510 if (q->mq_ops) {
511 blk_mq_drain_queue(q);
512 spin_lock_irq(lock);
513 } else {
514 spin_lock_irq(lock);
515 __blk_drain_queue(q, true);
516 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100517 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100518 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200519
520 /* @q won't process any more request, flush async actions */
521 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
522 blk_sync_queue(q);
523
Asias He5e5cfac2012-05-24 23:28:52 +0800524 spin_lock_irq(lock);
525 if (q->queue_lock != &q->__queue_lock)
526 q->queue_lock = &q->__queue_lock;
527 spin_unlock_irq(lock);
528
Tejun Heoc9a929d2011-10-19 14:42:16 +0200529 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500530 blk_put_queue(q);
531}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532EXPORT_SYMBOL(blk_cleanup_queue);
533
Tejun Heo5b788ce2012-06-04 20:40:59 -0700534int blk_init_rl(struct request_list *rl, struct request_queue *q,
535 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400537 if (unlikely(rl->rq_pool))
538 return 0;
539
Tejun Heo5b788ce2012-06-04 20:40:59 -0700540 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200541 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
542 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200543 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
544 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Christoph Lameter19460892005-06-23 00:08:19 -0700546 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
Tejun Heoa91a5ac2012-06-04 20:40:53 -0700547 mempool_free_slab, request_cachep,
Tejun Heo5b788ce2012-06-04 20:40:59 -0700548 gfp_mask, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (!rl->rq_pool)
550 return -ENOMEM;
551
552 return 0;
553}
554
Tejun Heo5b788ce2012-06-04 20:40:59 -0700555void blk_exit_rl(struct request_list *rl)
556{
557 if (rl->rq_pool)
558 mempool_destroy(rl->rq_pool);
559}
560
Jens Axboe165125e2007-07-24 09:28:11 +0200561struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100563 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700564}
565EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Jens Axboe165125e2007-07-24 09:28:11 +0200567struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700568{
Jens Axboe165125e2007-07-24 09:28:11 +0200569 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700570 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700571
Jens Axboe8324aa92008-01-29 14:51:59 +0100572 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700573 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (!q)
575 return NULL;
576
Jens Axboe320ae512013-10-24 09:20:05 +0100577 if (percpu_counter_init(&q->mq_usage_counter, 0))
578 goto fail_q;
579
Dan Carpenter00380a42012-03-23 09:58:54 +0100580 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100581 if (q->id < 0)
Jens Axboe320ae512013-10-24 09:20:05 +0100582 goto fail_c;
Tejun Heoa73f7302011-12-14 00:33:37 +0100583
Jens Axboe0989a022009-06-12 14:42:56 +0200584 q->backing_dev_info.ra_pages =
585 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
586 q->backing_dev_info.state = 0;
587 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Jens Axboed9938312009-06-12 14:45:52 +0200588 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100589 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200590
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700591 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100592 if (err)
593 goto fail_id;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700594
Matthew Garrett31373d02010-04-06 14:25:14 +0200595 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
596 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700597 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100598 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700599 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100600 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800601#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800602 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800603#endif
Tejun Heoae1b1532011-01-25 12:43:54 +0100604 INIT_LIST_HEAD(&q->flush_queue[0]);
605 INIT_LIST_HEAD(&q->flush_queue[1]);
606 INIT_LIST_HEAD(&q->flush_data_in_flight);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500607 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500608
Jens Axboe8324aa92008-01-29 14:51:59 +0100609 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Al Viro483f4af2006-03-18 18:34:37 -0500611 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700612 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500613
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500614 /*
615 * By default initialize queue_lock to internal lock and driver can
616 * override it later if need be.
617 */
618 q->queue_lock = &q->__queue_lock;
619
Tejun Heob82d4b12012-04-13 13:11:31 -0700620 /*
621 * A queue starts its life with bypass turned on to avoid
622 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700623 * init. The initial bypass will be finished when the queue is
624 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700625 */
626 q->bypass_depth = 1;
627 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
628
Jens Axboe320ae512013-10-24 09:20:05 +0100629 init_waitqueue_head(&q->mq_freeze_wq);
630
Tejun Heo5efd6112012-03-05 13:15:12 -0800631 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400632 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800633
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100635
Mikulas Patockafff49962013-10-14 12:11:36 -0400636fail_bdi:
637 bdi_destroy(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100638fail_id:
639 ida_simple_remove(&blk_queue_ida, q->id);
Jens Axboe320ae512013-10-24 09:20:05 +0100640fail_c:
641 percpu_counter_destroy(&q->mq_usage_counter);
Tejun Heoa73f7302011-12-14 00:33:37 +0100642fail_q:
643 kmem_cache_free(blk_requestq_cachep, q);
644 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
Christoph Lameter19460892005-06-23 00:08:19 -0700646EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648/**
649 * blk_init_queue - prepare a request queue for use with a block device
650 * @rfn: The function to be called to process requests that have been
651 * placed on the queue.
652 * @lock: Request queue spin lock
653 *
654 * Description:
655 * If a block device wishes to use the standard request handling procedures,
656 * which sorts requests and coalesces adjacent requests, then it must
657 * call blk_init_queue(). The function @rfn will be called when there
658 * are requests on the queue that need to be processed. If the device
659 * supports plugging, then @rfn may not be called immediately when requests
660 * are available on the queue, but may be called at some time later instead.
661 * Plugged queues are generally unplugged when a buffer belonging to one
662 * of the requests on the queue is needed, or due to memory pressure.
663 *
664 * @rfn is not required, or even expected, to remove all requests off the
665 * queue, but only as many as it can handle at a time. If it does leave
666 * requests on the queue, it is responsible for arranging that the requests
667 * get dealt with eventually.
668 *
669 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200670 * request queue; this lock will be taken also from interrupt context, so irq
671 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200673 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 * it didn't succeed.
675 *
676 * Note:
677 * blk_init_queue() must be paired with a blk_cleanup_queue() call
678 * when the block device is deactivated (such as at module unload).
679 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700680
Jens Axboe165125e2007-07-24 09:28:11 +0200681struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100683 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700684}
685EXPORT_SYMBOL(blk_init_queue);
686
Jens Axboe165125e2007-07-24 09:28:11 +0200687struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700688blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
689{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600690 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600692 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
693 if (!uninit_q)
694 return NULL;
695
Christoph Hellwig18741982014-02-10 09:29:00 -0700696 uninit_q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
697 if (!uninit_q->flush_rq)
698 goto out_cleanup_queue;
699
Mike Snitzer51514122011-11-23 10:59:13 +0100700 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600701 if (!q)
Christoph Hellwig18741982014-02-10 09:29:00 -0700702 goto out_free_flush_rq;
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600703 return q;
Christoph Hellwig18741982014-02-10 09:29:00 -0700704
705out_free_flush_rq:
706 kfree(uninit_q->flush_rq);
707out_cleanup_queue:
708 blk_cleanup_queue(uninit_q);
709 return NULL;
Mike Snitzer01effb02010-05-11 08:57:42 +0200710}
711EXPORT_SYMBOL(blk_init_queue_node);
712
713struct request_queue *
714blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
715 spinlock_t *lock)
716{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (!q)
718 return NULL;
719
Tejun Heoa0516612012-06-26 15:05:44 -0700720 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Al Viro8669aaf2006-03-18 13:50:00 -0500721 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900725 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700726 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500727
728 /* Override internal queue lock with supplied lock pointer */
729 if (lock)
730 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Jens Axboef3b144a2009-03-06 08:48:33 +0100732 /*
733 * This also sets hw/phys segments, boundary and size
734 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200735 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Alan Stern44ec9542007-02-20 11:01:57 -0500737 q->sg_reserved_size = INT_MAX;
738
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600739 /* Protect q->elevator from elevator_change */
740 mutex_lock(&q->sysfs_lock);
741
Tejun Heob82d4b12012-04-13 13:11:31 -0700742 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600743 if (elevator_init(q, NULL)) {
744 mutex_unlock(&q->sysfs_lock);
Tejun Heob82d4b12012-04-13 13:11:31 -0700745 return NULL;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600746 }
747
748 mutex_unlock(&q->sysfs_lock);
749
Tejun Heob82d4b12012-04-13 13:11:31 -0700750 return q;
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100851 * Determine if elevator data should be initialized when allocating the
852 * request associated with @bio.
853 */
854static bool blk_rq_should_init_elevator(struct bio *bio)
855{
856 if (!bio)
857 return true;
858
859 /*
860 * Flush requests do not use the elevator so skip initialization.
861 * This allows a request to share the flush and elevator data.
862 */
863 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
864 return false;
865
866 return true;
867}
868
Tejun Heoda8303c2011-10-19 14:33:05 +0200869/**
Tejun Heo852c7882012-03-05 13:15:27 -0800870 * rq_ioc - determine io_context for request allocation
871 * @bio: request being allocated is for this bio (can be %NULL)
872 *
873 * Determine io_context to use for request allocation for @bio. May return
874 * %NULL if %current->io_context doesn't exist.
875 */
876static struct io_context *rq_ioc(struct bio *bio)
877{
878#ifdef CONFIG_BLK_CGROUP
879 if (bio && bio->bi_ioc)
880 return bio->bi_ioc;
881#endif
882 return current->io_context;
883}
884
885/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700886 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700887 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200888 * @rw_flags: RW and SYNC flags
889 * @bio: bio to allocate request for (can be %NULL)
890 * @gfp_mask: allocation mask
891 *
892 * Get a free request from @q. This function may fail under memory
893 * pressure or if @q is dead.
894 *
895 * Must be callled with @q->queue_lock held and,
896 * Returns %NULL on failure, with @q->queue_lock held.
897 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700899static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700900 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700902 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -0800903 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700904 struct elevator_type *et = q->elevator->type;
905 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100906 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200907 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200908 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100910 if (unlikely(blk_queue_dying(q)))
Tejun Heoda8303c2011-10-19 14:33:05 +0200911 return NULL;
912
Jens Axboe7749a8d2006-12-13 13:02:26 +0100913 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100914 if (may_queue == ELV_MQUEUE_NO)
915 goto rq_starved;
916
Jens Axboe1faa16d2009-04-06 14:48:01 +0200917 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
918 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +0100919 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100920 * The queue will fill after this allocation, so set
921 * it as full, and mark this process as "batching".
922 * This process will be allowed to complete a batch of
923 * requests, others will be blocked.
924 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700925 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100926 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -0700927 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100928 } else {
929 if (may_queue != ELV_MQUEUE_MUST
930 && !ioc_batching(q, ioc)) {
931 /*
932 * The queue is full and the allocating
933 * process is not a "batcher", and not
934 * exempted by the IO scheduler
935 */
Tejun Heob6792812012-03-05 13:15:23 -0800936 return NULL;
Jens Axboe88ee5ef2005-11-12 11:09:12 +0100937 }
938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
Tejun Heoa0516612012-06-26 15:05:44 -0700940 /*
941 * bdi isn't aware of blkcg yet. As all async IOs end up
942 * root blkcg anyway, just use root blkcg state.
943 */
944 if (rl == &q->root_rl)
945 blk_set_queue_congested(q, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
947
Jens Axboe082cf692005-06-28 16:35:11 +0200948 /*
949 * Only allow batching queuers to allocate up to 50% over the defined
950 * limit of requests, otherwise we could have thousands of requests
951 * allocated with any setting of ->nr_requests
952 */
Jens Axboe1faa16d2009-04-06 14:48:01 +0200953 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Tejun Heob6792812012-03-05 13:15:23 -0800954 return NULL;
Hugh Dickinsfd782a42005-06-29 15:15:40 +0100955
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700956 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200957 rl->count[is_sync]++;
958 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +0200959
Tejun Heof1f8cc92011-12-14 00:33:42 +0100960 /*
961 * Decide whether the new request will be managed by elevator. If
962 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
963 * prevent the current elevator from being destroyed until the new
964 * request is freed. This guarantees icq's won't be destroyed and
965 * makes creating new ones safe.
966 *
967 * Also, lookup icq while holding queue_lock. If it doesn't exist,
968 * it will be created after releasing queue_lock.
969 */
Tejun Heod7325802012-03-05 13:14:58 -0800970 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +0200971 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700972 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +0100973 if (et->icq_cache && ioc)
974 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100975 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200976
Jens Axboef253b862010-10-24 22:06:02 +0200977 if (blk_queue_io_stat(q))
978 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 spin_unlock_irq(q->queue_lock);
980
Tejun Heo29e2b092012-04-19 16:29:21 -0700981 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700982 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -0700983 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -0800984 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
Tejun Heo29e2b092012-04-19 16:29:21 -0700986 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -0700987 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -0700988 rq->cmd_flags = rw_flags | REQ_ALLOCED;
989
Tejun Heoaaf7c682012-04-19 16:29:22 -0700990 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -0700991 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -0700992 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700993 if (ioc)
994 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -0700995 if (!icq)
996 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -0700997 }
Tejun Heoaaf7c682012-04-19 16:29:22 -0700998
999 rq->elv.icq = icq;
1000 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1001 goto fail_elvpriv;
1002
1003 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001004 if (icq)
1005 get_io_context(icq->ioc);
1006 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001007out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001008 /*
1009 * ioc may be NULL here, and ioc_batching will be false. That's
1010 * OK, if the queue is under the request limit then requests need
1011 * not count toward the nr_batch_requests limit. There will always
1012 * be some limit enforced by BLK_BATCH_TIME.
1013 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (ioc_batching(q, ioc))
1015 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001016
Jens Axboe1faa16d2009-04-06 14:48:01 +02001017 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001019
Tejun Heoaaf7c682012-04-19 16:29:22 -07001020fail_elvpriv:
1021 /*
1022 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1023 * and may fail indefinitely under memory pressure and thus
1024 * shouldn't stall IO. Treat this request as !elvpriv. This will
1025 * disturb iosched and blkcg but weird is bettern than dead.
1026 */
1027 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1028 dev_name(q->backing_dev_info.dev));
1029
1030 rq->cmd_flags &= ~REQ_ELVPRIV;
1031 rq->elv.icq = NULL;
1032
1033 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001034 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001035 spin_unlock_irq(q->queue_lock);
1036 goto out;
1037
Tejun Heob6792812012-03-05 13:15:23 -08001038fail_alloc:
1039 /*
1040 * Allocation failed presumably due to memory. Undo anything we
1041 * might have messed up.
1042 *
1043 * Allocating task should really be put onto the front of the wait
1044 * queue, but this is pretty rare.
1045 */
1046 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001047 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001048
1049 /*
1050 * in the very unlikely event that allocation failed and no
1051 * requests for this direction was pending, mark us starved so that
1052 * freeing of a request in the other direction will notice
1053 * us. another possible fix would be to split the rq mempool into
1054 * READ and WRITE
1055 */
1056rq_starved:
1057 if (unlikely(rl->count[is_sync] == 0))
1058 rl->starved[is_sync] = 1;
1059 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060}
1061
Tejun Heoda8303c2011-10-19 14:33:05 +02001062/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001063 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001064 * @q: request_queue to allocate request from
1065 * @rw_flags: RW and SYNC flags
1066 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001067 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001068 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001069 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1070 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001071 *
1072 * Must be callled with @q->queue_lock held and,
1073 * Returns %NULL on failure, with @q->queue_lock held.
1074 * Returns !%NULL on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001076static struct request *get_request(struct request_queue *q, int rw_flags,
1077 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001079 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001080 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001081 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001083
1084 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001085retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001086 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001087 if (rq)
1088 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001090 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001091 blk_put_rl(rl);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001092 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -07001093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Tejun Heoa06e05e2012-06-04 20:40:55 -07001095 /* wait on @rl and retry */
1096 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1097 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001098
Tejun Heoa06e05e2012-06-04 20:40:55 -07001099 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Tejun Heoa06e05e2012-06-04 20:40:55 -07001101 spin_unlock_irq(q->queue_lock);
1102 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Tejun Heoa06e05e2012-06-04 20:40:55 -07001104 /*
1105 * After sleeping, we become a "batching" process and will be able
1106 * to allocate at least one request, and up to a big batch of them
1107 * for a small period time. See ioc_batching, ioc_set_batching
1108 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001109 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Tejun Heoa06e05e2012-06-04 20:40:55 -07001111 spin_lock_irq(q->queue_lock);
1112 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001113
Tejun Heoa06e05e2012-06-04 20:40:55 -07001114 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115}
1116
Jens Axboe320ae512013-10-24 09:20:05 +01001117static struct request *blk_old_get_request(struct request_queue *q, int rw,
1118 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119{
1120 struct request *rq;
1121
1122 BUG_ON(rw != READ && rw != WRITE);
1123
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001124 /* create ioc upfront */
1125 create_io_context(gfp_mask, q->node);
1126
Nick Piggind6344532005-06-28 20:45:14 -07001127 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001128 rq = get_request(q, rw, NULL, gfp_mask);
Tejun Heoda8303c2011-10-19 14:33:05 +02001129 if (!rq)
1130 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001131 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
1133 return rq;
1134}
Jens Axboe320ae512013-10-24 09:20:05 +01001135
1136struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1137{
1138 if (q->mq_ops)
Christoph Hellwig18741982014-02-10 09:29:00 -07001139 return blk_mq_alloc_request(q, rw, gfp_mask);
Jens Axboe320ae512013-10-24 09:20:05 +01001140 else
1141 return blk_old_get_request(q, rw, gfp_mask);
1142}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143EXPORT_SYMBOL(blk_get_request);
1144
1145/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001146 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001147 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001148 * @bio: The bio describing the memory mappings that will be submitted for IO.
1149 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001150 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001151 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001152 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1153 * type commands. Where the struct request needs to be farther initialized by
1154 * the caller. It is passed a &struct bio, which describes the memory info of
1155 * the I/O transfer.
1156 *
1157 * The caller of blk_make_request must make sure that bi_io_vec
1158 * are set to describe the memory buffers. That bio_data_dir() will return
1159 * the needed direction of the request. (And all bio's in the passed bio-chain
1160 * are properly set accordingly)
1161 *
1162 * If called under none-sleepable conditions, mapped bio buffers must not
1163 * need bouncing, by calling the appropriate masked or flagged allocator,
1164 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1165 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001166 *
1167 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1168 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1169 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1170 * completion of a bio that hasn't been submitted yet, thus resulting in a
1171 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1172 * of bio_alloc(), as that avoids the mempool deadlock.
1173 * If possible a big IO should be split into smaller parts when allocation
1174 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001175 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001176struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1177 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001178{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001179 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1180
1181 if (unlikely(!rq))
1182 return ERR_PTR(-ENOMEM);
1183
1184 for_each_bio(bio) {
1185 struct bio *bounce_bio = bio;
1186 int ret;
1187
1188 blk_queue_bounce(q, &bounce_bio);
1189 ret = blk_rq_append_bio(q, rq, bounce_bio);
1190 if (unlikely(ret)) {
1191 blk_put_request(rq);
1192 return ERR_PTR(ret);
1193 }
1194 }
1195
1196 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001197}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001198EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001199
1200/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 * blk_requeue_request - put a request back on queue
1202 * @q: request queue where request should be inserted
1203 * @rq: request to be inserted
1204 *
1205 * Description:
1206 * Drivers often keep queueing requests until the hardware cannot accept
1207 * more, when that condition happens we need to put the request back
1208 * on the queue. Must be called with queue lock held.
1209 */
Jens Axboe165125e2007-07-24 09:28:11 +02001210void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001212 blk_delete_timer(rq);
1213 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001214 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (blk_rq_tagged(rq))
1217 blk_queue_end_tag(q, rq);
1218
James Bottomleyba396a62009-05-27 14:17:08 +02001219 BUG_ON(blk_queued_rq(rq));
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 elv_requeue_request(q, rq);
1222}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223EXPORT_SYMBOL(blk_requeue_request);
1224
Jens Axboe73c10102011-03-08 13:19:51 +01001225static void add_acct_request(struct request_queue *q, struct request *rq,
1226 int where)
1227{
Jens Axboe320ae512013-10-24 09:20:05 +01001228 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001229 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001230}
1231
Tejun Heo074a7ac2008-08-25 19:56:14 +09001232static void part_round_stats_single(int cpu, struct hd_struct *part,
1233 unsigned long now)
1234{
1235 if (now == part->stamp)
1236 return;
1237
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001238 if (part_in_flight(part)) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001239 __part_stat_add(cpu, part, time_in_queue,
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02001240 part_in_flight(part) * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001241 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1242 }
1243 part->stamp = now;
1244}
1245
1246/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001247 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1248 * @cpu: cpu number for stats access
1249 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 *
1251 * The average IO queue length and utilisation statistics are maintained
1252 * by observing the current state of the queue length and the amount of
1253 * time it has been in this state for.
1254 *
1255 * Normally, that accounting is done on IO completion, but that can result
1256 * in more than a second's worth of IO being accounted for within any one
1257 * second, leading to >100% utilisation. To deal with that, we call this
1258 * function to do a round-off before returning the results when reading
1259 * /proc/diskstats. This accounts immediately for all queue usage up to
1260 * the current jiffies and restarts the counters again.
1261 */
Tejun Heoc9959052008-08-25 19:47:21 +09001262void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001263{
1264 unsigned long now = jiffies;
1265
Tejun Heo074a7ac2008-08-25 19:56:14 +09001266 if (part->partno)
1267 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1268 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001269}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001270EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001271
Lin Mingc8158812013-03-23 11:42:27 +08001272#ifdef CONFIG_PM_RUNTIME
1273static void blk_pm_put_request(struct request *rq)
1274{
1275 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1276 pm_runtime_mark_last_busy(rq->q->dev);
1277}
1278#else
1279static inline void blk_pm_put_request(struct request *rq) {}
1280#endif
1281
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282/*
1283 * queue lock must be held
1284 */
Jens Axboe165125e2007-07-24 09:28:11 +02001285void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (unlikely(!q))
1288 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001290 if (q->mq_ops) {
1291 blk_mq_free_request(req);
1292 return;
1293 }
1294
Lin Mingc8158812013-03-23 11:42:27 +08001295 blk_pm_put_request(req);
1296
Tejun Heo8922e162005-10-20 16:23:44 +02001297 elv_completed_request(q, req);
1298
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001299 /* this is a bio leak */
1300 WARN_ON(req->bio != NULL);
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 /*
1303 * Request may not have originated from ll_rw_blk. if not,
1304 * it didn't come out of our reserved rq pools
1305 */
Jens Axboe49171e52006-08-10 08:59:11 +02001306 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001307 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001308 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02001311 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Tejun Heoa0516612012-06-26 15:05:44 -07001313 blk_free_request(rl, req);
1314 freed_request(rl, flags);
1315 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
1317}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001318EXPORT_SYMBOL_GPL(__blk_put_request);
1319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320void blk_put_request(struct request *req)
1321{
Jens Axboe165125e2007-07-24 09:28:11 +02001322 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Jens Axboe320ae512013-10-24 09:20:05 +01001324 if (q->mq_ops)
1325 blk_mq_free_request(req);
1326 else {
1327 unsigned long flags;
1328
1329 spin_lock_irqsave(q->queue_lock, flags);
1330 __blk_put_request(q, req);
1331 spin_unlock_irqrestore(q->queue_lock, flags);
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334EXPORT_SYMBOL(blk_put_request);
1335
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001336/**
1337 * blk_add_request_payload - add a payload to a request
1338 * @rq: request to update
1339 * @page: page backing the payload
1340 * @len: length of the payload.
1341 *
1342 * This allows to later add a payload to an already submitted request by
1343 * a block driver. The driver needs to take care of freeing the payload
1344 * itself.
1345 *
1346 * Note that this is a quite horrible hack and nothing but handling of
1347 * discard requests should ever use it.
1348 */
1349void blk_add_request_payload(struct request *rq, struct page *page,
1350 unsigned int len)
1351{
1352 struct bio *bio = rq->bio;
1353
1354 bio->bi_io_vec->bv_page = page;
1355 bio->bi_io_vec->bv_offset = 0;
1356 bio->bi_io_vec->bv_len = len;
1357
Kent Overstreet4f024f32013-10-11 15:44:27 -07001358 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001359 bio->bi_vcnt = 1;
1360 bio->bi_phys_segments = 1;
1361
1362 rq->__data_len = rq->resid_len = len;
1363 rq->nr_phys_segments = 1;
1364 rq->buffer = bio_data(bio);
1365}
1366EXPORT_SYMBOL_GPL(blk_add_request_payload);
1367
Jens Axboe320ae512013-10-24 09:20:05 +01001368bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1369 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001370{
1371 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1372
Jens Axboe73c10102011-03-08 13:19:51 +01001373 if (!ll_back_merge_fn(q, req, bio))
1374 return false;
1375
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001376 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001377
1378 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1379 blk_rq_set_mixed_merge(req);
1380
1381 req->biotail->bi_next = bio;
1382 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001383 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001384 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1385
Jens Axboe320ae512013-10-24 09:20:05 +01001386 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001387 return true;
1388}
1389
Jens Axboe320ae512013-10-24 09:20:05 +01001390bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1391 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001392{
1393 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001394
Jens Axboe73c10102011-03-08 13:19:51 +01001395 if (!ll_front_merge_fn(q, req, bio))
1396 return false;
1397
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001398 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001399
1400 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1401 blk_rq_set_mixed_merge(req);
1402
Jens Axboe73c10102011-03-08 13:19:51 +01001403 bio->bi_next = req->bio;
1404 req->bio = bio;
1405
1406 /*
1407 * may not be valid. if the low level driver said
1408 * it didn't need a bounce buffer then it better
1409 * not touch req->buffer either...
1410 */
1411 req->buffer = bio_data(bio);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001412 req->__sector = bio->bi_iter.bi_sector;
1413 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001414 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1415
Jens Axboe320ae512013-10-24 09:20:05 +01001416 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001417 return true;
1418}
1419
Tejun Heobd87b582011-10-19 14:33:08 +02001420/**
Jens Axboe320ae512013-10-24 09:20:05 +01001421 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001422 * @q: request_queue new bio is being queued at
1423 * @bio: new bio being queued
1424 * @request_count: out parameter for number of traversed plugged requests
1425 *
1426 * Determine whether @bio being queued on @q can be merged with a request
1427 * on %current's plugged list. Returns %true if merge was successful,
1428 * otherwise %false.
1429 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001430 * Plugging coalesces IOs from the same issuer for the same purpose without
1431 * going through @q->queue_lock. As such it's more of an issuing mechanism
1432 * than scheduling, and the request, while may have elvpriv data, is not
1433 * added on the elevator at this point. In addition, we don't have
1434 * reliable access to the elevator outside queue lock. Only check basic
1435 * merging parameters without querying the elevator.
Jens Axboe73c10102011-03-08 13:19:51 +01001436 */
Jens Axboe320ae512013-10-24 09:20:05 +01001437bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1438 unsigned int *request_count)
Jens Axboe73c10102011-03-08 13:19:51 +01001439{
1440 struct blk_plug *plug;
1441 struct request *rq;
1442 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001443 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001444
Alireza Haghdoost23779fb2013-10-23 17:08:16 +01001445 if (blk_queue_nomerges(q))
1446 goto out;
1447
Tejun Heobd87b582011-10-19 14:33:08 +02001448 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001449 if (!plug)
1450 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001451 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001452
Shaohua Li92f399c2013-10-29 12:01:03 -06001453 if (q->mq_ops)
1454 plug_list = &plug->mq_list;
1455 else
1456 plug_list = &plug->list;
1457
1458 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001459 int el_ret;
1460
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001461 if (rq->q == q)
1462 (*request_count)++;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001463
Tejun Heo07c2bd32012-02-08 09:19:42 +01001464 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001465 continue;
1466
Tejun Heo050c8ea2012-02-08 09:19:38 +01001467 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001468 if (el_ret == ELEVATOR_BACK_MERGE) {
1469 ret = bio_attempt_back_merge(q, rq, bio);
1470 if (ret)
1471 break;
1472 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1473 ret = bio_attempt_front_merge(q, rq, bio);
1474 if (ret)
1475 break;
1476 }
1477 }
1478out:
1479 return ret;
1480}
1481
Jens Axboe86db1e22008-01-29 14:53:40 +01001482void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001483{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001484 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001485
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001486 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1487 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001488 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001489
Tejun Heo52d9e672006-01-06 09:49:58 +01001490 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001491 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001492 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001493 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001494}
1495
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001496void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001498 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001499 struct blk_plug *plug;
1500 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1501 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001502 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /*
1505 * low level driver can indicate that it wants pages above a
1506 * certain limit bounced to low memory (ie for highmem, or even
1507 * ISA dma in theory)
1508 */
1509 blk_queue_bounce(q, &bio);
1510
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001511 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1512 bio_endio(bio, -EIO);
1513 return;
1514 }
1515
Tejun Heo4fed9472010-09-03 11:56:17 +02001516 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001517 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001518 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001519 goto get_rq;
1520 }
1521
Jens Axboe73c10102011-03-08 13:19:51 +01001522 /*
1523 * Check if we can merge with the plugged list before grabbing
1524 * any locks.
1525 */
Jens Axboe320ae512013-10-24 09:20:05 +01001526 if (blk_attempt_plug_merge(q, bio, &request_count))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001527 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001528
1529 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001532 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001533 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001534 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001535 if (!attempt_back_merge(q, req))
1536 elv_merged_request(q, req, el_ret);
1537 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001538 }
Jens Axboe73c10102011-03-08 13:19:51 +01001539 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001540 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001541 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001542 if (!attempt_front_merge(q, req))
1543 elv_merged_request(q, req, el_ret);
1544 goto out_unlock;
1545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
1547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001549 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001550 * This sync check and mask will be re-done in init_request_from_bio(),
1551 * but we need to set it earlier to expose the sync flag to the
1552 * rq allocator and io schedulers.
1553 */
1554 rw_flags = bio_data_dir(bio);
1555 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001556 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001557
1558 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001559 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001560 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001561 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001562 req = get_request(q, rw_flags, bio, GFP_NOIO);
Tejun Heoda8303c2011-10-19 14:33:05 +02001563 if (unlikely(!req)) {
1564 bio_endio(bio, -ENODEV); /* @q is dead */
1565 goto out_unlock;
1566 }
Nick Piggind6344532005-06-28 20:45:14 -07001567
Nick Piggin450991b2005-06-28 20:45:13 -07001568 /*
1569 * After dropping the lock and possibly sleeping here, our request
1570 * may now be mergeable after it had proven unmergeable (above).
1571 * We don't worry about that case for efficiency. It won't happen
1572 * often, and the elevators are able to handle it.
1573 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001574 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Tao Ma9562ad92011-10-24 16:11:30 +02001576 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001577 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001578
Jens Axboe73c10102011-03-08 13:19:51 +01001579 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001580 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001581 /*
1582 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001583 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001584 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001585 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001586 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001587 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001588 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001589 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001590 trace_block_plug(q);
1591 }
Jens Axboe73c10102011-03-08 13:19:51 +01001592 }
Shaohua Lia6327162011-08-24 16:04:32 +02001593 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001594 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001595 } else {
1596 spin_lock_irq(q->queue_lock);
1597 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001598 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001599out_unlock:
1600 spin_unlock_irq(q->queue_lock);
1601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
Jens Axboec20e8de2011-09-12 12:03:37 +02001603EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605/*
1606 * If bio->bi_dev is a partition, remap the location
1607 */
1608static inline void blk_partition_remap(struct bio *bio)
1609{
1610 struct block_device *bdev = bio->bi_bdev;
1611
Jens Axboebf2de6f2007-09-27 13:01:25 +02001612 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001614
Kent Overstreet4f024f32013-10-11 15:44:27 -07001615 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001617
Mike Snitzerd07335e2010-11-16 12:52:38 +01001618 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1619 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001620 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
1622}
1623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624static void handle_bad_sector(struct bio *bio)
1625{
1626 char b[BDEVNAME_SIZE];
1627
1628 printk(KERN_INFO "attempt to access beyond end of device\n");
1629 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1630 bdevname(bio->bi_bdev, b),
1631 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001632 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001633 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
1635 set_bit(BIO_EOF, &bio->bi_flags);
1636}
1637
Akinobu Mitac17bb492006-12-08 02:39:46 -08001638#ifdef CONFIG_FAIL_MAKE_REQUEST
1639
1640static DECLARE_FAULT_ATTR(fail_make_request);
1641
1642static int __init setup_fail_make_request(char *str)
1643{
1644 return setup_fault_attr(&fail_make_request, str);
1645}
1646__setup("fail_make_request=", setup_fail_make_request);
1647
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001648static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001649{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001650 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001651}
1652
1653static int __init fail_make_request_debugfs(void)
1654{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001655 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1656 NULL, &fail_make_request);
1657
1658 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001659}
1660
1661late_initcall(fail_make_request_debugfs);
1662
1663#else /* CONFIG_FAIL_MAKE_REQUEST */
1664
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001665static inline bool should_fail_request(struct hd_struct *part,
1666 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001667{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001668 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001669}
1670
1671#endif /* CONFIG_FAIL_MAKE_REQUEST */
1672
Jens Axboec07e2b42007-07-18 13:27:58 +02001673/*
1674 * Check whether this bio extends beyond the end of the device.
1675 */
1676static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1677{
1678 sector_t maxsector;
1679
1680 if (!nr_sectors)
1681 return 0;
1682
1683 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001684 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001685 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001686 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001687
1688 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1689 /*
1690 * This may well happen - the kernel calls bread()
1691 * without checking the size of the device, e.g., when
1692 * mounting a device.
1693 */
1694 handle_bad_sector(bio);
1695 return 1;
1696 }
1697 }
1698
1699 return 0;
1700}
1701
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001702static noinline_for_stack bool
1703generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
Jens Axboe165125e2007-07-24 09:28:11 +02001705 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001706 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001707 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001708 char b[BDEVNAME_SIZE];
1709 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
Jens Axboec07e2b42007-07-18 13:27:58 +02001713 if (bio_check_eod(bio, nr_sectors))
1714 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001716 q = bdev_get_queue(bio->bi_bdev);
1717 if (unlikely(!q)) {
1718 printk(KERN_ERR
1719 "generic_make_request: Trying to access "
1720 "nonexistent block-device %s (%Lu)\n",
1721 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001722 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001723 goto end_io;
1724 }
1725
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001726 if (likely(bio_is_rw(bio) &&
1727 nr_sectors > queue_max_hw_sectors(q))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001728 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1729 bdevname(bio->bi_bdev, b),
1730 bio_sectors(bio),
1731 queue_max_hw_sectors(q));
1732 goto end_io;
1733 }
1734
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001735 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001736 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001737 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001738 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001739 goto end_io;
1740
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001742 * If this device has partitions, remap block n
1743 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001745 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001747 if (bio_check_eod(bio, nr_sectors))
1748 goto end_io;
1749
1750 /*
1751 * Filter flush bio's early so that make_request based
1752 * drivers without flush support don't have to worry
1753 * about them.
1754 */
1755 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1756 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1757 if (!nr_sectors) {
1758 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001759 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001763 if ((bio->bi_rw & REQ_DISCARD) &&
1764 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001765 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001766 err = -EOPNOTSUPP;
1767 goto end_io;
1768 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001770 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 err = -EOPNOTSUPP;
1772 goto end_io;
1773 }
1774
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001775 /*
1776 * Various block parts want %current->io_context and lazy ioc
1777 * allocation ends up trading a lot of pain for a small amount of
1778 * memory. Just allocate it upfront. This may fail and block
1779 * layer knows how to live with it.
1780 */
1781 create_io_context(GFP_ATOMIC, q->node);
1782
Tejun Heobc16a4f2011-10-19 14:33:01 +02001783 if (blk_throtl_bio(q, bio))
1784 return false; /* throttled, will be resubmitted later */
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001785
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001786 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001787 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001788
1789end_io:
1790 bio_endio(bio, err);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001791 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792}
1793
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001794/**
1795 * generic_make_request - hand a buffer to its device driver for I/O
1796 * @bio: The bio describing the location in memory and on the device.
1797 *
1798 * generic_make_request() is used to make I/O requests of block
1799 * devices. It is passed a &struct bio, which describes the I/O that needs
1800 * to be done.
1801 *
1802 * generic_make_request() does not return any status. The
1803 * success/failure status of the request, along with notification of
1804 * completion, is delivered asynchronously through the bio->bi_end_io
1805 * function described (one day) else where.
1806 *
1807 * The caller of generic_make_request must make sure that bi_io_vec
1808 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1809 * set to describe the device address, and the
1810 * bi_end_io and optionally bi_private are set to describe how
1811 * completion notification should be signaled.
1812 *
1813 * generic_make_request and the drivers it calls may use bi_next if this
1814 * bio happens to be merged with someone else, and may resubmit the bio to
1815 * a lower device by calling into generic_make_request recursively, which
1816 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001817 */
1818void generic_make_request(struct bio *bio)
1819{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001820 struct bio_list bio_list_on_stack;
1821
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001822 if (!generic_make_request_checks(bio))
1823 return;
1824
1825 /*
1826 * We only want one ->make_request_fn to be active at a time, else
1827 * stack usage with stacked devices could be a problem. So use
1828 * current->bio_list to keep a list of requests submited by a
1829 * make_request_fn function. current->bio_list is also used as a
1830 * flag to say if generic_make_request is currently active in this
1831 * task or not. If it is NULL, then no make_request is active. If
1832 * it is non-NULL, then a make_request is active, and new requests
1833 * should be added at the tail
1834 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001835 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001836 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001837 return;
1838 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001839
Neil Brownd89d8792007-05-01 09:53:42 +02001840 /* following loop may be a bit non-obvious, and so deserves some
1841 * explanation.
1842 * Before entering the loop, bio->bi_next is NULL (as all callers
1843 * ensure that) so we have a list with a single bio.
1844 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001845 * we assign bio_list to a pointer to the bio_list_on_stack,
1846 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001847 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001848 * through a recursive call to generic_make_request. If it
1849 * did, we find a non-NULL value in bio_list and re-enter the loop
1850 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001851 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001852 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001853 */
1854 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001855 bio_list_init(&bio_list_on_stack);
1856 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001857 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001858 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1859
1860 q->make_request_fn(q, bio);
1861
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001862 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001863 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001864 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001865}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866EXPORT_SYMBOL(generic_make_request);
1867
1868/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001869 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1871 * @bio: The &struct bio which describes the I/O
1872 *
1873 * submit_bio() is very similar in purpose to generic_make_request(), and
1874 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001875 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 *
1877 */
1878void submit_bio(int rw, struct bio *bio)
1879{
Jens Axboe22e2c502005-06-27 10:55:12 +02001880 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
Jens Axboebf2de6f2007-09-27 13:01:25 +02001882 /*
1883 * If it's a regular read/write or a barrier with data attached,
1884 * go through the normal accounting stuff before submission.
1885 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001886 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001887 unsigned int count;
1888
1889 if (unlikely(rw & REQ_WRITE_SAME))
1890 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1891 else
1892 count = bio_sectors(bio);
1893
Jens Axboebf2de6f2007-09-27 13:01:25 +02001894 if (rw & WRITE) {
1895 count_vm_events(PGPGOUT, count);
1896 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001897 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001898 count_vm_events(PGPGIN, count);
1899 }
1900
1901 if (unlikely(block_dump)) {
1902 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001903 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001904 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02001905 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001906 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02001907 bdevname(bio->bi_bdev, b),
1908 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 }
1911
1912 generic_make_request(bio);
1913}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914EXPORT_SYMBOL(submit_bio);
1915
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001916/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001917 * blk_rq_check_limits - Helper function to check a request for the queue limit
1918 * @q: the queue
1919 * @rq: the request being checked
1920 *
1921 * Description:
1922 * @rq may have been made based on weaker limitations of upper-level queues
1923 * in request stacking drivers, and it may violate the limitation of @q.
1924 * Since the block layer and the underlying device driver trust @rq
1925 * after it is inserted to @q, it should be checked against @q before
1926 * the insertion using this generic function.
1927 *
1928 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02001929 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001930 * Request stacking drivers like request-based dm may change the queue
1931 * limits while requests are in the queue (e.g. dm's table swapping).
1932 * Such request stacking drivers should check those requests agaist
1933 * the new queue limits again when they dispatch those requests,
1934 * although such checkings are also done against the old queue limits
1935 * when submitting requests.
1936 */
1937int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1938{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001939 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04001940 return 0;
1941
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04001942 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001943 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1944 return -EIO;
1945 }
1946
1947 /*
1948 * queue's settings related to segment counting like q->bounce_pfn
1949 * may differ from that of other stacking queues.
1950 * Recalculate it to check the request correctly on this queue's
1951 * limitation.
1952 */
1953 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001954 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001955 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1956 return -EIO;
1957 }
1958
1959 return 0;
1960}
1961EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1962
1963/**
1964 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1965 * @q: the queue to submit the request
1966 * @rq: the request being queued
1967 */
1968int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1969{
1970 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02001971 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001972
1973 if (blk_rq_check_limits(q, rq))
1974 return -EIO;
1975
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001976 if (rq->rq_disk &&
1977 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001978 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001979
1980 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001981 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01001982 spin_unlock_irqrestore(q->queue_lock, flags);
1983 return -ENODEV;
1984 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001985
1986 /*
1987 * Submitting request must be dequeued before calling this function
1988 * because it will be linked to another request_queue
1989 */
1990 BUG_ON(blk_queued_rq(rq));
1991
Jeff Moyer4853aba2011-08-15 21:37:25 +02001992 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1993 where = ELEVATOR_INSERT_FLUSH;
1994
1995 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02001996 if (where == ELEVATOR_INSERT_FLUSH)
1997 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001998 spin_unlock_irqrestore(q->queue_lock, flags);
1999
2000 return 0;
2001}
2002EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2003
Tejun Heo80a761f2009-07-03 17:48:17 +09002004/**
2005 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2006 * @rq: request to examine
2007 *
2008 * Description:
2009 * A request could be merge of IOs which require different failure
2010 * handling. This function determines the number of bytes which
2011 * can be failed from the beginning of the request without
2012 * crossing into area which need to be retried further.
2013 *
2014 * Return:
2015 * The number of bytes to fail.
2016 *
2017 * Context:
2018 * queue_lock must be held.
2019 */
2020unsigned int blk_rq_err_bytes(const struct request *rq)
2021{
2022 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2023 unsigned int bytes = 0;
2024 struct bio *bio;
2025
2026 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2027 return blk_rq_bytes(rq);
2028
2029 /*
2030 * Currently the only 'mixing' which can happen is between
2031 * different fastfail types. We can safely fail portions
2032 * which have all the failfast bits that the first one has -
2033 * the ones which are at least as eager to fail as the first
2034 * one.
2035 */
2036 for (bio = rq->bio; bio; bio = bio->bi_next) {
2037 if ((bio->bi_rw & ff) != ff)
2038 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002039 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002040 }
2041
2042 /* this could lead to infinite loop */
2043 BUG_ON(blk_rq_bytes(rq) && !bytes);
2044 return bytes;
2045}
2046EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2047
Jens Axboe320ae512013-10-24 09:20:05 +01002048void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002049{
Jens Axboec2553b52009-04-24 08:10:11 +02002050 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002051 const int rw = rq_data_dir(req);
2052 struct hd_struct *part;
2053 int cpu;
2054
2055 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002056 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002057 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2058 part_stat_unlock();
2059 }
2060}
2061
Jens Axboe320ae512013-10-24 09:20:05 +01002062void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002063{
Jens Axboebc58ba92009-01-23 10:54:44 +01002064 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002065 * Account IO completion. flush_rq isn't accounted as a
2066 * normal IO on queueing nor completion. Accounting the
2067 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002068 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002069 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002070 unsigned long duration = jiffies - req->start_time;
2071 const int rw = rq_data_dir(req);
2072 struct hd_struct *part;
2073 int cpu;
2074
2075 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002076 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002077
2078 part_stat_inc(cpu, part, ios[rw]);
2079 part_stat_add(cpu, part, ticks[rw], duration);
2080 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002081 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002082
Jens Axboe6c23a962011-01-07 08:43:37 +01002083 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002084 part_stat_unlock();
2085 }
2086}
2087
Lin Mingc8158812013-03-23 11:42:27 +08002088#ifdef CONFIG_PM_RUNTIME
2089/*
2090 * Don't process normal requests when queue is suspended
2091 * or in the process of suspending/resuming
2092 */
2093static struct request *blk_pm_peek_request(struct request_queue *q,
2094 struct request *rq)
2095{
2096 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2097 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2098 return NULL;
2099 else
2100 return rq;
2101}
2102#else
2103static inline struct request *blk_pm_peek_request(struct request_queue *q,
2104 struct request *rq)
2105{
2106 return rq;
2107}
2108#endif
2109
Jens Axboe320ae512013-10-24 09:20:05 +01002110void blk_account_io_start(struct request *rq, bool new_io)
2111{
2112 struct hd_struct *part;
2113 int rw = rq_data_dir(rq);
2114 int cpu;
2115
2116 if (!blk_do_io_stat(rq))
2117 return;
2118
2119 cpu = part_stat_lock();
2120
2121 if (!new_io) {
2122 part = rq->part;
2123 part_stat_inc(cpu, part, merges[rw]);
2124 } else {
2125 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2126 if (!hd_struct_try_get(part)) {
2127 /*
2128 * The partition is already being removed,
2129 * the request will be accounted on the disk only
2130 *
2131 * We take a reference on disk->part0 although that
2132 * partition will never be deleted, so we can treat
2133 * it as any other partition.
2134 */
2135 part = &rq->rq_disk->part0;
2136 hd_struct_get(part);
2137 }
2138 part_round_stats(cpu, part);
2139 part_inc_in_flight(part, rw);
2140 rq->part = part;
2141 }
2142
2143 part_stat_unlock();
2144}
2145
Tejun Heo53a08802008-12-03 12:41:26 +01002146/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002147 * blk_peek_request - peek at the top of a request queue
2148 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002149 *
2150 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002151 * Return the request at the top of @q. The returned request
2152 * should be started using blk_start_request() before LLD starts
2153 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002154 *
2155 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002156 * Pointer to the request at the top of @q if available. Null
2157 * otherwise.
2158 *
2159 * Context:
2160 * queue_lock must be held.
2161 */
2162struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002163{
2164 struct request *rq;
2165 int ret;
2166
2167 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002168
2169 rq = blk_pm_peek_request(q, rq);
2170 if (!rq)
2171 break;
2172
Tejun Heo158dbda2009-04-23 11:05:18 +09002173 if (!(rq->cmd_flags & REQ_STARTED)) {
2174 /*
2175 * This is the first time the device driver
2176 * sees this request (possibly after
2177 * requeueing). Notify IO scheduler.
2178 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002179 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002180 elv_activate_rq(q, rq);
2181
2182 /*
2183 * just mark as started even if we don't start
2184 * it, a request that has been delayed should
2185 * not be passed by new incoming requests
2186 */
2187 rq->cmd_flags |= REQ_STARTED;
2188 trace_block_rq_issue(q, rq);
2189 }
2190
2191 if (!q->boundary_rq || q->boundary_rq == rq) {
2192 q->end_sector = rq_end_sector(rq);
2193 q->boundary_rq = NULL;
2194 }
2195
2196 if (rq->cmd_flags & REQ_DONTPREP)
2197 break;
2198
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002199 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002200 /*
2201 * make sure space for the drain appears we
2202 * know we can do this because max_hw_segments
2203 * has been adjusted to be one fewer than the
2204 * device can handle
2205 */
2206 rq->nr_phys_segments++;
2207 }
2208
2209 if (!q->prep_rq_fn)
2210 break;
2211
2212 ret = q->prep_rq_fn(q, rq);
2213 if (ret == BLKPREP_OK) {
2214 break;
2215 } else if (ret == BLKPREP_DEFER) {
2216 /*
2217 * the request may have been (partially) prepped.
2218 * we need to keep this request in the front to
2219 * avoid resource deadlock. REQ_STARTED will
2220 * prevent other fs requests from passing this one.
2221 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002222 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002223 !(rq->cmd_flags & REQ_DONTPREP)) {
2224 /*
2225 * remove the space for the drain we added
2226 * so that we don't add it again
2227 */
2228 --rq->nr_phys_segments;
2229 }
2230
2231 rq = NULL;
2232 break;
2233 } else if (ret == BLKPREP_KILL) {
2234 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002235 /*
2236 * Mark this request as started so we don't trigger
2237 * any debug logic in the end I/O path.
2238 */
2239 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002240 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002241 } else {
2242 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2243 break;
2244 }
2245 }
2246
2247 return rq;
2248}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002249EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002250
Tejun Heo9934c8c2009-05-08 11:54:16 +09002251void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002252{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002253 struct request_queue *q = rq->q;
2254
Tejun Heo158dbda2009-04-23 11:05:18 +09002255 BUG_ON(list_empty(&rq->queuelist));
2256 BUG_ON(ELV_ON_HASH(rq));
2257
2258 list_del_init(&rq->queuelist);
2259
2260 /*
2261 * the time frame between a request being removed from the lists
2262 * and to it is freed is accounted as io that is in progress at
2263 * the driver side.
2264 */
Divyesh Shah91952912010-04-01 15:01:41 -07002265 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002266 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002267 set_io_start_time_ns(rq);
2268 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002269}
2270
Tejun Heo5efccd12009-04-23 11:05:18 +09002271/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002272 * blk_start_request - start request processing on the driver
2273 * @req: request to dequeue
2274 *
2275 * Description:
2276 * Dequeue @req and start timeout timer on it. This hands off the
2277 * request to the driver.
2278 *
2279 * Block internal functions which don't want to start timer should
2280 * call blk_dequeue_request().
2281 *
2282 * Context:
2283 * queue_lock must be held.
2284 */
2285void blk_start_request(struct request *req)
2286{
2287 blk_dequeue_request(req);
2288
2289 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002290 * We are now handing the request to the hardware, initialize
2291 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002292 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002293 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002294 if (unlikely(blk_bidi_rq(req)))
2295 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2296
Jeff Moyer4912aa62013-10-08 14:36:41 -04002297 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002298 blk_add_timer(req);
2299}
2300EXPORT_SYMBOL(blk_start_request);
2301
2302/**
2303 * blk_fetch_request - fetch a request from a request queue
2304 * @q: request queue to fetch a request from
2305 *
2306 * Description:
2307 * Return the request at the top of @q. The request is started on
2308 * return and LLD can start processing it immediately.
2309 *
2310 * Return:
2311 * Pointer to the request at the top of @q if available. Null
2312 * otherwise.
2313 *
2314 * Context:
2315 * queue_lock must be held.
2316 */
2317struct request *blk_fetch_request(struct request_queue *q)
2318{
2319 struct request *rq;
2320
2321 rq = blk_peek_request(q);
2322 if (rq)
2323 blk_start_request(rq);
2324 return rq;
2325}
2326EXPORT_SYMBOL(blk_fetch_request);
2327
2328/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002329 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002330 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002331 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002332 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002333 *
2334 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002335 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2336 * the request structure even if @req doesn't have leftover.
2337 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002338 *
2339 * This special helper function is only for request stacking drivers
2340 * (e.g. request-based dm) so that they can handle partial completion.
2341 * Actual device drivers should use blk_end_request instead.
2342 *
2343 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2344 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002345 *
2346 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002347 * %false - this request doesn't have any more data
2348 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002349 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002350bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002352 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Tejun Heo2e60e022009-04-23 11:05:18 +09002354 if (!req->bio)
2355 return false;
2356
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01002357 trace_block_rq_complete(req->q, req);
Jens Axboe2056a782006-03-23 20:00:26 +01002358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002360 * For fs requests, rq is just carrier of independent bio's
2361 * and each partial completion should be handled separately.
2362 * Reset per-request error on each partial completion.
2363 *
2364 * TODO: tj: This is too subtle. It would be better to let
2365 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002367 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 req->errors = 0;
2369
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002370 if (error && req->cmd_type == REQ_TYPE_FS &&
2371 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002372 char *error_type;
2373
2374 switch (error) {
2375 case -ENOLINK:
2376 error_type = "recoverable transport";
2377 break;
2378 case -EREMOTEIO:
2379 error_type = "critical target";
2380 break;
2381 case -EBADE:
2382 error_type = "critical nexus";
2383 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002384 case -ETIMEDOUT:
2385 error_type = "timeout";
2386 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002387 case -ENOSPC:
2388 error_type = "critical space allocation";
2389 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002390 case -ENODATA:
2391 error_type = "critical medium";
2392 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002393 case -EIO:
2394 default:
2395 error_type = "I/O";
2396 break;
2397 }
Yi Zou37d7b342012-08-30 16:26:25 -07002398 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2399 error_type, req->rq_disk ?
2400 req->rq_disk->disk_name : "?",
2401 (unsigned long long)blk_rq_pos(req));
2402
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 }
2404
Jens Axboebc58ba92009-01-23 10:54:44 +01002405 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002406
Kent Overstreetf79ea412012-09-20 16:38:30 -07002407 total_bytes = 0;
2408 while (req->bio) {
2409 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002410 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Kent Overstreet4f024f32013-10-11 15:44:27 -07002412 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
Kent Overstreetf79ea412012-09-20 16:38:30 -07002415 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Kent Overstreetf79ea412012-09-20 16:38:30 -07002417 total_bytes += bio_bytes;
2418 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Kent Overstreetf79ea412012-09-20 16:38:30 -07002420 if (!nr_bytes)
2421 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 }
2423
2424 /*
2425 * completely done
2426 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002427 if (!req->bio) {
2428 /*
2429 * Reset counters so that the request stacking driver
2430 * can find how many bytes remain in the request
2431 * later.
2432 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002433 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002434 return false;
2435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002437 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002438 req->buffer = bio_data(req->bio);
2439
2440 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002441 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002442 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002443
Tejun Heo80a761f2009-07-03 17:48:17 +09002444 /* mixed attributes always follow the first bio */
2445 if (req->cmd_flags & REQ_MIXED_MERGE) {
2446 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2447 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2448 }
2449
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002450 /*
2451 * If total number of sectors is less than the first segment
2452 * size, something has gone terribly wrong.
2453 */
2454 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002455 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002456 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002457 }
2458
2459 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002461
Tejun Heo2e60e022009-04-23 11:05:18 +09002462 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463}
Tejun Heo2e60e022009-04-23 11:05:18 +09002464EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Tejun Heo2e60e022009-04-23 11:05:18 +09002466static bool blk_update_bidi_request(struct request *rq, int error,
2467 unsigned int nr_bytes,
2468 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002469{
Tejun Heo2e60e022009-04-23 11:05:18 +09002470 if (blk_update_request(rq, error, nr_bytes))
2471 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002472
Tejun Heo2e60e022009-04-23 11:05:18 +09002473 /* Bidi request must be completed as a whole */
2474 if (unlikely(blk_bidi_rq(rq)) &&
2475 blk_update_request(rq->next_rq, error, bidi_bytes))
2476 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002477
Jens Axboee2e1a142010-06-09 10:42:09 +02002478 if (blk_queue_add_random(rq->q))
2479 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002480
2481 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482}
2483
James Bottomley28018c22010-07-01 19:49:17 +09002484/**
2485 * blk_unprep_request - unprepare a request
2486 * @req: the request
2487 *
2488 * This function makes a request ready for complete resubmission (or
2489 * completion). It happens only after all error handling is complete,
2490 * so represents the appropriate moment to deallocate any resources
2491 * that were allocated to the request in the prep_rq_fn. The queue
2492 * lock is held when calling this.
2493 */
2494void blk_unprep_request(struct request *req)
2495{
2496 struct request_queue *q = req->q;
2497
2498 req->cmd_flags &= ~REQ_DONTPREP;
2499 if (q->unprep_rq_fn)
2500 q->unprep_rq_fn(q, req);
2501}
2502EXPORT_SYMBOL_GPL(blk_unprep_request);
2503
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504/*
2505 * queue lock must be held
2506 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002507static void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508{
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002509 if (blk_rq_tagged(req))
2510 blk_queue_end_tag(req->q, req);
2511
James Bottomleyba396a62009-05-27 14:17:08 +02002512 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002514 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002515 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
Mike Andersone78042e2008-10-30 02:16:20 -07002517 blk_delete_timer(req);
2518
James Bottomley28018c22010-07-01 19:49:17 +09002519 if (req->cmd_flags & REQ_DONTPREP)
2520 blk_unprep_request(req);
2521
Jens Axboebc58ba92009-01-23 10:54:44 +01002522 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002525 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002526 else {
2527 if (blk_bidi_rq(req))
2528 __blk_put_request(req->next_rq->q, req->next_rq);
2529
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 }
2532}
2533
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002534/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002535 * blk_end_bidi_request - Complete a bidi request
2536 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002537 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002538 * @nr_bytes: number of bytes to complete @rq
2539 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002540 *
2541 * Description:
2542 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002543 * Drivers that supports bidi can safely call this member for any
2544 * type of request, bidi or uni. In the later case @bidi_bytes is
2545 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002546 *
2547 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002548 * %false - we are done with this request
2549 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002550 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002551static bool blk_end_bidi_request(struct request *rq, int error,
2552 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002553{
2554 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002555 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002556
Tejun Heo2e60e022009-04-23 11:05:18 +09002557 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2558 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002559
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002560 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002561 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002562 spin_unlock_irqrestore(q->queue_lock, flags);
2563
Tejun Heo2e60e022009-04-23 11:05:18 +09002564 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002565}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002566
2567/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002568 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2569 * @rq: the request to complete
2570 * @error: %0 for success, < %0 for error
2571 * @nr_bytes: number of bytes to complete @rq
2572 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002573 *
2574 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002575 * Identical to blk_end_bidi_request() except that queue lock is
2576 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002577 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002578 * Return:
2579 * %false - we are done with this request
2580 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002581 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002582bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002583 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002584{
Tejun Heo2e60e022009-04-23 11:05:18 +09002585 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2586 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002587
Tejun Heo2e60e022009-04-23 11:05:18 +09002588 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002589
Tejun Heo2e60e022009-04-23 11:05:18 +09002590 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002591}
2592
2593/**
2594 * blk_end_request - Helper function for drivers to complete the request.
2595 * @rq: the request being processed
2596 * @error: %0 for success, < %0 for error
2597 * @nr_bytes: number of bytes to complete
2598 *
2599 * Description:
2600 * Ends I/O on a number of bytes attached to @rq.
2601 * If @rq has leftover, sets it up for the next range of segments.
2602 *
2603 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002604 * %false - we are done with this request
2605 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002606 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002607bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002608{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002609 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002610}
Jens Axboe56ad1742009-07-28 22:11:24 +02002611EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002612
2613/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002614 * blk_end_request_all - Helper function for drives to finish the request.
2615 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002616 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002617 *
2618 * Description:
2619 * Completely finish @rq.
2620 */
2621void blk_end_request_all(struct request *rq, int error)
2622{
2623 bool pending;
2624 unsigned int bidi_bytes = 0;
2625
2626 if (unlikely(blk_bidi_rq(rq)))
2627 bidi_bytes = blk_rq_bytes(rq->next_rq);
2628
2629 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2630 BUG_ON(pending);
2631}
Jens Axboe56ad1742009-07-28 22:11:24 +02002632EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002633
2634/**
2635 * blk_end_request_cur - Helper function to finish the current request chunk.
2636 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002637 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002638 *
2639 * Description:
2640 * Complete the current consecutively mapped chunk from @rq.
2641 *
2642 * Return:
2643 * %false - we are done with this request
2644 * %true - still buffers pending for this request
2645 */
2646bool blk_end_request_cur(struct request *rq, int error)
2647{
2648 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2649}
Jens Axboe56ad1742009-07-28 22:11:24 +02002650EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002651
2652/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002653 * blk_end_request_err - Finish a request till the next failure boundary.
2654 * @rq: the request to finish till the next failure boundary for
2655 * @error: must be negative errno
2656 *
2657 * Description:
2658 * Complete @rq till the next failure boundary.
2659 *
2660 * Return:
2661 * %false - we are done with this request
2662 * %true - still buffers pending for this request
2663 */
2664bool blk_end_request_err(struct request *rq, int error)
2665{
2666 WARN_ON(error >= 0);
2667 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2668}
2669EXPORT_SYMBOL_GPL(blk_end_request_err);
2670
2671/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002672 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002673 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002674 * @error: %0 for success, < %0 for error
2675 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002676 *
2677 * Description:
2678 * Must be called with queue lock held unlike blk_end_request().
2679 *
2680 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002681 * %false - we are done with this request
2682 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002683 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002684bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002685{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002686 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002687}
Jens Axboe56ad1742009-07-28 22:11:24 +02002688EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002689
2690/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002691 * __blk_end_request_all - Helper function for drives to finish the request.
2692 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002693 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002694 *
2695 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002696 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002697 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002698void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002699{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002700 bool pending;
2701 unsigned int bidi_bytes = 0;
2702
2703 if (unlikely(blk_bidi_rq(rq)))
2704 bidi_bytes = blk_rq_bytes(rq->next_rq);
2705
2706 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2707 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002708}
Jens Axboe56ad1742009-07-28 22:11:24 +02002709EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002710
2711/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002712 * __blk_end_request_cur - Helper function to finish the current request chunk.
2713 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002714 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002715 *
2716 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002717 * Complete the current consecutively mapped chunk from @rq. Must
2718 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002719 *
2720 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002721 * %false - we are done with this request
2722 * %true - still buffers pending for this request
2723 */
2724bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002725{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002726 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002727}
Jens Axboe56ad1742009-07-28 22:11:24 +02002728EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002729
Tejun Heo80a761f2009-07-03 17:48:17 +09002730/**
2731 * __blk_end_request_err - Finish a request till the next failure boundary.
2732 * @rq: the request to finish till the next failure boundary for
2733 * @error: must be negative errno
2734 *
2735 * Description:
2736 * Complete @rq till the next failure boundary. Must be called
2737 * with queue lock held.
2738 *
2739 * Return:
2740 * %false - we are done with this request
2741 * %true - still buffers pending for this request
2742 */
2743bool __blk_end_request_err(struct request *rq, int error)
2744{
2745 WARN_ON(error >= 0);
2746 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2747}
2748EXPORT_SYMBOL_GPL(__blk_end_request_err);
2749
Jens Axboe86db1e22008-01-29 14:53:40 +01002750void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2751 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002753 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002754 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
David Woodhousefb2dce82008-08-05 18:01:53 +01002756 if (bio_has_data(bio)) {
2757 rq->nr_phys_segments = bio_phys_segments(q, bio);
David Woodhousefb2dce82008-08-05 18:01:53 +01002758 rq->buffer = bio_data(bio);
2759 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07002760 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
NeilBrown66846572007-08-16 13:31:28 +02002763 if (bio->bi_bdev)
2764 rq->rq_disk = bio->bi_bdev->bd_disk;
2765}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002767#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2768/**
2769 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2770 * @rq: the request to be flushed
2771 *
2772 * Description:
2773 * Flush all pages in @rq.
2774 */
2775void rq_flush_dcache_pages(struct request *rq)
2776{
2777 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002778 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002779
2780 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002781 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002782}
2783EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2784#endif
2785
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002786/**
2787 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2788 * @q : the queue of the device being checked
2789 *
2790 * Description:
2791 * Check if underlying low-level drivers of a device are busy.
2792 * If the drivers want to export their busy state, they must set own
2793 * exporting function using blk_queue_lld_busy() first.
2794 *
2795 * Basically, this function is used only by request stacking drivers
2796 * to stop dispatching requests to underlying devices when underlying
2797 * devices are busy. This behavior helps more I/O merging on the queue
2798 * of the request stacking driver and prevents I/O throughput regression
2799 * on burst I/O load.
2800 *
2801 * Return:
2802 * 0 - Not busy (The request stacking driver should dispatch request)
2803 * 1 - Busy (The request stacking driver should stop dispatching request)
2804 */
2805int blk_lld_busy(struct request_queue *q)
2806{
2807 if (q->lld_busy_fn)
2808 return q->lld_busy_fn(q);
2809
2810 return 0;
2811}
2812EXPORT_SYMBOL_GPL(blk_lld_busy);
2813
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002814/**
2815 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2816 * @rq: the clone request to be cleaned up
2817 *
2818 * Description:
2819 * Free all bios in @rq for a cloned request.
2820 */
2821void blk_rq_unprep_clone(struct request *rq)
2822{
2823 struct bio *bio;
2824
2825 while ((bio = rq->bio) != NULL) {
2826 rq->bio = bio->bi_next;
2827
2828 bio_put(bio);
2829 }
2830}
2831EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2832
2833/*
2834 * Copy attributes of the original request to the clone request.
2835 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2836 */
2837static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2838{
2839 dst->cpu = src->cpu;
Tejun Heo3a2edd02010-09-03 11:56:18 +02002840 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002841 dst->cmd_type = src->cmd_type;
2842 dst->__sector = blk_rq_pos(src);
2843 dst->__data_len = blk_rq_bytes(src);
2844 dst->nr_phys_segments = src->nr_phys_segments;
2845 dst->ioprio = src->ioprio;
2846 dst->extra_len = src->extra_len;
2847}
2848
2849/**
2850 * blk_rq_prep_clone - Helper function to setup clone request
2851 * @rq: the request to be setup
2852 * @rq_src: original request to be cloned
2853 * @bs: bio_set that bios for clone are allocated from
2854 * @gfp_mask: memory allocation mask for bio
2855 * @bio_ctr: setup function to be called for each clone bio.
2856 * Returns %0 for success, non %0 for failure.
2857 * @data: private data to be passed to @bio_ctr
2858 *
2859 * Description:
2860 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2861 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2862 * are not copied, and copying such parts is the caller's responsibility.
2863 * Also, pages which the original bios are pointing to are not copied
2864 * and the cloned bios just point same pages.
2865 * So cloned bios must be completed before original bios, which means
2866 * the caller must complete @rq before @rq_src.
2867 */
2868int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2869 struct bio_set *bs, gfp_t gfp_mask,
2870 int (*bio_ctr)(struct bio *, struct bio *, void *),
2871 void *data)
2872{
2873 struct bio *bio, *bio_src;
2874
2875 if (!bs)
2876 bs = fs_bio_set;
2877
2878 blk_rq_init(NULL, rq);
2879
2880 __rq_for_each_bio(bio_src, rq_src) {
Kent Overstreetbf800ef2012-09-06 15:35:02 -07002881 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002882 if (!bio)
2883 goto free_and_out;
2884
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002885 if (bio_ctr && bio_ctr(bio, bio_src, data))
2886 goto free_and_out;
2887
2888 if (rq->bio) {
2889 rq->biotail->bi_next = bio;
2890 rq->biotail = bio;
2891 } else
2892 rq->bio = rq->biotail = bio;
2893 }
2894
2895 __blk_rq_prep_clone(rq, rq_src);
2896
2897 return 0;
2898
2899free_and_out:
2900 if (bio)
Kent Overstreet4254bba2012-09-06 15:35:00 -07002901 bio_put(bio);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002902 blk_rq_unprep_clone(rq);
2903
2904 return -ENOMEM;
2905}
2906EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2907
Jens Axboe18887ad2008-07-28 13:08:45 +02002908int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909{
2910 return queue_work(kblockd_workqueue, work);
2911}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912EXPORT_SYMBOL(kblockd_schedule_work);
2913
Vivek Goyale43473b2010-09-15 17:06:35 -04002914int kblockd_schedule_delayed_work(struct request_queue *q,
2915 struct delayed_work *dwork, unsigned long delay)
2916{
2917 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2918}
2919EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2920
Jens Axboe73c10102011-03-08 13:19:51 +01002921#define PLUG_MAGIC 0x91827364
2922
Suresh Jayaraman75df7132011-09-21 10:00:16 +02002923/**
2924 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2925 * @plug: The &struct blk_plug that needs to be initialized
2926 *
2927 * Description:
2928 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2929 * pending I/O should the task end up blocking between blk_start_plug() and
2930 * blk_finish_plug(). This is important from a performance perspective, but
2931 * also ensures that we don't deadlock. For instance, if the task is blocking
2932 * for a memory allocation, memory reclaim could end up wanting to free a
2933 * page belonging to that request that is currently residing in our private
2934 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2935 * this kind of deadlock.
2936 */
Jens Axboe73c10102011-03-08 13:19:51 +01002937void blk_start_plug(struct blk_plug *plug)
2938{
2939 struct task_struct *tsk = current;
2940
2941 plug->magic = PLUG_MAGIC;
2942 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01002943 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02002944 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01002945
2946 /*
2947 * If this is a nested plug, don't actually assign it. It will be
2948 * flushed on its own.
2949 */
2950 if (!tsk->plug) {
2951 /*
2952 * Store ordering should not be needed here, since a potential
2953 * preempt will imply a full memory barrier
2954 */
2955 tsk->plug = plug;
2956 }
2957}
2958EXPORT_SYMBOL(blk_start_plug);
2959
2960static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2961{
2962 struct request *rqa = container_of(a, struct request, queuelist);
2963 struct request *rqb = container_of(b, struct request, queuelist);
2964
Jianpeng Ma975927b2012-10-25 21:58:17 +02002965 return !(rqa->q < rqb->q ||
2966 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01002967}
2968
Jens Axboe49cac012011-04-16 13:51:05 +02002969/*
2970 * If 'from_schedule' is true, then postpone the dispatch of requests
2971 * until a safe kblockd context. We due this to avoid accidental big
2972 * additional stack usage in driver dispatch, in places where the originally
2973 * plugger did not intend it.
2974 */
Jens Axboef6603782011-04-15 15:49:07 +02002975static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02002976 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02002977 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02002978{
Jens Axboe49cac012011-04-16 13:51:05 +02002979 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02002980
Bart Van Assche70460572012-11-28 13:45:56 +01002981 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002982 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002983 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02002984 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01002985 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02002986}
2987
NeilBrown74018dc2012-07-31 09:08:15 +02002988static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02002989{
2990 LIST_HEAD(callbacks);
2991
Shaohua Li2a7d5552012-07-31 09:08:15 +02002992 while (!list_empty(&plug->cb_list)) {
2993 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02002994
Shaohua Li2a7d5552012-07-31 09:08:15 +02002995 while (!list_empty(&callbacks)) {
2996 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02002997 struct blk_plug_cb,
2998 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02002999 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003000 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003001 }
NeilBrown048c9372011-04-18 09:52:22 +02003002 }
3003}
3004
NeilBrown9cbb1752012-07-31 09:08:14 +02003005struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3006 int size)
3007{
3008 struct blk_plug *plug = current->plug;
3009 struct blk_plug_cb *cb;
3010
3011 if (!plug)
3012 return NULL;
3013
3014 list_for_each_entry(cb, &plug->cb_list, list)
3015 if (cb->callback == unplug && cb->data == data)
3016 return cb;
3017
3018 /* Not currently on the callback list */
3019 BUG_ON(size < sizeof(*cb));
3020 cb = kzalloc(size, GFP_ATOMIC);
3021 if (cb) {
3022 cb->data = data;
3023 cb->callback = unplug;
3024 list_add(&cb->list, &plug->cb_list);
3025 }
3026 return cb;
3027}
3028EXPORT_SYMBOL(blk_check_plugged);
3029
Jens Axboe49cac012011-04-16 13:51:05 +02003030void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003031{
3032 struct request_queue *q;
3033 unsigned long flags;
3034 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003035 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003036 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003037
3038 BUG_ON(plug->magic != PLUG_MAGIC);
3039
NeilBrown74018dc2012-07-31 09:08:15 +02003040 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003041
3042 if (!list_empty(&plug->mq_list))
3043 blk_mq_flush_plug_list(plug, from_schedule);
3044
Jens Axboe73c10102011-03-08 13:19:51 +01003045 if (list_empty(&plug->list))
3046 return;
3047
NeilBrown109b8122011-04-11 14:13:10 +02003048 list_splice_init(&plug->list, &list);
3049
Jianpeng Ma422765c2013-01-11 14:46:09 +01003050 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003051
3052 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003053 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003054
3055 /*
3056 * Save and disable interrupts here, to avoid doing it for every
3057 * queue lock we have to take.
3058 */
Jens Axboe73c10102011-03-08 13:19:51 +01003059 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003060 while (!list_empty(&list)) {
3061 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003062 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003063 BUG_ON(!rq->q);
3064 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003065 /*
3066 * This drops the queue lock
3067 */
3068 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003069 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003070 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003071 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003072 spin_lock(q->queue_lock);
3073 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003074
3075 /*
3076 * Short-circuit if @q is dead
3077 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003078 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003079 __blk_end_request_all(rq, -ENODEV);
3080 continue;
3081 }
3082
Jens Axboe73c10102011-03-08 13:19:51 +01003083 /*
3084 * rq is already accounted, so use raw insert
3085 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003086 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3087 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3088 else
3089 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003090
3091 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003092 }
3093
Jens Axboe99e22592011-04-18 09:59:55 +02003094 /*
3095 * This drops the queue lock
3096 */
3097 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003098 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003099
Jens Axboe73c10102011-03-08 13:19:51 +01003100 local_irq_restore(flags);
3101}
Jens Axboe73c10102011-03-08 13:19:51 +01003102
3103void blk_finish_plug(struct blk_plug *plug)
3104{
Jens Axboef6603782011-04-15 15:49:07 +02003105 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003106
3107 if (plug == current->plug)
3108 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003109}
3110EXPORT_SYMBOL(blk_finish_plug);
3111
Lin Ming6c954662013-03-23 11:42:26 +08003112#ifdef CONFIG_PM_RUNTIME
3113/**
3114 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3115 * @q: the queue of the device
3116 * @dev: the device the queue belongs to
3117 *
3118 * Description:
3119 * Initialize runtime-PM-related fields for @q and start auto suspend for
3120 * @dev. Drivers that want to take advantage of request-based runtime PM
3121 * should call this function after @dev has been initialized, and its
3122 * request queue @q has been allocated, and runtime PM for it can not happen
3123 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3124 * cases, driver should call this function before any I/O has taken place.
3125 *
3126 * This function takes care of setting up using auto suspend for the device,
3127 * the autosuspend delay is set to -1 to make runtime suspend impossible
3128 * until an updated value is either set by user or by driver. Drivers do
3129 * not need to touch other autosuspend settings.
3130 *
3131 * The block layer runtime PM is request based, so only works for drivers
3132 * that use request as their IO unit instead of those directly use bio's.
3133 */
3134void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3135{
3136 q->dev = dev;
3137 q->rpm_status = RPM_ACTIVE;
3138 pm_runtime_set_autosuspend_delay(q->dev, -1);
3139 pm_runtime_use_autosuspend(q->dev);
3140}
3141EXPORT_SYMBOL(blk_pm_runtime_init);
3142
3143/**
3144 * blk_pre_runtime_suspend - Pre runtime suspend check
3145 * @q: the queue of the device
3146 *
3147 * Description:
3148 * This function will check if runtime suspend is allowed for the device
3149 * by examining if there are any requests pending in the queue. If there
3150 * are requests pending, the device can not be runtime suspended; otherwise,
3151 * the queue's status will be updated to SUSPENDING and the driver can
3152 * proceed to suspend the device.
3153 *
3154 * For the not allowed case, we mark last busy for the device so that
3155 * runtime PM core will try to autosuspend it some time later.
3156 *
3157 * This function should be called near the start of the device's
3158 * runtime_suspend callback.
3159 *
3160 * Return:
3161 * 0 - OK to runtime suspend the device
3162 * -EBUSY - Device should not be runtime suspended
3163 */
3164int blk_pre_runtime_suspend(struct request_queue *q)
3165{
3166 int ret = 0;
3167
3168 spin_lock_irq(q->queue_lock);
3169 if (q->nr_pending) {
3170 ret = -EBUSY;
3171 pm_runtime_mark_last_busy(q->dev);
3172 } else {
3173 q->rpm_status = RPM_SUSPENDING;
3174 }
3175 spin_unlock_irq(q->queue_lock);
3176 return ret;
3177}
3178EXPORT_SYMBOL(blk_pre_runtime_suspend);
3179
3180/**
3181 * blk_post_runtime_suspend - Post runtime suspend processing
3182 * @q: the queue of the device
3183 * @err: return value of the device's runtime_suspend function
3184 *
3185 * Description:
3186 * Update the queue's runtime status according to the return value of the
3187 * device's runtime suspend function and mark last busy for the device so
3188 * that PM core will try to auto suspend the device at a later time.
3189 *
3190 * This function should be called near the end of the device's
3191 * runtime_suspend callback.
3192 */
3193void blk_post_runtime_suspend(struct request_queue *q, int err)
3194{
3195 spin_lock_irq(q->queue_lock);
3196 if (!err) {
3197 q->rpm_status = RPM_SUSPENDED;
3198 } else {
3199 q->rpm_status = RPM_ACTIVE;
3200 pm_runtime_mark_last_busy(q->dev);
3201 }
3202 spin_unlock_irq(q->queue_lock);
3203}
3204EXPORT_SYMBOL(blk_post_runtime_suspend);
3205
3206/**
3207 * blk_pre_runtime_resume - Pre runtime resume processing
3208 * @q: the queue of the device
3209 *
3210 * Description:
3211 * Update the queue's runtime status to RESUMING in preparation for the
3212 * runtime resume of the device.
3213 *
3214 * This function should be called near the start of the device's
3215 * runtime_resume callback.
3216 */
3217void blk_pre_runtime_resume(struct request_queue *q)
3218{
3219 spin_lock_irq(q->queue_lock);
3220 q->rpm_status = RPM_RESUMING;
3221 spin_unlock_irq(q->queue_lock);
3222}
3223EXPORT_SYMBOL(blk_pre_runtime_resume);
3224
3225/**
3226 * blk_post_runtime_resume - Post runtime resume processing
3227 * @q: the queue of the device
3228 * @err: return value of the device's runtime_resume function
3229 *
3230 * Description:
3231 * Update the queue's runtime status according to the return value of the
3232 * device's runtime_resume function. If it is successfully resumed, process
3233 * the requests that are queued into the device's queue when it is resuming
3234 * and then mark last busy and initiate autosuspend for it.
3235 *
3236 * This function should be called near the end of the device's
3237 * runtime_resume callback.
3238 */
3239void blk_post_runtime_resume(struct request_queue *q, int err)
3240{
3241 spin_lock_irq(q->queue_lock);
3242 if (!err) {
3243 q->rpm_status = RPM_ACTIVE;
3244 __blk_run_queue(q);
3245 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003246 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003247 } else {
3248 q->rpm_status = RPM_SUSPENDED;
3249 }
3250 spin_unlock_irq(q->queue_lock);
3251}
3252EXPORT_SYMBOL(blk_post_runtime_resume);
3253#endif
3254
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255int __init blk_dev_init(void)
3256{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003257 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3258 sizeof(((struct request *)0)->cmd_flags));
3259
Tejun Heo89b90be2011-01-03 15:01:47 +01003260 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3261 kblockd_workqueue = alloc_workqueue("kblockd",
Viresh Kumar695588f2013-04-24 17:12:56 +05303262 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3263 WQ_POWER_EFFICIENT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 if (!kblockd_workqueue)
3265 panic("Failed to create kblockd\n");
3266
3267 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003268 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269
Jens Axboe8324aa92008-01-29 14:51:59 +01003270 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003271 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 return 0;
3274}