blob: d8fba675a6e4277911eb3634fa78e4c0e12ad28a [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>
Tejun Heoeea8f412015-05-22 17:13:17 -040035#include <linux/blk-cgroup.h>
Li Zefan55782132009-06-09 13:43:05 +080036
37#define CREATE_TRACE_POINTS
38#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Jens Axboe8324aa92008-01-29 14:51:59 +010040#include "blk.h"
Ming Lei43a5e4e2013-12-26 21:31:35 +080041#include "blk-mq.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010042
Mohan Srinivasane2d88782016-12-14 15:55:36 -080043#include <linux/math64.h>
44
Mike Snitzerd07335e2010-11-16 12:52:38 +010045EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020046EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070047EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060048EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010049EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010050
Tejun Heoa73f7302011-12-14 00:33:37 +010051DEFINE_IDA(blk_queue_ida);
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
54 * For the allocated request tables
55 */
Wei Tangd674d412015-11-24 09:58:45 +080056struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58/*
59 * For queue allocation
60 */
Jens Axboe6728cb02008-01-31 13:03:55 +010061struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * Controlling structure to kblockd
65 */
Jens Axboeff856ba2006-01-09 16:02:34 +010066static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Tejun Heod40f75a2015-05-22 17:13:42 -040068static void blk_clear_congested(struct request_list *rl, int sync)
69{
Tejun Heod40f75a2015-05-22 17:13:42 -040070#ifdef CONFIG_CGROUP_WRITEBACK
71 clear_wb_congested(rl->blkg->wb_congested, sync);
72#else
Tejun Heo482cf792015-05-22 17:13:43 -040073 /*
74 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
75 * flip its congestion state for events on other blkcgs.
76 */
77 if (rl == &rl->q->root_rl)
78 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
Tejun Heod40f75a2015-05-22 17:13:42 -040079#endif
80}
81
82static void blk_set_congested(struct request_list *rl, int sync)
83{
Tejun Heod40f75a2015-05-22 17:13:42 -040084#ifdef CONFIG_CGROUP_WRITEBACK
85 set_wb_congested(rl->blkg->wb_congested, sync);
86#else
Tejun Heo482cf792015-05-22 17:13:43 -040087 /* see blk_clear_congested() */
88 if (rl == &rl->q->root_rl)
89 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
Tejun Heod40f75a2015-05-22 17:13:42 -040090#endif
91}
92
Jens Axboe8324aa92008-01-29 14:51:59 +010093void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
95 int nr;
96
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
99 nr = q->nr_requests;
100 q->nr_congestion_on = nr;
101
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 if (nr < 1)
104 nr = 1;
105 q->nr_congestion_off = nr;
106}
107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108/**
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * @bdev: device
111 *
112 * Locates the passed device's request queue and returns the address of its
Tejun Heoff9ea322014-09-08 08:03:56 +0900113 * backing_dev_info. This function can only be called if @bdev is opened
114 * and the return value is never NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 */
116struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
117{
Jens Axboe165125e2007-07-24 09:28:11 +0200118 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Tejun Heoff9ea322014-09-08 08:03:56 +0900120 return &q->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122EXPORT_SYMBOL(blk_get_backing_dev_info);
123
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200124void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200126 memset(rq, 0, sizeof(*rq));
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700129 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200130 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100131 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900132 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200133 INIT_HLIST_NODE(&rq->hash);
134 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200135 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800136 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100137 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900138 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700139 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100140 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200142EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
NeilBrown5bb23a62007-09-27 12:46:13 +0200144static void req_bio_endio(struct request *rq, struct bio *bio,
145 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100146{
Mike Snitzer78d8e582015-06-26 10:01:13 -0400147 if (error)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200148 bio->bi_error = error;
Tejun Heo797e7db2006-01-06 09:51:03 +0100149
Tejun Heo143a87f2011-01-25 12:43:52 +0100150 if (unlikely(rq->cmd_flags & REQ_QUIET))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600151 bio_set_flag(bio, BIO_QUIET);
Tejun Heo143a87f2011-01-25 12:43:52 +0100152
Kent Overstreetf79ea412012-09-20 16:38:30 -0700153 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100154
155 /* don't actually finish bio if it's part of flush sequence */
Mike Snitzer78d8e582015-06-26 10:01:13 -0400156 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200157 bio_endio(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +0100158}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160void blk_dump_rq_flags(struct request *rq, char *msg)
161{
162 int bit;
163
Jens Axboe59533162013-05-23 12:25:08 +0200164 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200165 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200166 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Tejun Heo83096eb2009-05-07 22:24:39 +0900168 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
169 (unsigned long long)blk_rq_pos(rq),
170 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600171 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
172 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200174 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100175 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200176 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 printk("%02x ", rq->cmd[bit]);
178 printk("\n");
179 }
180}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181EXPORT_SYMBOL(blk_dump_rq_flags);
182
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500183static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200184{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500185 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200186
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500187 q = container_of(work, struct request_queue, delay_work.work);
188 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200189 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500190 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500194 * blk_delay_queue - restart queueing after defined interval
195 * @q: The &struct request_queue in question
196 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 *
198 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500199 * Sometimes queueing needs to be postponed for a little while, to allow
200 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100201 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500202 */
203void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Bart Van Assche70460572012-11-28 13:45:56 +0100205 if (likely(!blk_queue_dead(q)))
206 queue_delayed_work(kblockd_workqueue, &q->delay_work,
207 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500209EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211/**
Jens Axboe21491412015-12-28 13:01:22 -0700212 * blk_start_queue_async - asynchronously restart a previously stopped queue
213 * @q: The &struct request_queue in question
214 *
215 * Description:
216 * blk_start_queue_async() will clear the stop flag on the queue, and
217 * ensure that the request_fn for the queue is run from an async
218 * context.
219 **/
220void blk_start_queue_async(struct request_queue *q)
221{
222 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
223 blk_run_queue_async(q);
224}
225EXPORT_SYMBOL(blk_start_queue_async);
226
227/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200229 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 *
231 * Description:
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
235 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200236void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200238 WARN_ON(!irqs_disabled());
239
Nick Piggin75ad23b2008-04-29 14:48:33 +0200240 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200241 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243EXPORT_SYMBOL(blk_start_queue);
244
245/**
246 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200247 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 *
249 * Description:
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
258 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200259void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Tejun Heo136b5722012-08-21 13:18:24 -0700261 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200262 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264EXPORT_SYMBOL(blk_stop_queue);
265
266/**
267 * blk_sync_queue - cancel any pending callbacks on a queue
268 * @q: the queue
269 *
270 * Description:
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200275 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * that its ->make_request_fn will not re-add plugging prior to calling
277 * this function.
278 *
Vivek Goyalda527772011-03-02 19:05:33 -0500279 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900280 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800281 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500282 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 */
284void blk_sync_queue(struct request_queue *q)
285{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100286 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800287
288 if (q->mq_ops) {
289 struct blk_mq_hw_ctx *hctx;
290 int i;
291
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600292 queue_for_each_hw_ctx(q, hctx, i) {
Jens Axboe27489a32016-08-24 15:54:25 -0600293 cancel_work_sync(&hctx->run_work);
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600294 cancel_delayed_work_sync(&hctx->delay_work);
295 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800296 } else {
297 cancel_delayed_work_sync(&q->delay_work);
298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300EXPORT_SYMBOL(blk_sync_queue);
301
302/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100303 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
304 * @q: The queue to run
305 *
306 * Description:
307 * Invoke request handling on a queue if there are any pending requests.
308 * May be used to restart request handling after a request has completed.
309 * This variant runs the queue whether or not the queue has been
310 * stopped. Must be called with the queue lock held and interrupts
311 * disabled. See also @blk_run_queue.
312 */
313inline void __blk_run_queue_uncond(struct request_queue *q)
314{
315 if (unlikely(blk_queue_dead(q)))
316 return;
317
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100318 /*
319 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
320 * the queue lock internally. As a result multiple threads may be
321 * running such a request function concurrently. Keep track of the
322 * number of active request_fn invocations such that blk_drain_queue()
323 * can wait until all these request_fn calls have finished.
324 */
325 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100326 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100327 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100328}
Christoph Hellwiga7928c12015-04-17 22:37:20 +0200329EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
Bart Van Asschec246e802012-12-06 14:32:01 +0100330
331/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200332 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200334 *
335 * Description:
336 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200337 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200339void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
Tejun Heoa538cd02009-04-23 11:05:17 +0900341 if (unlikely(blk_queue_stopped(q)))
342 return;
343
Bart Van Asschec246e802012-12-06 14:32:01 +0100344 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200345}
346EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200347
Nick Piggin75ad23b2008-04-29 14:48:33 +0200348/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200349 * blk_run_queue_async - run a single device queue in workqueue context
350 * @q: The queue to run
351 *
352 * Description:
353 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100354 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200355 */
356void blk_run_queue_async(struct request_queue *q)
357{
Bart Van Assche70460572012-11-28 13:45:56 +0100358 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700359 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200360}
Jens Axboec21e6be2011-04-19 13:32:46 +0200361EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200362
363/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200364 * blk_run_queue - run a single device queue
365 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200366 *
367 * Description:
368 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900369 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200370 */
371void blk_run_queue(struct request_queue *q)
372{
373 unsigned long flags;
374
375 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200376 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 spin_unlock_irqrestore(q->queue_lock, flags);
378}
379EXPORT_SYMBOL(blk_run_queue);
380
Jens Axboe165125e2007-07-24 09:28:11 +0200381void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500382{
383 kobject_put(&q->kobj);
384}
Jens Axboed86e0e82011-05-27 07:44:43 +0200385EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500386
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200387/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100388 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200389 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200390 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200391 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200392 * Drain requests from @q. If @drain_all is set, all requests are drained.
393 * If not, only ELVPRIV requests are drained. The caller is responsible
394 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200395 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100396static void __blk_drain_queue(struct request_queue *q, bool drain_all)
397 __releases(q->queue_lock)
398 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200399{
Asias He458f27a2012-06-15 08:45:25 +0200400 int i;
401
Bart Van Assche807592a2012-11-28 13:43:38 +0100402 lockdep_assert_held(q->queue_lock);
403
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100405 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200406
Tejun Heob855b042012-03-06 21:24:55 +0100407 /*
408 * The caller might be trying to drain @q before its
409 * elevator is initialized.
410 */
411 if (q->elevator)
412 elv_drain_elevator(q);
413
Tejun Heo5efd6112012-03-05 13:15:12 -0800414 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200415
Tejun Heo4eabc942011-12-15 20:03:04 +0100416 /*
417 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100418 * driver init after queue creation or is not yet fully
419 * active yet. Some drivers (e.g. fd and loop) get unhappy
420 * in such cases. Kick queue iff dispatch queue has
421 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100422 */
Tejun Heob855b042012-03-06 21:24:55 +0100423 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100424 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200425
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700426 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100427 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100428
429 /*
430 * Unfortunately, requests are queued at and tracked from
431 * multiple places and there's no single counter which can
432 * be drained. Check all the queues and counters.
433 */
434 if (drain_all) {
Ming Leie97c2932014-09-25 23:23:46 +0800435 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
Tejun Heo481a7d62011-12-14 00:33:37 +0100436 drain |= !list_empty(&q->queue_head);
437 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700438 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100439 drain |= q->in_flight[i];
Ming Lei7c94e1c2014-09-25 23:23:43 +0800440 if (fq)
441 drain |= !list_empty(&fq->flush_queue[i]);
Tejun Heo481a7d62011-12-14 00:33:37 +0100442 }
443 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200444
Tejun Heo481a7d62011-12-14 00:33:37 +0100445 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200446 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100447
448 spin_unlock_irq(q->queue_lock);
449
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200450 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100451
452 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200453 }
Asias He458f27a2012-06-15 08:45:25 +0200454
455 /*
456 * With queue marked dead, any woken up waiter will fail the
457 * allocation path, so the wakeup chaining is lost and we're
458 * left with hung waiters. We need to wake up those waiters.
459 */
460 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700461 struct request_list *rl;
462
Tejun Heoa0516612012-06-26 15:05:44 -0700463 blk_queue_for_each_rl(rl, q)
464 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
465 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200466 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200467}
468
Tejun Heoc9a929d2011-10-19 14:42:16 +0200469/**
Tejun Heod7325802012-03-05 13:14:58 -0800470 * blk_queue_bypass_start - enter queue bypass mode
471 * @q: queue of interest
472 *
473 * In bypass mode, only the dispatch FIFO queue of @q is used. This
474 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800475 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700476 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
477 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800478 */
479void blk_queue_bypass_start(struct request_queue *q)
480{
481 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600482 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800483 queue_flag_set(QUEUE_FLAG_BYPASS, q);
484 spin_unlock_irq(q->queue_lock);
485
Tejun Heo776687b2014-07-01 10:29:17 -0600486 /*
487 * Queues start drained. Skip actual draining till init is
488 * complete. This avoids lenghty delays during queue init which
489 * can happen many times during boot.
490 */
491 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100492 spin_lock_irq(q->queue_lock);
493 __blk_drain_queue(q, false);
494 spin_unlock_irq(q->queue_lock);
495
Tejun Heob82d4b12012-04-13 13:11:31 -0700496 /* ensure blk_queue_bypass() is %true inside RCU read lock */
497 synchronize_rcu();
498 }
Tejun Heod7325802012-03-05 13:14:58 -0800499}
500EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
501
502/**
503 * blk_queue_bypass_end - leave queue bypass mode
504 * @q: queue of interest
505 *
506 * Leave bypass mode and restore the normal queueing behavior.
507 */
508void blk_queue_bypass_end(struct request_queue *q)
509{
510 spin_lock_irq(q->queue_lock);
511 if (!--q->bypass_depth)
512 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
513 WARN_ON_ONCE(q->bypass_depth < 0);
514 spin_unlock_irq(q->queue_lock);
515}
516EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
517
Jens Axboeaed3ea92014-12-22 14:04:42 -0700518void blk_set_queue_dying(struct request_queue *q)
519{
Bart Van Assche1b856082016-08-16 16:48:36 -0700520 spin_lock_irq(q->queue_lock);
521 queue_flag_set(QUEUE_FLAG_DYING, q);
522 spin_unlock_irq(q->queue_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700523
524 if (q->mq_ops)
525 blk_mq_wake_waiters(q);
526 else {
527 struct request_list *rl;
528
529 blk_queue_for_each_rl(rl, q) {
530 if (rl->rq_pool) {
531 wake_up(&rl->wait[BLK_RW_SYNC]);
532 wake_up(&rl->wait[BLK_RW_ASYNC]);
533 }
534 }
535 }
536}
537EXPORT_SYMBOL_GPL(blk_set_queue_dying);
538
Tejun Heod7325802012-03-05 13:14:58 -0800539/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200540 * blk_cleanup_queue - shutdown a request queue
541 * @q: request queue to shutdown
542 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100543 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
544 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500545 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100546void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500547{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200548 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700549
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100550 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500551 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700552 blk_set_queue_dying(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200553 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800554
Tejun Heo80fd9972012-04-13 14:50:53 -0700555 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100556 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700557 * that, unlike blk_queue_bypass_start(), we aren't performing
558 * synchronize_rcu() after entering bypass mode to avoid the delay
559 * as some drivers create and destroy a lot of queues while
560 * probing. This is still safe because blk_release_queue() will be
561 * called only after the queue refcnt drops to zero and nothing,
562 * RCU or not, would be traversing the queue by then.
563 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800564 q->bypass_depth++;
565 queue_flag_set(QUEUE_FLAG_BYPASS, q);
566
Tejun Heoc9a929d2011-10-19 14:42:16 +0200567 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
568 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100569 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200570 spin_unlock_irq(lock);
571 mutex_unlock(&q->sysfs_lock);
572
Bart Van Asschec246e802012-12-06 14:32:01 +0100573 /*
574 * Drain all requests queued before DYING marking. Set DEAD flag to
575 * prevent that q->request_fn() gets invoked after draining finished.
576 */
Dan Williams3ef28e82015-10-21 13:20:12 -0400577 blk_freeze_queue(q);
578 spin_lock_irq(lock);
579 if (!q->mq_ops)
Ming Lei43a5e4e2013-12-26 21:31:35 +0800580 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100581 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100582 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200583
Dan Williams5a48fc12015-10-21 13:20:23 -0400584 /* for synchronous bio-based driver finish in-flight integrity i/o */
585 blk_flush_integrity();
586
Tejun Heoc9a929d2011-10-19 14:42:16 +0200587 /* @q won't process any more request, flush async actions */
588 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
589 blk_sync_queue(q);
590
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100591 if (q->mq_ops)
592 blk_mq_free_queue(q);
Dan Williams3ef28e82015-10-21 13:20:12 -0400593 percpu_ref_exit(&q->q_usage_counter);
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100594
Asias He5e5cfac2012-05-24 23:28:52 +0800595 spin_lock_irq(lock);
596 if (q->queue_lock != &q->__queue_lock)
597 q->queue_lock = &q->__queue_lock;
598 spin_unlock_irq(lock);
599
Tejun Heob02176f2015-09-08 12:20:22 -0400600 bdi_unregister(&q->backing_dev_info);
NeilBrown6cd18e72015-04-27 14:12:22 +1000601
Tejun Heoc9a929d2011-10-19 14:42:16 +0200602 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500603 blk_put_queue(q);
604}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605EXPORT_SYMBOL(blk_cleanup_queue);
606
David Rientjes271508d2015-03-24 16:21:16 -0700607/* Allocate memory local to the request queue */
608static void *alloc_request_struct(gfp_t gfp_mask, void *data)
609{
610 int nid = (int)(long)data;
611 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
612}
613
614static void free_request_struct(void *element, void *unused)
615{
616 kmem_cache_free(request_cachep, element);
617}
618
Tejun Heo5b788ce2012-06-04 20:40:59 -0700619int blk_init_rl(struct request_list *rl, struct request_queue *q,
620 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400622 if (unlikely(rl->rq_pool))
623 return 0;
624
Tejun Heo5b788ce2012-06-04 20:40:59 -0700625 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200626 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
627 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200628 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
629 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
David Rientjes271508d2015-03-24 16:21:16 -0700631 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
632 free_request_struct,
633 (void *)(long)q->node, gfp_mask,
634 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (!rl->rq_pool)
636 return -ENOMEM;
637
638 return 0;
639}
640
Tejun Heo5b788ce2012-06-04 20:40:59 -0700641void blk_exit_rl(struct request_list *rl)
642{
643 if (rl->rq_pool)
644 mempool_destroy(rl->rq_pool);
645}
646
Jens Axboe165125e2007-07-24 09:28:11 +0200647struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100649 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700650}
651EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100653int blk_queue_enter(struct request_queue *q, bool nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400654{
655 while (true) {
656 int ret;
657
658 if (percpu_ref_tryget_live(&q->q_usage_counter))
659 return 0;
660
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100661 if (nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400662 return -EBUSY;
663
664 ret = wait_event_interruptible(q->mq_freeze_wq,
665 !atomic_read(&q->mq_freeze_depth) ||
666 blk_queue_dying(q));
667 if (blk_queue_dying(q))
668 return -ENODEV;
669 if (ret)
670 return ret;
671 }
672}
673
674void blk_queue_exit(struct request_queue *q)
675{
676 percpu_ref_put(&q->q_usage_counter);
677}
678
679static void blk_queue_usage_counter_release(struct percpu_ref *ref)
680{
681 struct request_queue *q =
682 container_of(ref, struct request_queue, q_usage_counter);
683
684 wake_up_all(&q->mq_freeze_wq);
685}
686
Christoph Hellwig287922e2015-10-30 20:57:30 +0800687static void blk_rq_timed_out_timer(unsigned long data)
688{
689 struct request_queue *q = (struct request_queue *)data;
690
691 kblockd_schedule_work(&q->timeout_work);
692}
693
Jens Axboe165125e2007-07-24 09:28:11 +0200694struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700695{
Jens Axboe165125e2007-07-24 09:28:11 +0200696 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700697 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700698
Jens Axboe8324aa92008-01-29 14:51:59 +0100699 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700700 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (!q)
702 return NULL;
703
Dan Carpenter00380a42012-03-23 09:58:54 +0100704 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100705 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800706 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100707
Kent Overstreet54efd502015-04-23 22:37:18 -0700708 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
709 if (!q->bio_split)
710 goto fail_id;
711
Jens Axboe0989a022009-06-12 14:42:56 +0200712 q->backing_dev_info.ra_pages =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300713 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
Tejun Heo89e9b9e2015-05-22 17:13:36 -0400714 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
Jens Axboed9938312009-06-12 14:45:52 +0200715 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100716 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200717
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700718 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100719 if (err)
Kent Overstreet54efd502015-04-23 22:37:18 -0700720 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700721
Matthew Garrett31373d02010-04-06 14:25:14 +0200722 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
723 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700724 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100725 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700726 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100727 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800728#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800729 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800730#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500731 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500732
Jens Axboe8324aa92008-01-29 14:51:59 +0100733 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Al Viro483f4af2006-03-18 18:34:37 -0500735 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700736 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500737
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500738 /*
739 * By default initialize queue_lock to internal lock and driver can
740 * override it later if need be.
741 */
742 q->queue_lock = &q->__queue_lock;
743
Tejun Heob82d4b12012-04-13 13:11:31 -0700744 /*
745 * A queue starts its life with bypass turned on to avoid
746 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700747 * init. The initial bypass will be finished when the queue is
748 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700749 */
750 q->bypass_depth = 1;
751 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
752
Jens Axboe320ae512013-10-24 09:20:05 +0100753 init_waitqueue_head(&q->mq_freeze_wq);
754
Dan Williams3ef28e82015-10-21 13:20:12 -0400755 /*
756 * Init percpu_ref in atomic mode so that it's faster to shutdown.
757 * See blk_register_queue() for details.
758 */
759 if (percpu_ref_init(&q->q_usage_counter,
760 blk_queue_usage_counter_release,
761 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
Mikulas Patockafff49962013-10-14 12:11:36 -0400762 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800763
Dan Williams3ef28e82015-10-21 13:20:12 -0400764 if (blkcg_init_queue(q))
765 goto fail_ref;
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100768
Dan Williams3ef28e82015-10-21 13:20:12 -0400769fail_ref:
770 percpu_ref_exit(&q->q_usage_counter);
Mikulas Patockafff49962013-10-14 12:11:36 -0400771fail_bdi:
772 bdi_destroy(&q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700773fail_split:
774 bioset_free(q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100775fail_id:
776 ida_simple_remove(&blk_queue_ida, q->id);
777fail_q:
778 kmem_cache_free(blk_requestq_cachep, q);
779 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
Christoph Lameter19460892005-06-23 00:08:19 -0700781EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783/**
784 * blk_init_queue - prepare a request queue for use with a block device
785 * @rfn: The function to be called to process requests that have been
786 * placed on the queue.
787 * @lock: Request queue spin lock
788 *
789 * Description:
790 * If a block device wishes to use the standard request handling procedures,
791 * which sorts requests and coalesces adjacent requests, then it must
792 * call blk_init_queue(). The function @rfn will be called when there
793 * are requests on the queue that need to be processed. If the device
794 * supports plugging, then @rfn may not be called immediately when requests
795 * are available on the queue, but may be called at some time later instead.
796 * Plugged queues are generally unplugged when a buffer belonging to one
797 * of the requests on the queue is needed, or due to memory pressure.
798 *
799 * @rfn is not required, or even expected, to remove all requests off the
800 * queue, but only as many as it can handle at a time. If it does leave
801 * requests on the queue, it is responsible for arranging that the requests
802 * get dealt with eventually.
803 *
804 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200805 * request queue; this lock will be taken also from interrupt context, so irq
806 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200808 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 * it didn't succeed.
810 *
811 * Note:
812 * blk_init_queue() must be paired with a blk_cleanup_queue() call
813 * when the block device is deactivated (such as at module unload).
814 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700815
Jens Axboe165125e2007-07-24 09:28:11 +0200816struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100818 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700819}
820EXPORT_SYMBOL(blk_init_queue);
821
Jens Axboe165125e2007-07-24 09:28:11 +0200822struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700823blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
824{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600825 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600827 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
828 if (!uninit_q)
829 return NULL;
830
Mike Snitzer51514122011-11-23 10:59:13 +0100831 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600832 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700833 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700834
Mike Snitzer7982e902014-03-08 17:20:01 -0700835 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200836}
837EXPORT_SYMBOL(blk_init_queue_node);
838
Jens Axboedece1632015-11-05 10:41:16 -0700839static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
Mike Snitzer336b7e12015-05-11 14:06:32 -0400840
Mike Snitzer01effb02010-05-11 08:57:42 +0200841struct request_queue *
842blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
843 spinlock_t *lock)
844{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (!q)
846 return NULL;
847
Ming Leif70ced02014-09-25 23:23:47 +0800848 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800849 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700850 return NULL;
851
Tejun Heoa0516612012-06-26 15:05:44 -0700852 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600853 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Christoph Hellwig287922e2015-10-30 20:57:30 +0800855 INIT_WORK(&q->timeout_work, blk_timeout_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900858 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700859 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500860
861 /* Override internal queue lock with supplied lock pointer */
862 if (lock)
863 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Jens Axboef3b144a2009-03-06 08:48:33 +0100865 /*
866 * This also sets hw/phys segments, boundary and size
867 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200868 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Alan Stern44ec9542007-02-20 11:01:57 -0500870 q->sg_reserved_size = INT_MAX;
871
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600872 /* Protect q->elevator from elevator_change */
873 mutex_lock(&q->sysfs_lock);
874
Tejun Heob82d4b12012-04-13 13:11:31 -0700875 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600876 if (elevator_init(q, NULL)) {
877 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600878 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600879 }
880
881 mutex_unlock(&q->sysfs_lock);
882
Tejun Heob82d4b12012-04-13 13:11:31 -0700883 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600884
885fail:
Ming Leiba483382014-09-25 23:23:44 +0800886 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600887 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
Mike Snitzer51514122011-11-23 10:59:13 +0100889EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Tejun Heo09ac46c2011-12-14 00:33:38 +0100891bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100893 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100894 __blk_get_queue(q);
895 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 }
897
Tejun Heo09ac46c2011-12-14 00:33:38 +0100898 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
Jens Axboed86e0e82011-05-27 07:44:43 +0200900EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Tejun Heo5b788ce2012-06-04 20:40:59 -0700902static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100904 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700905 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100906 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100907 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100908 }
909
Tejun Heo5b788ce2012-06-04 20:40:59 -0700910 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911}
912
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913/*
914 * ioc_batching returns true if the ioc is a valid batching request and
915 * should be given priority access to a request.
916 */
Jens Axboe165125e2007-07-24 09:28:11 +0200917static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918{
919 if (!ioc)
920 return 0;
921
922 /*
923 * Make sure the process is able to allocate at least 1 request
924 * even if the batch times out, otherwise we could theoretically
925 * lose wakeups.
926 */
927 return ioc->nr_batch_requests == q->nr_batching ||
928 (ioc->nr_batch_requests > 0
929 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
930}
931
932/*
933 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
934 * will cause the process to be a "batcher" on all queues in the system. This
935 * is the behaviour we want though - once it gets a wakeup it should be given
936 * a nice run.
937 */
Jens Axboe165125e2007-07-24 09:28:11 +0200938static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
940 if (!ioc || ioc_batching(q, ioc))
941 return;
942
943 ioc->nr_batch_requests = q->nr_batching;
944 ioc->last_waited = jiffies;
945}
946
Tejun Heo5b788ce2012-06-04 20:40:59 -0700947static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700949 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Tejun Heod40f75a2015-05-22 17:13:42 -0400951 if (rl->count[sync] < queue_congestion_off_threshold(q))
952 blk_clear_congested(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Jens Axboe1faa16d2009-04-06 14:48:01 +0200954 if (rl->count[sync] + 1 <= q->nr_requests) {
955 if (waitqueue_active(&rl->wait[sync]))
956 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Tejun Heo5b788ce2012-06-04 20:40:59 -0700958 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960}
961
962/*
963 * A request has just been released. Account for it, update the full and
964 * congestion status, wake up any waiters. Called under q->queue_lock.
965 */
Mike Christiee6a40b02016-06-05 14:32:11 -0500966static void freed_request(struct request_list *rl, int op, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700968 struct request_queue *q = rl->q;
Mike Christied9d8c5c2016-06-05 14:32:16 -0500969 int sync = rw_is_sync(op, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700971 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200972 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200973 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700974 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Tejun Heo5b788ce2012-06-04 20:40:59 -0700976 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Jens Axboe1faa16d2009-04-06 14:48:01 +0200978 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700979 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980}
981
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600982int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
983{
984 struct request_list *rl;
Tejun Heod40f75a2015-05-22 17:13:42 -0400985 int on_thresh, off_thresh;
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600986
987 spin_lock_irq(q->queue_lock);
988 q->nr_requests = nr;
989 blk_queue_congestion_threshold(q);
Tejun Heod40f75a2015-05-22 17:13:42 -0400990 on_thresh = queue_congestion_on_threshold(q);
991 off_thresh = queue_congestion_off_threshold(q);
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600992
993 blk_queue_for_each_rl(rl, q) {
Tejun Heod40f75a2015-05-22 17:13:42 -0400994 if (rl->count[BLK_RW_SYNC] >= on_thresh)
995 blk_set_congested(rl, BLK_RW_SYNC);
996 else if (rl->count[BLK_RW_SYNC] < off_thresh)
997 blk_clear_congested(rl, BLK_RW_SYNC);
998
999 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1000 blk_set_congested(rl, BLK_RW_ASYNC);
1001 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1002 blk_clear_congested(rl, BLK_RW_ASYNC);
1003
Jens Axboee3a2b3f2014-05-20 11:49:02 -06001004 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1005 blk_set_rl_full(rl, BLK_RW_SYNC);
1006 } else {
1007 blk_clear_rl_full(rl, BLK_RW_SYNC);
1008 wake_up(&rl->wait[BLK_RW_SYNC]);
1009 }
1010
1011 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1012 blk_set_rl_full(rl, BLK_RW_ASYNC);
1013 } else {
1014 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1015 wake_up(&rl->wait[BLK_RW_ASYNC]);
1016 }
1017 }
1018
1019 spin_unlock_irq(q->queue_lock);
1020 return 0;
1021}
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001024 * Determine if elevator data should be initialized when allocating the
1025 * request associated with @bio.
1026 */
1027static bool blk_rq_should_init_elevator(struct bio *bio)
1028{
1029 if (!bio)
1030 return true;
1031
1032 /*
1033 * Flush requests do not use the elevator so skip initialization.
1034 * This allows a request to share the flush and elevator data.
1035 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001036 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001037 return false;
1038
1039 return true;
1040}
1041
Tejun Heoda8303c2011-10-19 14:33:05 +02001042/**
Tejun Heo852c7882012-03-05 13:15:27 -08001043 * rq_ioc - determine io_context for request allocation
1044 * @bio: request being allocated is for this bio (can be %NULL)
1045 *
1046 * Determine io_context to use for request allocation for @bio. May return
1047 * %NULL if %current->io_context doesn't exist.
1048 */
1049static struct io_context *rq_ioc(struct bio *bio)
1050{
1051#ifdef CONFIG_BLK_CGROUP
1052 if (bio && bio->bi_ioc)
1053 return bio->bi_ioc;
1054#endif
1055 return current->io_context;
1056}
1057
1058/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001059 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -07001060 * @rl: request list to allocate from
Mike Christiee6a40b02016-06-05 14:32:11 -05001061 * @op: REQ_OP_READ/REQ_OP_WRITE
1062 * @op_flags: rq_flag_bits
Tejun Heoda8303c2011-10-19 14:33:05 +02001063 * @bio: bio to allocate request for (can be %NULL)
1064 * @gfp_mask: allocation mask
1065 *
1066 * Get a free request from @q. This function may fail under memory
1067 * pressure or if @q is dead.
1068 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001069 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001070 * Returns ERR_PTR on failure, with @q->queue_lock held.
1071 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001073static struct request *__get_request(struct request_list *rl, int op,
1074 int op_flags, struct bio *bio,
1075 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
Tejun Heo5b788ce2012-06-04 20:40:59 -07001077 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -08001078 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001079 struct elevator_type *et = q->elevator->type;
1080 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +01001081 struct io_cq *icq = NULL;
Mike Christied9d8c5c2016-06-05 14:32:16 -05001082 const bool is_sync = rw_is_sync(op, op_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +02001083 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001085 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -06001086 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +02001087
Mike Christieba568ea2016-06-05 14:32:13 -05001088 may_queue = elv_may_queue(q, op, op_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001089 if (may_queue == ELV_MQUEUE_NO)
1090 goto rq_starved;
1091
Jens Axboe1faa16d2009-04-06 14:48:01 +02001092 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1093 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +01001094 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001095 * The queue will fill after this allocation, so set
1096 * it as full, and mark this process as "batching".
1097 * This process will be allowed to complete a batch of
1098 * requests, others will be blocked.
1099 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001100 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001101 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001102 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001103 } else {
1104 if (may_queue != ELV_MQUEUE_MUST
1105 && !ioc_batching(q, ioc)) {
1106 /*
1107 * The queue is full and the allocating
1108 * process is not a "batcher", and not
1109 * exempted by the IO scheduler
1110 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001111 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001112 }
1113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
Tejun Heod40f75a2015-05-22 17:13:42 -04001115 blk_set_congested(rl, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 }
1117
Jens Axboe082cf692005-06-28 16:35:11 +02001118 /*
1119 * Only allow batching queuers to allocate up to 50% over the defined
1120 * limit of requests, otherwise we could have thousands of requests
1121 * allocated with any setting of ->nr_requests
1122 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001123 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001124 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001125
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001126 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001127 rl->count[is_sync]++;
1128 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001129
Tejun Heof1f8cc92011-12-14 00:33:42 +01001130 /*
1131 * Decide whether the new request will be managed by elevator. If
Mike Christiee6a40b02016-06-05 14:32:11 -05001132 * so, mark @op_flags and increment elvpriv. Non-zero elvpriv will
Tejun Heof1f8cc92011-12-14 00:33:42 +01001133 * prevent the current elevator from being destroyed until the new
1134 * request is freed. This guarantees icq's won't be destroyed and
1135 * makes creating new ones safe.
1136 *
1137 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1138 * it will be created after releasing queue_lock.
1139 */
Tejun Heod7325802012-03-05 13:14:58 -08001140 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Mike Christiee6a40b02016-06-05 14:32:11 -05001141 op_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001142 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001143 if (et->icq_cache && ioc)
1144 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001145 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001146
Jens Axboef253b862010-10-24 22:06:02 +02001147 if (blk_queue_io_stat(q))
Mike Christiee6a40b02016-06-05 14:32:11 -05001148 op_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 spin_unlock_irq(q->queue_lock);
1150
Tejun Heo29e2b092012-04-19 16:29:21 -07001151 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001152 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001153 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001154 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Tejun Heo29e2b092012-04-19 16:29:21 -07001156 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001157 blk_rq_set_rl(rq, rl);
Mike Christiee6a40b02016-06-05 14:32:11 -05001158 req_set_op_attrs(rq, op, op_flags | REQ_ALLOCED);
Tejun Heo29e2b092012-04-19 16:29:21 -07001159
Tejun Heoaaf7c682012-04-19 16:29:22 -07001160 /* init elvpriv */
Mike Christiee6a40b02016-06-05 14:32:11 -05001161 if (op_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001162 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001163 if (ioc)
1164 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001165 if (!icq)
1166 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001167 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001168
1169 rq->elv.icq = icq;
1170 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1171 goto fail_elvpriv;
1172
1173 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001174 if (icq)
1175 get_io_context(icq->ioc);
1176 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001177out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001178 /*
1179 * ioc may be NULL here, and ioc_batching will be false. That's
1180 * OK, if the queue is under the request limit then requests need
1181 * not count toward the nr_batch_requests limit. There will always
1182 * be some limit enforced by BLK_BATCH_TIME.
1183 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (ioc_batching(q, ioc))
1185 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001186
Mike Christiee6a40b02016-06-05 14:32:11 -05001187 trace_block_getrq(q, bio, op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001189
Tejun Heoaaf7c682012-04-19 16:29:22 -07001190fail_elvpriv:
1191 /*
1192 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1193 * and may fail indefinitely under memory pressure and thus
1194 * shouldn't stall IO. Treat this request as !elvpriv. This will
1195 * disturb iosched and blkcg but weird is bettern than dead.
1196 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001197 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1198 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001199
1200 rq->cmd_flags &= ~REQ_ELVPRIV;
1201 rq->elv.icq = NULL;
1202
1203 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001204 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001205 spin_unlock_irq(q->queue_lock);
1206 goto out;
1207
Tejun Heob6792812012-03-05 13:15:23 -08001208fail_alloc:
1209 /*
1210 * Allocation failed presumably due to memory. Undo anything we
1211 * might have messed up.
1212 *
1213 * Allocating task should really be put onto the front of the wait
1214 * queue, but this is pretty rare.
1215 */
1216 spin_lock_irq(q->queue_lock);
Mike Christiee6a40b02016-06-05 14:32:11 -05001217 freed_request(rl, op, op_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001218
1219 /*
1220 * in the very unlikely event that allocation failed and no
1221 * requests for this direction was pending, mark us starved so that
1222 * freeing of a request in the other direction will notice
1223 * us. another possible fix would be to split the rq mempool into
1224 * READ and WRITE
1225 */
1226rq_starved:
1227 if (unlikely(rl->count[is_sync] == 0))
1228 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001229 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
Tejun Heoda8303c2011-10-19 14:33:05 +02001232/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001233 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001234 * @q: request_queue to allocate request from
Mike Christiee6a40b02016-06-05 14:32:11 -05001235 * @op: REQ_OP_READ/REQ_OP_WRITE
1236 * @op_flags: rq_flag_bits
Tejun Heoda8303c2011-10-19 14:33:05 +02001237 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001238 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001239 *
Mel Gormand0164ad2015-11-06 16:28:21 -08001240 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1241 * this function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001242 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001243 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001244 * Returns ERR_PTR on failure, with @q->queue_lock held.
1245 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001247static struct request *get_request(struct request_queue *q, int op,
1248 int op_flags, struct bio *bio,
1249 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250{
Mike Christied9d8c5c2016-06-05 14:32:16 -05001251 const bool is_sync = rw_is_sync(op, op_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001252 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001253 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001255
1256 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001257retry:
Mike Christiee6a40b02016-06-05 14:32:11 -05001258 rq = __get_request(rl, op, op_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001259 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001260 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Mel Gormand0164ad2015-11-06 16:28:21 -08001262 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001263 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001264 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Tejun Heoa06e05e2012-06-04 20:40:55 -07001267 /* wait on @rl and retry */
1268 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1269 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001270
Mike Christiee6a40b02016-06-05 14:32:11 -05001271 trace_block_sleeprq(q, bio, op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Tejun Heoa06e05e2012-06-04 20:40:55 -07001273 spin_unlock_irq(q->queue_lock);
1274 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Tejun Heoa06e05e2012-06-04 20:40:55 -07001276 /*
1277 * After sleeping, we become a "batching" process and will be able
1278 * to allocate at least one request, and up to a big batch of them
1279 * for a small period time. See ioc_batching, ioc_set_batching
1280 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001281 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Tejun Heoa06e05e2012-06-04 20:40:55 -07001283 spin_lock_irq(q->queue_lock);
1284 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001285
Tejun Heoa06e05e2012-06-04 20:40:55 -07001286 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
Jens Axboe320ae512013-10-24 09:20:05 +01001289static struct request *blk_old_get_request(struct request_queue *q, int rw,
1290 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291{
1292 struct request *rq;
1293
1294 BUG_ON(rw != READ && rw != WRITE);
1295
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001296 /* create ioc upfront */
1297 create_io_context(gfp_mask, q->node);
1298
Nick Piggind6344532005-06-28 20:45:14 -07001299 spin_lock_irq(q->queue_lock);
Mike Christiee6a40b02016-06-05 14:32:11 -05001300 rq = get_request(q, rw, 0, NULL, gfp_mask);
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001301 if (IS_ERR(rq)) {
Tejun Heoda8303c2011-10-19 14:33:05 +02001302 spin_unlock_irq(q->queue_lock);
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001303 return rq;
1304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001306 /* q->queue_lock is unlocked at this point */
1307 rq->__data_len = 0;
1308 rq->__sector = (sector_t) -1;
1309 rq->bio = rq->biotail = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return rq;
1311}
Jens Axboe320ae512013-10-24 09:20:05 +01001312
1313struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1314{
1315 if (q->mq_ops)
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01001316 return blk_mq_alloc_request(q, rw,
1317 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1318 0 : BLK_MQ_REQ_NOWAIT);
Jens Axboe320ae512013-10-24 09:20:05 +01001319 else
1320 return blk_old_get_request(q, rw, gfp_mask);
1321}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322EXPORT_SYMBOL(blk_get_request);
1323
1324/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001325 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001326 * @rq: request to be initialized
1327 *
1328 */
1329void blk_rq_set_block_pc(struct request *rq)
1330{
1331 rq->cmd_type = REQ_TYPE_BLOCK_PC;
Jens Axboef27b0872014-06-06 07:57:37 -06001332 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001333}
1334EXPORT_SYMBOL(blk_rq_set_block_pc);
1335
1336/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * blk_requeue_request - put a request back on queue
1338 * @q: request queue where request should be inserted
1339 * @rq: request to be inserted
1340 *
1341 * Description:
1342 * Drivers often keep queueing requests until the hardware cannot accept
1343 * more, when that condition happens we need to put the request back
1344 * on the queue. Must be called with queue lock held.
1345 */
Jens Axboe165125e2007-07-24 09:28:11 +02001346void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001348 blk_delete_timer(rq);
1349 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001350 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001351
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001352 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 blk_queue_end_tag(q, rq);
1354
James Bottomleyba396a62009-05-27 14:17:08 +02001355 BUG_ON(blk_queued_rq(rq));
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 elv_requeue_request(q, rq);
1358}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359EXPORT_SYMBOL(blk_requeue_request);
1360
Jens Axboe73c10102011-03-08 13:19:51 +01001361static void add_acct_request(struct request_queue *q, struct request *rq,
1362 int where)
1363{
Jens Axboe320ae512013-10-24 09:20:05 +01001364 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001365 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001366}
1367
Tejun Heo074a7ac2008-08-25 19:56:14 +09001368static void part_round_stats_single(int cpu, struct hd_struct *part,
1369 unsigned long now)
1370{
Jens Axboe7276d022014-05-09 15:48:23 -06001371 int inflight;
1372
Tejun Heo074a7ac2008-08-25 19:56:14 +09001373 if (now == part->stamp)
1374 return;
1375
Jens Axboe7276d022014-05-09 15:48:23 -06001376 inflight = part_in_flight(part);
1377 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001378 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001379 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001380 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1381 }
1382 part->stamp = now;
1383}
1384
1385/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001386 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1387 * @cpu: cpu number for stats access
1388 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 *
1390 * The average IO queue length and utilisation statistics are maintained
1391 * by observing the current state of the queue length and the amount of
1392 * time it has been in this state for.
1393 *
1394 * Normally, that accounting is done on IO completion, but that can result
1395 * in more than a second's worth of IO being accounted for within any one
1396 * second, leading to >100% utilisation. To deal with that, we call this
1397 * function to do a round-off before returning the results when reading
1398 * /proc/diskstats. This accounts immediately for all queue usage up to
1399 * the current jiffies and restarts the counters again.
1400 */
Tejun Heoc9959052008-08-25 19:47:21 +09001401void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001402{
1403 unsigned long now = jiffies;
1404
Tejun Heo074a7ac2008-08-25 19:56:14 +09001405 if (part->partno)
1406 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1407 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001408}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001409EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001410
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001411#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001412static void blk_pm_put_request(struct request *rq)
1413{
1414 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1415 pm_runtime_mark_last_busy(rq->q->dev);
1416}
1417#else
1418static inline void blk_pm_put_request(struct request *rq) {}
1419#endif
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421/*
1422 * queue lock must be held
1423 */
Jens Axboe165125e2007-07-24 09:28:11 +02001424void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (unlikely(!q))
1427 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001429 if (q->mq_ops) {
1430 blk_mq_free_request(req);
1431 return;
1432 }
1433
Lin Mingc8158812013-03-23 11:42:27 +08001434 blk_pm_put_request(req);
1435
Tejun Heo8922e162005-10-20 16:23:44 +02001436 elv_completed_request(q, req);
1437
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001438 /* this is a bio leak */
1439 WARN_ON(req->bio != NULL);
1440
AnilKumar Chimata7214d7e2017-06-23 03:09:59 -07001441 /* this is a bio leak if the bio is not tagged with BIO_DONTFREE */
1442 WARN_ON(req->bio && !bio_flagged(req->bio, BIO_DONTFREE));
1443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 /*
1445 * Request may not have originated from ll_rw_blk. if not,
1446 * it didn't come out of our reserved rq pools
1447 */
Jens Axboe49171e52006-08-10 08:59:11 +02001448 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001449 unsigned int flags = req->cmd_flags;
Mike Christiee6a40b02016-06-05 14:32:11 -05001450 int op = req_op(req);
Tejun Heoa0516612012-06-26 15:05:44 -07001451 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001454 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Tejun Heoa0516612012-06-26 15:05:44 -07001456 blk_free_request(rl, req);
Mike Christiee6a40b02016-06-05 14:32:11 -05001457 freed_request(rl, op, flags);
Tejun Heoa0516612012-06-26 15:05:44 -07001458 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001461EXPORT_SYMBOL_GPL(__blk_put_request);
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463void blk_put_request(struct request *req)
1464{
Jens Axboe165125e2007-07-24 09:28:11 +02001465 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Jens Axboe320ae512013-10-24 09:20:05 +01001467 if (q->mq_ops)
1468 blk_mq_free_request(req);
1469 else {
1470 unsigned long flags;
1471
1472 spin_lock_irqsave(q->queue_lock, flags);
1473 __blk_put_request(q, req);
1474 spin_unlock_irqrestore(q->queue_lock, flags);
1475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477EXPORT_SYMBOL(blk_put_request);
1478
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001479/**
1480 * blk_add_request_payload - add a payload to a request
1481 * @rq: request to update
1482 * @page: page backing the payload
Ming Lin37e58232016-03-22 00:24:44 -07001483 * @offset: offset in page
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001484 * @len: length of the payload.
1485 *
1486 * This allows to later add a payload to an already submitted request by
1487 * a block driver. The driver needs to take care of freeing the payload
1488 * itself.
1489 *
1490 * Note that this is a quite horrible hack and nothing but handling of
1491 * discard requests should ever use it.
1492 */
1493void blk_add_request_payload(struct request *rq, struct page *page,
Ming Lin37e58232016-03-22 00:24:44 -07001494 int offset, unsigned int len)
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001495{
1496 struct bio *bio = rq->bio;
1497
1498 bio->bi_io_vec->bv_page = page;
Ming Lin37e58232016-03-22 00:24:44 -07001499 bio->bi_io_vec->bv_offset = offset;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001500 bio->bi_io_vec->bv_len = len;
1501
Kent Overstreet4f024f32013-10-11 15:44:27 -07001502 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001503 bio->bi_vcnt = 1;
1504 bio->bi_phys_segments = 1;
1505
1506 rq->__data_len = rq->resid_len = len;
1507 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001508}
1509EXPORT_SYMBOL_GPL(blk_add_request_payload);
1510
Jens Axboe320ae512013-10-24 09:20:05 +01001511bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1512 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001513{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001514 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001515
Jens Axboe73c10102011-03-08 13:19:51 +01001516 if (!ll_back_merge_fn(q, req, bio))
1517 return false;
1518
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001519 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001520
1521 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1522 blk_rq_set_mixed_merge(req);
1523
1524 req->biotail->bi_next = bio;
1525 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001526 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001527 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1528
Jens Axboe320ae512013-10-24 09:20:05 +01001529 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001530 return true;
1531}
1532
Jens Axboe320ae512013-10-24 09:20:05 +01001533bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1534 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001535{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001536 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001537
Jens Axboe73c10102011-03-08 13:19:51 +01001538 if (!ll_front_merge_fn(q, req, bio))
1539 return false;
1540
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001541 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001542
1543 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1544 blk_rq_set_mixed_merge(req);
1545
Jens Axboe73c10102011-03-08 13:19:51 +01001546 bio->bi_next = req->bio;
1547 req->bio = bio;
1548
Kent Overstreet4f024f32013-10-11 15:44:27 -07001549 req->__sector = bio->bi_iter.bi_sector;
1550 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001551 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1552
Jens Axboe320ae512013-10-24 09:20:05 +01001553 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001554 return true;
1555}
1556
Tejun Heobd87b582011-10-19 14:33:08 +02001557/**
Jens Axboe320ae512013-10-24 09:20:05 +01001558 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001559 * @q: request_queue new bio is being queued at
1560 * @bio: new bio being queued
1561 * @request_count: out parameter for number of traversed plugged requests
Randy Dunlapccc26002015-10-30 18:36:16 -07001562 * @same_queue_rq: pointer to &struct request that gets filled in when
1563 * another request associated with @q is found on the plug list
1564 * (optional, may be %NULL)
Tejun Heobd87b582011-10-19 14:33:08 +02001565 *
1566 * Determine whether @bio being queued on @q can be merged with a request
1567 * on %current's plugged list. Returns %true if merge was successful,
1568 * otherwise %false.
1569 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001570 * Plugging coalesces IOs from the same issuer for the same purpose without
1571 * going through @q->queue_lock. As such it's more of an issuing mechanism
1572 * than scheduling, and the request, while may have elvpriv data, is not
1573 * added on the elevator at this point. In addition, we don't have
1574 * reliable access to the elevator outside queue lock. Only check basic
1575 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001576 *
1577 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001578 */
Jens Axboe320ae512013-10-24 09:20:05 +01001579bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Shaohua Li5b3f3412015-05-08 10:51:33 -07001580 unsigned int *request_count,
1581 struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +01001582{
1583 struct blk_plug *plug;
1584 struct request *rq;
1585 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001586 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001587
Tejun Heobd87b582011-10-19 14:33:08 +02001588 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001589 if (!plug)
1590 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001591 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001592
Shaohua Li92f399c2013-10-29 12:01:03 -06001593 if (q->mq_ops)
1594 plug_list = &plug->mq_list;
1595 else
1596 plug_list = &plug->list;
1597
1598 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001599 int el_ret;
1600
Shaohua Li5b3f3412015-05-08 10:51:33 -07001601 if (rq->q == q) {
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001602 (*request_count)++;
Shaohua Li5b3f3412015-05-08 10:51:33 -07001603 /*
1604 * Only blk-mq multiple hardware queues case checks the
1605 * rq in the same queue, there should be only one such
1606 * rq in a queue
1607 **/
1608 if (same_queue_rq)
1609 *same_queue_rq = rq;
1610 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001611
Tejun Heo07c2bd32012-02-08 09:19:42 +01001612 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001613 continue;
1614
Tejun Heo050c8ea2012-02-08 09:19:38 +01001615 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001616 if (el_ret == ELEVATOR_BACK_MERGE) {
1617 ret = bio_attempt_back_merge(q, rq, bio);
1618 if (ret)
1619 break;
1620 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1621 ret = bio_attempt_front_merge(q, rq, bio);
1622 if (ret)
1623 break;
1624 }
1625 }
1626out:
1627 return ret;
1628}
1629
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001630unsigned int blk_plug_queued_count(struct request_queue *q)
1631{
1632 struct blk_plug *plug;
1633 struct request *rq;
1634 struct list_head *plug_list;
1635 unsigned int ret = 0;
1636
1637 plug = current->plug;
1638 if (!plug)
1639 goto out;
1640
1641 if (q->mq_ops)
1642 plug_list = &plug->mq_list;
1643 else
1644 plug_list = &plug->list;
1645
1646 list_for_each_entry(rq, plug_list, queuelist) {
1647 if (rq->q == q)
1648 ret++;
1649 }
1650out:
1651 return ret;
1652}
1653
Jens Axboe86db1e22008-01-29 14:53:40 +01001654void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001655{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001656 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001657
Jens Axboe1eff9d32016-08-05 15:35:16 -06001658 req->cmd_flags |= bio->bi_opf & REQ_COMMON_MASK;
1659 if (bio->bi_opf & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001660 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001661
Tejun Heo52d9e672006-01-06 09:49:58 +01001662 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001663 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001664 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001665 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001666}
1667
Jens Axboedece1632015-11-05 10:41:16 -07001668static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001670 const bool sync = !!(bio->bi_opf & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001671 struct blk_plug *plug;
Mike Christiee6a40b02016-06-05 14:32:11 -05001672 int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
Jens Axboe73c10102011-03-08 13:19:51 +01001673 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001674 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 /*
1677 * low level driver can indicate that it wants pages above a
1678 * certain limit bounced to low memory (ie for highmem, or even
1679 * ISA dma in theory)
1680 */
1681 blk_queue_bounce(q, &bio);
1682
Junichi Nomura23688bf2015-12-22 10:23:44 -07001683 blk_queue_split(q, &bio, q->bio_split);
1684
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001685 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001686 bio->bi_error = -EIO;
1687 bio_endio(bio);
Jens Axboedece1632015-11-05 10:41:16 -07001688 return BLK_QC_T_NONE;
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001689 }
1690
Jens Axboe1eff9d32016-08-05 15:35:16 -06001691 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001692 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001693 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001694 goto get_rq;
1695 }
1696
Jens Axboe73c10102011-03-08 13:19:51 +01001697 /*
1698 * Check if we can merge with the plugged list before grabbing
1699 * any locks.
1700 */
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001701 if (!blk_queue_nomerges(q)) {
1702 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
Jens Axboedece1632015-11-05 10:41:16 -07001703 return BLK_QC_T_NONE;
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001704 } else
1705 request_count = blk_plug_queued_count(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001706
1707 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001710 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001711 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001712 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001713 if (!attempt_back_merge(q, req))
1714 elv_merged_request(q, req, el_ret);
1715 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001716 }
Jens Axboe73c10102011-03-08 13:19:51 +01001717 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001718 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001719 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001720 if (!attempt_front_merge(q, req))
1721 elv_merged_request(q, req, el_ret);
1722 goto out_unlock;
1723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 }
1725
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001727 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001728 * This sync check and mask will be re-done in init_request_from_bio(),
1729 * but we need to set it earlier to expose the sync flag to the
1730 * rq allocator and io schedulers.
1731 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001732 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001733 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001734
1735 /*
Jens Axboeb8269db2016-06-09 15:47:29 -06001736 * Add in META/PRIO flags, if set, before we get to the IO scheduler
1737 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001738 rw_flags |= (bio->bi_opf & (REQ_META | REQ_PRIO));
Jens Axboeb8269db2016-06-09 15:47:29 -06001739
1740 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001741 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001742 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001743 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001744 req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001745 if (IS_ERR(req)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001746 bio->bi_error = PTR_ERR(req);
1747 bio_endio(bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001748 goto out_unlock;
1749 }
Nick Piggind6344532005-06-28 20:45:14 -07001750
Nick Piggin450991b2005-06-28 20:45:13 -07001751 /*
1752 * After dropping the lock and possibly sleeping here, our request
1753 * may now be mergeable after it had proven unmergeable (above).
1754 * We don't worry about that case for efficiency. It won't happen
1755 * often, and the elevators are able to handle it.
1756 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001757 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Tao Ma9562ad92011-10-24 16:11:30 +02001759 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001760 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001761
Jens Axboe73c10102011-03-08 13:19:51 +01001762 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001763 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001764 /*
1765 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001766 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001767 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001768 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001769 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001770 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001771 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001772 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001773 trace_block_plug(q);
1774 }
Jens Axboe73c10102011-03-08 13:19:51 +01001775 }
Shaohua Lia6327162011-08-24 16:04:32 +02001776 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001777 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001778 } else {
1779 spin_lock_irq(q->queue_lock);
1780 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001781 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001782out_unlock:
1783 spin_unlock_irq(q->queue_lock);
1784 }
Jens Axboedece1632015-11-05 10:41:16 -07001785
1786 return BLK_QC_T_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787}
1788
1789/*
1790 * If bio->bi_dev is a partition, remap the location
1791 */
1792static inline void blk_partition_remap(struct bio *bio)
1793{
1794 struct block_device *bdev = bio->bi_bdev;
1795
Jens Axboebf2de6f2007-09-27 13:01:25 +02001796 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001798
Kent Overstreet4f024f32013-10-11 15:44:27 -07001799 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001801
Mike Snitzerd07335e2010-11-16 12:52:38 +01001802 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1803 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001804 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 }
1806}
1807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808static void handle_bad_sector(struct bio *bio)
1809{
1810 char b[BDEVNAME_SIZE];
1811
1812 printk(KERN_INFO "attempt to access beyond end of device\n");
Mike Christie6296b962016-06-05 14:32:21 -05001813 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 bdevname(bio->bi_bdev, b),
Jens Axboe1eff9d32016-08-05 15:35:16 -06001815 bio->bi_opf,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001816 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001817 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818}
1819
Akinobu Mitac17bb492006-12-08 02:39:46 -08001820#ifdef CONFIG_FAIL_MAKE_REQUEST
1821
1822static DECLARE_FAULT_ATTR(fail_make_request);
1823
1824static int __init setup_fail_make_request(char *str)
1825{
1826 return setup_fault_attr(&fail_make_request, str);
1827}
1828__setup("fail_make_request=", setup_fail_make_request);
1829
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001830static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001831{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001832 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001833}
1834
1835static int __init fail_make_request_debugfs(void)
1836{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001837 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1838 NULL, &fail_make_request);
1839
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001840 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001841}
1842
1843late_initcall(fail_make_request_debugfs);
1844
1845#else /* CONFIG_FAIL_MAKE_REQUEST */
1846
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001847static inline bool should_fail_request(struct hd_struct *part,
1848 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001849{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001850 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001851}
1852
1853#endif /* CONFIG_FAIL_MAKE_REQUEST */
1854
Jens Axboec07e2b42007-07-18 13:27:58 +02001855/*
1856 * Check whether this bio extends beyond the end of the device.
1857 */
1858static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1859{
1860 sector_t maxsector;
1861
1862 if (!nr_sectors)
1863 return 0;
1864
1865 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001866 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001867 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001868 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001869
1870 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1871 /*
1872 * This may well happen - the kernel calls bread()
1873 * without checking the size of the device, e.g., when
1874 * mounting a device.
1875 */
1876 handle_bad_sector(bio);
1877 return 1;
1878 }
1879 }
1880
1881 return 0;
1882}
1883
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001884static noinline_for_stack bool
1885generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886{
Jens Axboe165125e2007-07-24 09:28:11 +02001887 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001888 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001889 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001890 char b[BDEVNAME_SIZE];
1891 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
1893 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
Jens Axboec07e2b42007-07-18 13:27:58 +02001895 if (bio_check_eod(bio, nr_sectors))
1896 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001898 q = bdev_get_queue(bio->bi_bdev);
1899 if (unlikely(!q)) {
1900 printk(KERN_ERR
1901 "generic_make_request: Trying to access "
1902 "nonexistent block-device %s (%Lu)\n",
1903 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001904 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001905 goto end_io;
1906 }
1907
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001908 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001909 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001910 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001911 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001912 goto end_io;
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001915 * If this device has partitions, remap block n
1916 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001918 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001920 if (bio_check_eod(bio, nr_sectors))
1921 goto end_io;
1922
1923 /*
1924 * Filter flush bio's early so that make_request based
1925 * drivers without flush support don't have to worry
1926 * about them.
1927 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001928 if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
Jens Axboec888a8f2016-04-13 13:33:19 -06001929 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
Jens Axboe1eff9d32016-08-05 15:35:16 -06001930 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001931 if (!nr_sectors) {
1932 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001933 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
Christoph Hellwig288dab82016-06-09 16:00:36 +02001937 switch (bio_op(bio)) {
1938 case REQ_OP_DISCARD:
1939 if (!blk_queue_discard(q))
1940 goto not_supported;
1941 break;
1942 case REQ_OP_SECURE_ERASE:
1943 if (!blk_queue_secure_erase(q))
1944 goto not_supported;
1945 break;
1946 case REQ_OP_WRITE_SAME:
1947 if (!bdev_write_same(bio->bi_bdev))
1948 goto not_supported;
1949 break;
1950 default:
1951 break;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001954 /*
1955 * Various block parts want %current->io_context and lazy ioc
1956 * allocation ends up trading a lot of pain for a small amount of
1957 * memory. Just allocate it upfront. This may fail and block
1958 * layer knows how to live with it.
1959 */
1960 create_io_context(GFP_ATOMIC, q->node);
1961
Tejun Heoae118892015-08-18 14:55:20 -07001962 if (!blkcg_bio_issue_check(q, bio))
1963 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001964
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001965 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001966 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001967
Christoph Hellwig288dab82016-06-09 16:00:36 +02001968not_supported:
1969 err = -EOPNOTSUPP;
Tejun Heoa7384672008-11-28 13:32:03 +09001970end_io:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001971 bio->bi_error = err;
1972 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001973 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974}
1975
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001976/**
1977 * generic_make_request - hand a buffer to its device driver for I/O
1978 * @bio: The bio describing the location in memory and on the device.
1979 *
1980 * generic_make_request() is used to make I/O requests of block
1981 * devices. It is passed a &struct bio, which describes the I/O that needs
1982 * to be done.
1983 *
1984 * generic_make_request() does not return any status. The
1985 * success/failure status of the request, along with notification of
1986 * completion, is delivered asynchronously through the bio->bi_end_io
1987 * function described (one day) else where.
1988 *
1989 * The caller of generic_make_request must make sure that bi_io_vec
1990 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1991 * set to describe the device address, and the
1992 * bi_end_io and optionally bi_private are set to describe how
1993 * completion notification should be signaled.
1994 *
1995 * generic_make_request and the drivers it calls may use bi_next if this
1996 * bio happens to be merged with someone else, and may resubmit the bio to
1997 * a lower device by calling into generic_make_request recursively, which
1998 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001999 */
Jens Axboedece1632015-11-05 10:41:16 -07002000blk_qc_t generic_make_request(struct bio *bio)
Neil Brownd89d8792007-05-01 09:53:42 +02002001{
NeilBrown5959cde2017-03-10 17:00:47 +11002002 /*
2003 * bio_list_on_stack[0] contains bios submitted by the current
2004 * make_request_fn.
2005 * bio_list_on_stack[1] contains bios that were submitted before
2006 * the current make_request_fn, but that haven't been processed
2007 * yet.
2008 */
2009 struct bio_list bio_list_on_stack[2];
Jens Axboedece1632015-11-05 10:41:16 -07002010 blk_qc_t ret = BLK_QC_T_NONE;
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002011
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002012 if (!generic_make_request_checks(bio))
Jens Axboedece1632015-11-05 10:41:16 -07002013 goto out;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002014
2015 /*
2016 * We only want one ->make_request_fn to be active at a time, else
2017 * stack usage with stacked devices could be a problem. So use
2018 * current->bio_list to keep a list of requests submited by a
2019 * make_request_fn function. current->bio_list is also used as a
2020 * flag to say if generic_make_request is currently active in this
2021 * task or not. If it is NULL, then no make_request is active. If
2022 * it is non-NULL, then a make_request is active, and new requests
2023 * should be added at the tail
2024 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002025 if (current->bio_list) {
NeilBrown5959cde2017-03-10 17:00:47 +11002026 bio_list_add(&current->bio_list[0], bio);
Jens Axboedece1632015-11-05 10:41:16 -07002027 goto out;
Neil Brownd89d8792007-05-01 09:53:42 +02002028 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002029
Neil Brownd89d8792007-05-01 09:53:42 +02002030 /* following loop may be a bit non-obvious, and so deserves some
2031 * explanation.
2032 * Before entering the loop, bio->bi_next is NULL (as all callers
2033 * ensure that) so we have a list with a single bio.
2034 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002035 * we assign bio_list to a pointer to the bio_list_on_stack,
2036 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002037 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02002038 * through a recursive call to generic_make_request. If it
2039 * did, we find a non-NULL value in bio_list and re-enter the loop
2040 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002041 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002042 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02002043 */
2044 BUG_ON(bio->bi_next);
NeilBrown5959cde2017-03-10 17:00:47 +11002045 bio_list_init(&bio_list_on_stack[0]);
2046 current->bio_list = bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02002047 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002048 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2049
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01002050 if (likely(blk_queue_enter(q, false) == 0)) {
NeilBrownd5986e02017-03-08 07:38:05 +11002051 struct bio_list lower, same;
2052
2053 /* Create a fresh bio_list for all subordinate requests */
NeilBrown5959cde2017-03-10 17:00:47 +11002054 bio_list_on_stack[1] = bio_list_on_stack[0];
2055 bio_list_init(&bio_list_on_stack[0]);
Jens Axboedece1632015-11-05 10:41:16 -07002056 ret = q->make_request_fn(q, bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002057
2058 blk_queue_exit(q);
2059
NeilBrownd5986e02017-03-08 07:38:05 +11002060 /* sort new bios into those for a lower level
2061 * and those for the same level
2062 */
2063 bio_list_init(&lower);
2064 bio_list_init(&same);
NeilBrown5959cde2017-03-10 17:00:47 +11002065 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
NeilBrownd5986e02017-03-08 07:38:05 +11002066 if (q == bdev_get_queue(bio->bi_bdev))
2067 bio_list_add(&same, bio);
2068 else
2069 bio_list_add(&lower, bio);
2070 /* now assemble so we handle the lowest level first */
NeilBrown5959cde2017-03-10 17:00:47 +11002071 bio_list_merge(&bio_list_on_stack[0], &lower);
2072 bio_list_merge(&bio_list_on_stack[0], &same);
2073 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
Dan Williams3ef28e82015-10-21 13:20:12 -04002074 } else {
Dan Williams3ef28e82015-10-21 13:20:12 -04002075 bio_io_error(bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002076 }
NeilBrown5959cde2017-03-10 17:00:47 +11002077 bio = bio_list_pop(&bio_list_on_stack[0]);
Neil Brownd89d8792007-05-01 09:53:42 +02002078 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002079 current->bio_list = NULL; /* deactivate */
Jens Axboedece1632015-11-05 10:41:16 -07002080
2081out:
2082 return ret;
Neil Brownd89d8792007-05-01 09:53:42 +02002083}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084EXPORT_SYMBOL(generic_make_request);
2085
2086/**
Randy Dunlap710027a2008-08-19 20:13:11 +02002087 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 * @bio: The &struct bio which describes the I/O
2089 *
2090 * submit_bio() is very similar in purpose to generic_make_request(), and
2091 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02002092 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 *
2094 */
Mike Christie4e49ea42016-06-05 14:31:41 -05002095blk_qc_t submit_bio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096{
Jens Axboebf2de6f2007-09-27 13:01:25 +02002097 /*
2098 * If it's a regular read/write or a barrier with data attached,
2099 * go through the normal accounting stuff before submission.
2100 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002101 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002102 unsigned int count;
2103
Mike Christie95fe6c12016-06-05 14:31:48 -05002104 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002105 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2106 else
2107 count = bio_sectors(bio);
2108
Mike Christiea8ebb052016-06-05 14:31:45 -05002109 if (op_is_write(bio_op(bio))) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02002110 count_vm_events(PGPGOUT, count);
2111 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07002112 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002113 count_vm_events(PGPGIN, count);
2114 }
2115
2116 if (unlikely(block_dump)) {
2117 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02002118 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07002119 current->comm, task_pid_nr(current),
Mike Christiea8ebb052016-06-05 14:31:45 -05002120 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002121 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02002122 bdevname(bio->bi_bdev, b),
2123 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 }
2126
Jens Axboedece1632015-11-05 10:41:16 -07002127 return generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129EXPORT_SYMBOL(submit_bio);
2130
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002131/**
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002132 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2133 * for new the queue limits
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002134 * @q: the queue
2135 * @rq: the request being checked
2136 *
2137 * Description:
2138 * @rq may have been made based on weaker limitations of upper-level queues
2139 * in request stacking drivers, and it may violate the limitation of @q.
2140 * Since the block layer and the underlying device driver trust @rq
2141 * after it is inserted to @q, it should be checked against @q before
2142 * the insertion using this generic function.
2143 *
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002144 * Request stacking drivers like request-based dm may change the queue
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002145 * limits when retrying requests on other queues. Those requests need
2146 * to be checked against the new queue limits again during dispatch.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002147 */
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002148static int blk_cloned_rq_check_limits(struct request_queue *q,
2149 struct request *rq)
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002150{
Mike Christie8fe0d472016-06-05 14:32:15 -05002151 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002152 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2153 return -EIO;
2154 }
2155
2156 /*
2157 * queue's settings related to segment counting like q->bounce_pfn
2158 * may differ from that of other stacking queues.
2159 * Recalculate it to check the request correctly on this queue's
2160 * limitation.
2161 */
2162 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002163 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002164 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2165 return -EIO;
2166 }
2167
2168 return 0;
2169}
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002170
2171/**
2172 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2173 * @q: the queue to submit the request
2174 * @rq: the request being queued
2175 */
2176int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2177{
2178 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002179 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002180
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002181 if (blk_cloned_rq_check_limits(q, rq))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002182 return -EIO;
2183
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002184 if (rq->rq_disk &&
2185 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002186 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002187
Keith Busch7fb48982014-10-17 17:46:38 -06002188 if (q->mq_ops) {
2189 if (blk_queue_io_stat(q))
2190 blk_account_io_start(rq, true);
Mike Snitzer6acfe682016-02-05 08:49:01 -05002191 blk_mq_insert_request(rq, false, true, false);
Keith Busch7fb48982014-10-17 17:46:38 -06002192 return 0;
2193 }
2194
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002195 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002196 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002197 spin_unlock_irqrestore(q->queue_lock, flags);
2198 return -ENODEV;
2199 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002200
2201 /*
2202 * Submitting request must be dequeued before calling this function
2203 * because it will be linked to another request_queue
2204 */
2205 BUG_ON(blk_queued_rq(rq));
2206
Mike Christie28a8f0d2016-06-05 14:32:25 -05002207 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
Jeff Moyer4853aba2011-08-15 21:37:25 +02002208 where = ELEVATOR_INSERT_FLUSH;
2209
2210 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002211 if (where == ELEVATOR_INSERT_FLUSH)
2212 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002213 spin_unlock_irqrestore(q->queue_lock, flags);
2214
2215 return 0;
2216}
2217EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2218
Tejun Heo80a761f2009-07-03 17:48:17 +09002219/**
2220 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2221 * @rq: request to examine
2222 *
2223 * Description:
2224 * A request could be merge of IOs which require different failure
2225 * handling. This function determines the number of bytes which
2226 * can be failed from the beginning of the request without
2227 * crossing into area which need to be retried further.
2228 *
2229 * Return:
2230 * The number of bytes to fail.
2231 *
2232 * Context:
2233 * queue_lock must be held.
2234 */
2235unsigned int blk_rq_err_bytes(const struct request *rq)
2236{
2237 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2238 unsigned int bytes = 0;
2239 struct bio *bio;
2240
2241 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2242 return blk_rq_bytes(rq);
2243
2244 /*
2245 * Currently the only 'mixing' which can happen is between
2246 * different fastfail types. We can safely fail portions
2247 * which have all the failfast bits that the first one has -
2248 * the ones which are at least as eager to fail as the first
2249 * one.
2250 */
2251 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -06002252 if ((bio->bi_opf & ff) != ff)
Tejun Heo80a761f2009-07-03 17:48:17 +09002253 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002254 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002255 }
2256
2257 /* this could lead to infinite loop */
2258 BUG_ON(blk_rq_bytes(rq) && !bytes);
2259 return bytes;
2260}
2261EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2262
Jens Axboe320ae512013-10-24 09:20:05 +01002263void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002264{
Jens Axboec2553b52009-04-24 08:10:11 +02002265 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002266 const int rw = rq_data_dir(req);
2267 struct hd_struct *part;
2268 int cpu;
2269
2270 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002271 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002272 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2273 part_stat_unlock();
2274 }
2275}
2276
Jens Axboe320ae512013-10-24 09:20:05 +01002277void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002278{
Jens Axboebc58ba92009-01-23 10:54:44 +01002279 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002280 * Account IO completion. flush_rq isn't accounted as a
2281 * normal IO on queueing nor completion. Accounting the
2282 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002283 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002284 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002285 unsigned long duration = jiffies - req->start_time;
2286 const int rw = rq_data_dir(req);
2287 struct hd_struct *part;
2288 int cpu;
2289
2290 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002291 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002292
2293 part_stat_inc(cpu, part, ios[rw]);
2294 part_stat_add(cpu, part, ticks[rw], duration);
2295 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002296 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002297
Jens Axboe6c23a962011-01-07 08:43:37 +01002298 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002299 part_stat_unlock();
2300 }
2301}
2302
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002303#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002304/*
2305 * Don't process normal requests when queue is suspended
2306 * or in the process of suspending/resuming
2307 */
2308static struct request *blk_pm_peek_request(struct request_queue *q,
2309 struct request *rq)
2310{
2311 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2312 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2313 return NULL;
2314 else
2315 return rq;
2316}
2317#else
2318static inline struct request *blk_pm_peek_request(struct request_queue *q,
2319 struct request *rq)
2320{
2321 return rq;
2322}
2323#endif
2324
Jens Axboe320ae512013-10-24 09:20:05 +01002325void blk_account_io_start(struct request *rq, bool new_io)
2326{
2327 struct hd_struct *part;
2328 int rw = rq_data_dir(rq);
2329 int cpu;
2330
2331 if (!blk_do_io_stat(rq))
2332 return;
2333
2334 cpu = part_stat_lock();
2335
2336 if (!new_io) {
2337 part = rq->part;
2338 part_stat_inc(cpu, part, merges[rw]);
2339 } else {
2340 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2341 if (!hd_struct_try_get(part)) {
2342 /*
2343 * The partition is already being removed,
2344 * the request will be accounted on the disk only
2345 *
2346 * We take a reference on disk->part0 although that
2347 * partition will never be deleted, so we can treat
2348 * it as any other partition.
2349 */
2350 part = &rq->rq_disk->part0;
2351 hd_struct_get(part);
2352 }
2353 part_round_stats(cpu, part);
2354 part_inc_in_flight(part, rw);
2355 rq->part = part;
2356 }
2357
2358 part_stat_unlock();
2359}
2360
Tejun Heo53a08802008-12-03 12:41:26 +01002361/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002362 * blk_peek_request - peek at the top of a request queue
2363 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002364 *
2365 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002366 * Return the request at the top of @q. The returned request
2367 * should be started using blk_start_request() before LLD starts
2368 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002369 *
2370 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002371 * Pointer to the request at the top of @q if available. Null
2372 * otherwise.
2373 *
2374 * Context:
2375 * queue_lock must be held.
2376 */
2377struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002378{
2379 struct request *rq;
2380 int ret;
2381
2382 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002383
2384 rq = blk_pm_peek_request(q, rq);
2385 if (!rq)
2386 break;
2387
Tejun Heo158dbda2009-04-23 11:05:18 +09002388 if (!(rq->cmd_flags & REQ_STARTED)) {
2389 /*
2390 * This is the first time the device driver
2391 * sees this request (possibly after
2392 * requeueing). Notify IO scheduler.
2393 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002394 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002395 elv_activate_rq(q, rq);
2396
2397 /*
2398 * just mark as started even if we don't start
2399 * it, a request that has been delayed should
2400 * not be passed by new incoming requests
2401 */
2402 rq->cmd_flags |= REQ_STARTED;
2403 trace_block_rq_issue(q, rq);
2404 }
2405
2406 if (!q->boundary_rq || q->boundary_rq == rq) {
2407 q->end_sector = rq_end_sector(rq);
2408 q->boundary_rq = NULL;
2409 }
2410
2411 if (rq->cmd_flags & REQ_DONTPREP)
2412 break;
2413
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002414 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002415 /*
2416 * make sure space for the drain appears we
2417 * know we can do this because max_hw_segments
2418 * has been adjusted to be one fewer than the
2419 * device can handle
2420 */
2421 rq->nr_phys_segments++;
2422 }
2423
2424 if (!q->prep_rq_fn)
2425 break;
2426
2427 ret = q->prep_rq_fn(q, rq);
2428 if (ret == BLKPREP_OK) {
2429 break;
2430 } else if (ret == BLKPREP_DEFER) {
2431 /*
2432 * the request may have been (partially) prepped.
2433 * we need to keep this request in the front to
2434 * avoid resource deadlock. REQ_STARTED will
2435 * prevent other fs requests from passing this one.
2436 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002437 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002438 !(rq->cmd_flags & REQ_DONTPREP)) {
2439 /*
2440 * remove the space for the drain we added
2441 * so that we don't add it again
2442 */
2443 --rq->nr_phys_segments;
2444 }
2445
2446 rq = NULL;
2447 break;
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002448 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2449 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2450
Tejun Heo158dbda2009-04-23 11:05:18 +09002451 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002452 /*
2453 * Mark this request as started so we don't trigger
2454 * any debug logic in the end I/O path.
2455 */
2456 blk_start_request(rq);
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002457 __blk_end_request_all(rq, err);
Tejun Heo158dbda2009-04-23 11:05:18 +09002458 } else {
2459 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2460 break;
2461 }
2462 }
2463
2464 return rq;
2465}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002466EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002467
Tejun Heo9934c8c2009-05-08 11:54:16 +09002468void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002469{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002470 struct request_queue *q = rq->q;
2471
Tejun Heo158dbda2009-04-23 11:05:18 +09002472 BUG_ON(list_empty(&rq->queuelist));
2473 BUG_ON(ELV_ON_HASH(rq));
2474
2475 list_del_init(&rq->queuelist);
2476
2477 /*
2478 * the time frame between a request being removed from the lists
2479 * and to it is freed is accounted as io that is in progress at
2480 * the driver side.
2481 */
Divyesh Shah91952912010-04-01 15:01:41 -07002482 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002483 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002484 set_io_start_time_ns(rq);
2485 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002486}
2487
Tejun Heo5efccd12009-04-23 11:05:18 +09002488/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002489 * blk_start_request - start request processing on the driver
2490 * @req: request to dequeue
2491 *
2492 * Description:
2493 * Dequeue @req and start timeout timer on it. This hands off the
2494 * request to the driver.
2495 *
2496 * Block internal functions which don't want to start timer should
2497 * call blk_dequeue_request().
2498 *
2499 * Context:
2500 * queue_lock must be held.
2501 */
2502void blk_start_request(struct request *req)
2503{
2504 blk_dequeue_request(req);
2505
2506 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002507 * We are now handing the request to the hardware, initialize
2508 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002509 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002510 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002511 if (unlikely(blk_bidi_rq(req)))
2512 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2513
Jeff Moyer4912aa62013-10-08 14:36:41 -04002514 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002515 blk_add_timer(req);
2516}
2517EXPORT_SYMBOL(blk_start_request);
2518
2519/**
2520 * blk_fetch_request - fetch a request from a request queue
2521 * @q: request queue to fetch a request from
2522 *
2523 * Description:
2524 * Return the request at the top of @q. The request is started on
2525 * return and LLD can start processing it immediately.
2526 *
2527 * Return:
2528 * Pointer to the request at the top of @q if available. Null
2529 * otherwise.
2530 *
2531 * Context:
2532 * queue_lock must be held.
2533 */
2534struct request *blk_fetch_request(struct request_queue *q)
2535{
2536 struct request *rq;
2537
2538 rq = blk_peek_request(q);
2539 if (rq)
2540 blk_start_request(rq);
2541 return rq;
2542}
2543EXPORT_SYMBOL(blk_fetch_request);
2544
2545/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002546 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002547 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002548 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002549 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002550 *
2551 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002552 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2553 * the request structure even if @req doesn't have leftover.
2554 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002555 *
2556 * This special helper function is only for request stacking drivers
2557 * (e.g. request-based dm) so that they can handle partial completion.
2558 * Actual device drivers should use blk_end_request instead.
2559 *
2560 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2561 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002562 *
2563 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002564 * %false - this request doesn't have any more data
2565 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002566 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002567bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002569 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002571 trace_block_rq_complete(req->q, req, nr_bytes);
2572
Tejun Heo2e60e022009-04-23 11:05:18 +09002573 if (!req->bio)
2574 return false;
2575
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002577 * For fs requests, rq is just carrier of independent bio's
2578 * and each partial completion should be handled separately.
2579 * Reset per-request error on each partial completion.
2580 *
2581 * TODO: tj: This is too subtle. It would be better to let
2582 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002584 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 req->errors = 0;
2586
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002587 if (error && req->cmd_type == REQ_TYPE_FS &&
2588 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002589 char *error_type;
2590
2591 switch (error) {
2592 case -ENOLINK:
2593 error_type = "recoverable transport";
2594 break;
2595 case -EREMOTEIO:
2596 error_type = "critical target";
2597 break;
2598 case -EBADE:
2599 error_type = "critical nexus";
2600 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002601 case -ETIMEDOUT:
2602 error_type = "timeout";
2603 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002604 case -ENOSPC:
2605 error_type = "critical space allocation";
2606 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002607 case -ENODATA:
2608 error_type = "critical medium";
2609 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002610 case -EIO:
2611 default:
2612 error_type = "I/O";
2613 break;
2614 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002615 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2616 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002617 req->rq_disk->disk_name : "?",
2618 (unsigned long long)blk_rq_pos(req));
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 }
2621
Jens Axboebc58ba92009-01-23 10:54:44 +01002622 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002623
Kent Overstreetf79ea412012-09-20 16:38:30 -07002624 total_bytes = 0;
AnilKumar Chimata7214d7e2017-06-23 03:09:59 -07002625
2626 /*
2627 * Check for this if flagged, Req based dm needs to perform
2628 * post processing, hence dont end bios or request.DM
2629 * layer takes care.
2630 */
2631 if (bio_flagged(req->bio, BIO_DONTFREE))
2632 return false;
2633
Kent Overstreetf79ea412012-09-20 16:38:30 -07002634 while (req->bio) {
2635 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002636 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
Kent Overstreet4f024f32013-10-11 15:44:27 -07002638 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
Kent Overstreetf79ea412012-09-20 16:38:30 -07002641 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Kent Overstreetf79ea412012-09-20 16:38:30 -07002643 total_bytes += bio_bytes;
2644 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
Kent Overstreetf79ea412012-09-20 16:38:30 -07002646 if (!nr_bytes)
2647 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 }
2649
2650 /*
2651 * completely done
2652 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002653 if (!req->bio) {
2654 /*
2655 * Reset counters so that the request stacking driver
2656 * can find how many bytes remain in the request
2657 * later.
2658 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002659 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002660 return false;
2661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002663 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002664
2665 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002666 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002667 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002668
Tejun Heo80a761f2009-07-03 17:48:17 +09002669 /* mixed attributes always follow the first bio */
2670 if (req->cmd_flags & REQ_MIXED_MERGE) {
2671 req->cmd_flags &= ~REQ_FAILFAST_MASK;
Jens Axboe1eff9d32016-08-05 15:35:16 -06002672 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
Tejun Heo80a761f2009-07-03 17:48:17 +09002673 }
2674
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002675 /*
2676 * If total number of sectors is less than the first segment
2677 * size, something has gone terribly wrong.
2678 */
2679 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002680 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002681 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002682 }
2683
2684 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002686
Tejun Heo2e60e022009-04-23 11:05:18 +09002687 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688}
Tejun Heo2e60e022009-04-23 11:05:18 +09002689EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
Tejun Heo2e60e022009-04-23 11:05:18 +09002691static bool blk_update_bidi_request(struct request *rq, int error,
2692 unsigned int nr_bytes,
2693 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002694{
Tejun Heo2e60e022009-04-23 11:05:18 +09002695 if (blk_update_request(rq, error, nr_bytes))
2696 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002697
Tejun Heo2e60e022009-04-23 11:05:18 +09002698 /* Bidi request must be completed as a whole */
2699 if (unlikely(blk_bidi_rq(rq)) &&
2700 blk_update_request(rq->next_rq, error, bidi_bytes))
2701 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002702
Jens Axboee2e1a142010-06-09 10:42:09 +02002703 if (blk_queue_add_random(rq->q))
2704 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002705
2706 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707}
2708
James Bottomley28018c22010-07-01 19:49:17 +09002709/**
2710 * blk_unprep_request - unprepare a request
2711 * @req: the request
2712 *
2713 * This function makes a request ready for complete resubmission (or
2714 * completion). It happens only after all error handling is complete,
2715 * so represents the appropriate moment to deallocate any resources
2716 * that were allocated to the request in the prep_rq_fn. The queue
2717 * lock is held when calling this.
2718 */
2719void blk_unprep_request(struct request *req)
2720{
2721 struct request_queue *q = req->q;
2722
2723 req->cmd_flags &= ~REQ_DONTPREP;
2724 if (q->unprep_rq_fn)
2725 q->unprep_rq_fn(q, req);
2726}
2727EXPORT_SYMBOL_GPL(blk_unprep_request);
2728
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729/*
2730 * queue lock must be held
2731 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002732void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002734 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002735 blk_queue_end_tag(req->q, req);
2736
James Bottomleyba396a62009-05-27 14:17:08 +02002737 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002739 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002740 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741
Mike Andersone78042e2008-10-30 02:16:20 -07002742 blk_delete_timer(req);
2743
James Bottomley28018c22010-07-01 19:49:17 +09002744 if (req->cmd_flags & REQ_DONTPREP)
2745 blk_unprep_request(req);
2746
Jens Axboebc58ba92009-01-23 10:54:44 +01002747 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002750 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002751 else {
2752 if (blk_bidi_rq(req))
2753 __blk_put_request(req->next_rq->q, req->next_rq);
2754
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 }
2757}
Christoph Hellwig12120072014-04-16 09:44:59 +02002758EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002760/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002761 * blk_end_bidi_request - Complete a bidi request
2762 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002763 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002764 * @nr_bytes: number of bytes to complete @rq
2765 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002766 *
2767 * Description:
2768 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002769 * Drivers that supports bidi can safely call this member for any
2770 * type of request, bidi or uni. In the later case @bidi_bytes is
2771 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002772 *
2773 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002774 * %false - we are done with this request
2775 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002776 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002777static bool blk_end_bidi_request(struct request *rq, int error,
2778 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002779{
2780 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002781 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002782
Tejun Heo2e60e022009-04-23 11:05:18 +09002783 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2784 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002785
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002786 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002787 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002788 spin_unlock_irqrestore(q->queue_lock, flags);
2789
Tejun Heo2e60e022009-04-23 11:05:18 +09002790 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002791}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002792
2793/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002794 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2795 * @rq: the request to complete
2796 * @error: %0 for success, < %0 for error
2797 * @nr_bytes: number of bytes to complete @rq
2798 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002799 *
2800 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002801 * Identical to blk_end_bidi_request() except that queue lock is
2802 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002803 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002804 * Return:
2805 * %false - we are done with this request
2806 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002807 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002808bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002809 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002810{
Tejun Heo2e60e022009-04-23 11:05:18 +09002811 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2812 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002813
Tejun Heo2e60e022009-04-23 11:05:18 +09002814 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002815
Tejun Heo2e60e022009-04-23 11:05:18 +09002816 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002817}
2818
2819/**
2820 * blk_end_request - Helper function for drivers to complete the request.
2821 * @rq: the request being processed
2822 * @error: %0 for success, < %0 for error
2823 * @nr_bytes: number of bytes to complete
2824 *
2825 * Description:
2826 * Ends I/O on a number of bytes attached to @rq.
2827 * If @rq has leftover, sets it up for the next range of segments.
2828 *
2829 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002830 * %false - we are done with this request
2831 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002832 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002833bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002834{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002835 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002836}
Jens Axboe56ad1742009-07-28 22:11:24 +02002837EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002838
2839/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002840 * blk_end_request_all - Helper function for drives to finish the request.
2841 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002842 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002843 *
2844 * Description:
2845 * Completely finish @rq.
2846 */
2847void blk_end_request_all(struct request *rq, int error)
2848{
2849 bool pending;
2850 unsigned int bidi_bytes = 0;
2851
2852 if (unlikely(blk_bidi_rq(rq)))
2853 bidi_bytes = blk_rq_bytes(rq->next_rq);
2854
2855 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2856 BUG_ON(pending);
2857}
Jens Axboe56ad1742009-07-28 22:11:24 +02002858EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002859
2860/**
2861 * blk_end_request_cur - Helper function to finish the current request chunk.
2862 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002863 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002864 *
2865 * Description:
2866 * Complete the current consecutively mapped chunk from @rq.
2867 *
2868 * Return:
2869 * %false - we are done with this request
2870 * %true - still buffers pending for this request
2871 */
2872bool blk_end_request_cur(struct request *rq, int error)
2873{
2874 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2875}
Jens Axboe56ad1742009-07-28 22:11:24 +02002876EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002877
2878/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002879 * blk_end_request_err - Finish a request till the next failure boundary.
2880 * @rq: the request to finish till the next failure boundary for
2881 * @error: must be negative errno
2882 *
2883 * Description:
2884 * Complete @rq till the next failure boundary.
2885 *
2886 * Return:
2887 * %false - we are done with this request
2888 * %true - still buffers pending for this request
2889 */
2890bool blk_end_request_err(struct request *rq, int error)
2891{
2892 WARN_ON(error >= 0);
2893 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2894}
2895EXPORT_SYMBOL_GPL(blk_end_request_err);
2896
2897/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002898 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002899 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002900 * @error: %0 for success, < %0 for error
2901 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002902 *
2903 * Description:
2904 * Must be called with queue lock held unlike blk_end_request().
2905 *
2906 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002907 * %false - we are done with this request
2908 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002909 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002910bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002911{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002912 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002913}
Jens Axboe56ad1742009-07-28 22:11:24 +02002914EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002915
2916/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002917 * __blk_end_request_all - Helper function for drives to finish the request.
2918 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002919 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002920 *
2921 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002922 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002923 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002924void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002925{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002926 bool pending;
2927 unsigned int bidi_bytes = 0;
2928
2929 if (unlikely(blk_bidi_rq(rq)))
2930 bidi_bytes = blk_rq_bytes(rq->next_rq);
2931
2932 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2933 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002934}
Jens Axboe56ad1742009-07-28 22:11:24 +02002935EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002936
2937/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002938 * __blk_end_request_cur - Helper function to finish the current request chunk.
2939 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002940 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002941 *
2942 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002943 * Complete the current consecutively mapped chunk from @rq. Must
2944 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002945 *
2946 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002947 * %false - we are done with this request
2948 * %true - still buffers pending for this request
2949 */
2950bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002951{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002952 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002953}
Jens Axboe56ad1742009-07-28 22:11:24 +02002954EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002955
Tejun Heo80a761f2009-07-03 17:48:17 +09002956/**
2957 * __blk_end_request_err - Finish a request till the next failure boundary.
2958 * @rq: the request to finish till the next failure boundary for
2959 * @error: must be negative errno
2960 *
2961 * Description:
2962 * Complete @rq till the next failure boundary. Must be called
2963 * with queue lock held.
2964 *
2965 * Return:
2966 * %false - we are done with this request
2967 * %true - still buffers pending for this request
2968 */
2969bool __blk_end_request_err(struct request *rq, int error)
2970{
2971 WARN_ON(error >= 0);
2972 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2973}
2974EXPORT_SYMBOL_GPL(__blk_end_request_err);
2975
Jens Axboe86db1e22008-01-29 14:53:40 +01002976void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2977 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978{
Mike Christie4993b772016-06-05 14:32:10 -05002979 req_set_op(rq, bio_op(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980
Jens Axboeb4f42e22014-04-10 09:46:28 -06002981 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002982 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002983
Kent Overstreet4f024f32013-10-11 15:44:27 -07002984 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986
NeilBrown66846572007-08-16 13:31:28 +02002987 if (bio->bi_bdev)
2988 rq->rq_disk = bio->bi_bdev->bd_disk;
2989}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002991#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2992/**
2993 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2994 * @rq: the request to be flushed
2995 *
2996 * Description:
2997 * Flush all pages in @rq.
2998 */
2999void rq_flush_dcache_pages(struct request *rq)
3000{
3001 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08003002 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01003003
3004 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08003005 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01003006}
3007EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3008#endif
3009
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02003010/**
3011 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3012 * @q : the queue of the device being checked
3013 *
3014 * Description:
3015 * Check if underlying low-level drivers of a device are busy.
3016 * If the drivers want to export their busy state, they must set own
3017 * exporting function using blk_queue_lld_busy() first.
3018 *
3019 * Basically, this function is used only by request stacking drivers
3020 * to stop dispatching requests to underlying devices when underlying
3021 * devices are busy. This behavior helps more I/O merging on the queue
3022 * of the request stacking driver and prevents I/O throughput regression
3023 * on burst I/O load.
3024 *
3025 * Return:
3026 * 0 - Not busy (The request stacking driver should dispatch request)
3027 * 1 - Busy (The request stacking driver should stop dispatching request)
3028 */
3029int blk_lld_busy(struct request_queue *q)
3030{
3031 if (q->lld_busy_fn)
3032 return q->lld_busy_fn(q);
3033
3034 return 0;
3035}
3036EXPORT_SYMBOL_GPL(blk_lld_busy);
3037
Mike Snitzer78d8e582015-06-26 10:01:13 -04003038/**
3039 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3040 * @rq: the clone request to be cleaned up
3041 *
3042 * Description:
3043 * Free all bios in @rq for a cloned request.
3044 */
3045void blk_rq_unprep_clone(struct request *rq)
3046{
3047 struct bio *bio;
3048
3049 while ((bio = rq->bio) != NULL) {
3050 rq->bio = bio->bi_next;
3051
3052 bio_put(bio);
3053 }
3054}
3055EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3056
3057/*
3058 * Copy attributes of the original request to the clone request.
3059 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3060 */
3061static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003062{
3063 dst->cpu = src->cpu;
Mike Christie4993b772016-06-05 14:32:10 -05003064 req_set_op_attrs(dst, req_op(src),
3065 (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003066 dst->cmd_type = src->cmd_type;
3067 dst->__sector = blk_rq_pos(src);
3068 dst->__data_len = blk_rq_bytes(src);
3069 dst->nr_phys_segments = src->nr_phys_segments;
3070 dst->ioprio = src->ioprio;
3071 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003072}
3073
3074/**
3075 * blk_rq_prep_clone - Helper function to setup clone request
3076 * @rq: the request to be setup
3077 * @rq_src: original request to be cloned
3078 * @bs: bio_set that bios for clone are allocated from
3079 * @gfp_mask: memory allocation mask for bio
3080 * @bio_ctr: setup function to be called for each clone bio.
3081 * Returns %0 for success, non %0 for failure.
3082 * @data: private data to be passed to @bio_ctr
3083 *
3084 * Description:
3085 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3086 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3087 * are not copied, and copying such parts is the caller's responsibility.
3088 * Also, pages which the original bios are pointing to are not copied
3089 * and the cloned bios just point same pages.
3090 * So cloned bios must be completed before original bios, which means
3091 * the caller must complete @rq before @rq_src.
3092 */
3093int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3094 struct bio_set *bs, gfp_t gfp_mask,
3095 int (*bio_ctr)(struct bio *, struct bio *, void *),
3096 void *data)
3097{
3098 struct bio *bio, *bio_src;
3099
3100 if (!bs)
3101 bs = fs_bio_set;
3102
3103 __rq_for_each_bio(bio_src, rq_src) {
3104 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3105 if (!bio)
3106 goto free_and_out;
3107
3108 if (bio_ctr && bio_ctr(bio, bio_src, data))
3109 goto free_and_out;
3110
3111 if (rq->bio) {
3112 rq->biotail->bi_next = bio;
3113 rq->biotail = bio;
3114 } else
3115 rq->bio = rq->biotail = bio;
3116 }
3117
3118 __blk_rq_prep_clone(rq, rq_src);
3119
3120 return 0;
3121
3122free_and_out:
3123 if (bio)
3124 bio_put(bio);
3125 blk_rq_unprep_clone(rq);
3126
3127 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003128}
3129EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3130
Jens Axboe59c3d452014-04-08 09:15:35 -06003131int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132{
3133 return queue_work(kblockd_workqueue, work);
3134}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135EXPORT_SYMBOL(kblockd_schedule_work);
3136
Jens Axboeee63cfa2016-08-24 15:52:48 -06003137int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3138{
3139 return queue_work_on(cpu, kblockd_workqueue, work);
3140}
3141EXPORT_SYMBOL(kblockd_schedule_work_on);
3142
Jens Axboe59c3d452014-04-08 09:15:35 -06003143int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3144 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003145{
3146 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3147}
3148EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3149
Jens Axboe8ab14592014-04-08 09:17:40 -06003150int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3151 unsigned long delay)
3152{
3153 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3154}
3155EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3156
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003157/**
3158 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3159 * @plug: The &struct blk_plug that needs to be initialized
3160 *
3161 * Description:
3162 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3163 * pending I/O should the task end up blocking between blk_start_plug() and
3164 * blk_finish_plug(). This is important from a performance perspective, but
3165 * also ensures that we don't deadlock. For instance, if the task is blocking
3166 * for a memory allocation, memory reclaim could end up wanting to free a
3167 * page belonging to that request that is currently residing in our private
3168 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3169 * this kind of deadlock.
3170 */
Jens Axboe73c10102011-03-08 13:19:51 +01003171void blk_start_plug(struct blk_plug *plug)
3172{
3173 struct task_struct *tsk = current;
3174
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003175 /*
3176 * If this is a nested plug, don't actually assign it.
3177 */
3178 if (tsk->plug)
3179 return;
3180
Jens Axboe73c10102011-03-08 13:19:51 +01003181 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003182 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003183 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003184 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003185 * Store ordering should not be needed here, since a potential
3186 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01003187 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003188 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01003189}
3190EXPORT_SYMBOL(blk_start_plug);
3191
3192static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3193{
3194 struct request *rqa = container_of(a, struct request, queuelist);
3195 struct request *rqb = container_of(b, struct request, queuelist);
3196
Jianpeng Ma975927b2012-10-25 21:58:17 +02003197 return !(rqa->q < rqb->q ||
3198 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003199}
3200
Jens Axboe49cac012011-04-16 13:51:05 +02003201/*
3202 * If 'from_schedule' is true, then postpone the dispatch of requests
3203 * until a safe kblockd context. We due this to avoid accidental big
3204 * additional stack usage in driver dispatch, in places where the originally
3205 * plugger did not intend it.
3206 */
Jens Axboef6603782011-04-15 15:49:07 +02003207static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003208 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003209 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003210{
Jens Axboe49cac012011-04-16 13:51:05 +02003211 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003212
Bart Van Assche70460572012-11-28 13:45:56 +01003213 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003214 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003215 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003216 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003217 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003218}
3219
NeilBrown74018dc2012-07-31 09:08:15 +02003220static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003221{
3222 LIST_HEAD(callbacks);
3223
Shaohua Li2a7d5552012-07-31 09:08:15 +02003224 while (!list_empty(&plug->cb_list)) {
3225 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003226
Shaohua Li2a7d5552012-07-31 09:08:15 +02003227 while (!list_empty(&callbacks)) {
3228 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003229 struct blk_plug_cb,
3230 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003231 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003232 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003233 }
NeilBrown048c9372011-04-18 09:52:22 +02003234 }
3235}
3236
NeilBrown9cbb1752012-07-31 09:08:14 +02003237struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3238 int size)
3239{
3240 struct blk_plug *plug = current->plug;
3241 struct blk_plug_cb *cb;
3242
3243 if (!plug)
3244 return NULL;
3245
3246 list_for_each_entry(cb, &plug->cb_list, list)
3247 if (cb->callback == unplug && cb->data == data)
3248 return cb;
3249
3250 /* Not currently on the callback list */
3251 BUG_ON(size < sizeof(*cb));
3252 cb = kzalloc(size, GFP_ATOMIC);
3253 if (cb) {
3254 cb->data = data;
3255 cb->callback = unplug;
3256 list_add(&cb->list, &plug->cb_list);
3257 }
3258 return cb;
3259}
3260EXPORT_SYMBOL(blk_check_plugged);
3261
Jens Axboe49cac012011-04-16 13:51:05 +02003262void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003263{
3264 struct request_queue *q;
3265 unsigned long flags;
3266 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003267 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003268 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003269
NeilBrown74018dc2012-07-31 09:08:15 +02003270 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003271
3272 if (!list_empty(&plug->mq_list))
3273 blk_mq_flush_plug_list(plug, from_schedule);
3274
Jens Axboe73c10102011-03-08 13:19:51 +01003275 if (list_empty(&plug->list))
3276 return;
3277
NeilBrown109b8122011-04-11 14:13:10 +02003278 list_splice_init(&plug->list, &list);
3279
Jianpeng Ma422765c2013-01-11 14:46:09 +01003280 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003281
3282 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003283 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003284
3285 /*
3286 * Save and disable interrupts here, to avoid doing it for every
3287 * queue lock we have to take.
3288 */
Jens Axboe73c10102011-03-08 13:19:51 +01003289 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003290 while (!list_empty(&list)) {
3291 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003292 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003293 BUG_ON(!rq->q);
3294 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003295 /*
3296 * This drops the queue lock
3297 */
3298 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003299 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003300 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003301 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003302 spin_lock(q->queue_lock);
3303 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003304
3305 /*
3306 * Short-circuit if @q is dead
3307 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003308 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003309 __blk_end_request_all(rq, -ENODEV);
3310 continue;
3311 }
3312
Jens Axboe73c10102011-03-08 13:19:51 +01003313 /*
3314 * rq is already accounted, so use raw insert
3315 */
Mike Christie28a8f0d2016-06-05 14:32:25 -05003316 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
Jens Axboe401a18e2011-03-25 16:57:52 +01003317 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3318 else
3319 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003320
3321 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003322 }
3323
Jens Axboe99e22592011-04-18 09:59:55 +02003324 /*
3325 * This drops the queue lock
3326 */
3327 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003328 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003329
Jens Axboe73c10102011-03-08 13:19:51 +01003330 local_irq_restore(flags);
3331}
Jens Axboe73c10102011-03-08 13:19:51 +01003332
3333void blk_finish_plug(struct blk_plug *plug)
3334{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003335 if (plug != current->plug)
3336 return;
Jens Axboef6603782011-04-15 15:49:07 +02003337 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003338
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003339 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003340}
3341EXPORT_SYMBOL(blk_finish_plug);
3342
Jens Axboe05229be2015-11-05 10:44:55 -07003343bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3344{
3345 struct blk_plug *plug;
3346 long state;
Stephen Bates6e219352016-09-13 12:23:15 -06003347 unsigned int queue_num;
3348 struct blk_mq_hw_ctx *hctx;
Jens Axboe05229be2015-11-05 10:44:55 -07003349
3350 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3351 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3352 return false;
3353
Stephen Bates6e219352016-09-13 12:23:15 -06003354 queue_num = blk_qc_t_to_queue_num(cookie);
3355 hctx = q->queue_hw_ctx[queue_num];
3356 hctx->poll_considered++;
3357
Jens Axboe05229be2015-11-05 10:44:55 -07003358 plug = current->plug;
3359 if (plug)
3360 blk_flush_plug_list(plug, false);
3361
3362 state = current->state;
3363 while (!need_resched()) {
Jens Axboe05229be2015-11-05 10:44:55 -07003364 int ret;
3365
3366 hctx->poll_invoked++;
3367
3368 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3369 if (ret > 0) {
3370 hctx->poll_success++;
3371 set_current_state(TASK_RUNNING);
3372 return true;
3373 }
3374
3375 if (signal_pending_state(state, current))
3376 set_current_state(TASK_RUNNING);
3377
3378 if (current->state == TASK_RUNNING)
3379 return true;
3380 if (ret < 0)
3381 break;
3382 cpu_relax();
3383 }
3384
3385 return false;
3386}
Sagi Grimberg9645c1a2016-06-21 18:04:19 +02003387EXPORT_SYMBOL_GPL(blk_poll);
Jens Axboe05229be2015-11-05 10:44:55 -07003388
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003389#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003390/**
3391 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3392 * @q: the queue of the device
3393 * @dev: the device the queue belongs to
3394 *
3395 * Description:
3396 * Initialize runtime-PM-related fields for @q and start auto suspend for
3397 * @dev. Drivers that want to take advantage of request-based runtime PM
3398 * should call this function after @dev has been initialized, and its
3399 * request queue @q has been allocated, and runtime PM for it can not happen
3400 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3401 * cases, driver should call this function before any I/O has taken place.
3402 *
3403 * This function takes care of setting up using auto suspend for the device,
3404 * the autosuspend delay is set to -1 to make runtime suspend impossible
3405 * until an updated value is either set by user or by driver. Drivers do
3406 * not need to touch other autosuspend settings.
3407 *
3408 * The block layer runtime PM is request based, so only works for drivers
3409 * that use request as their IO unit instead of those directly use bio's.
3410 */
3411void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3412{
3413 q->dev = dev;
3414 q->rpm_status = RPM_ACTIVE;
3415 pm_runtime_set_autosuspend_delay(q->dev, -1);
3416 pm_runtime_use_autosuspend(q->dev);
3417}
3418EXPORT_SYMBOL(blk_pm_runtime_init);
3419
3420/**
3421 * blk_pre_runtime_suspend - Pre runtime suspend check
3422 * @q: the queue of the device
3423 *
3424 * Description:
3425 * This function will check if runtime suspend is allowed for the device
3426 * by examining if there are any requests pending in the queue. If there
3427 * are requests pending, the device can not be runtime suspended; otherwise,
3428 * the queue's status will be updated to SUSPENDING and the driver can
3429 * proceed to suspend the device.
3430 *
3431 * For the not allowed case, we mark last busy for the device so that
3432 * runtime PM core will try to autosuspend it some time later.
3433 *
3434 * This function should be called near the start of the device's
3435 * runtime_suspend callback.
3436 *
3437 * Return:
3438 * 0 - OK to runtime suspend the device
3439 * -EBUSY - Device should not be runtime suspended
3440 */
3441int blk_pre_runtime_suspend(struct request_queue *q)
3442{
3443 int ret = 0;
3444
Ken Xue4fd41a852015-12-01 14:45:46 +08003445 if (!q->dev)
3446 return ret;
3447
Lin Ming6c954662013-03-23 11:42:26 +08003448 spin_lock_irq(q->queue_lock);
3449 if (q->nr_pending) {
3450 ret = -EBUSY;
3451 pm_runtime_mark_last_busy(q->dev);
3452 } else {
3453 q->rpm_status = RPM_SUSPENDING;
3454 }
3455 spin_unlock_irq(q->queue_lock);
3456 return ret;
3457}
3458EXPORT_SYMBOL(blk_pre_runtime_suspend);
3459
3460/**
3461 * blk_post_runtime_suspend - Post runtime suspend processing
3462 * @q: the queue of the device
3463 * @err: return value of the device's runtime_suspend function
3464 *
3465 * Description:
3466 * Update the queue's runtime status according to the return value of the
3467 * device's runtime suspend function and mark last busy for the device so
3468 * that PM core will try to auto suspend the device at a later time.
3469 *
3470 * This function should be called near the end of the device's
3471 * runtime_suspend callback.
3472 */
3473void blk_post_runtime_suspend(struct request_queue *q, int err)
3474{
Ken Xue4fd41a852015-12-01 14:45:46 +08003475 if (!q->dev)
3476 return;
3477
Lin Ming6c954662013-03-23 11:42:26 +08003478 spin_lock_irq(q->queue_lock);
3479 if (!err) {
3480 q->rpm_status = RPM_SUSPENDED;
3481 } else {
3482 q->rpm_status = RPM_ACTIVE;
3483 pm_runtime_mark_last_busy(q->dev);
3484 }
3485 spin_unlock_irq(q->queue_lock);
3486}
3487EXPORT_SYMBOL(blk_post_runtime_suspend);
3488
3489/**
3490 * blk_pre_runtime_resume - Pre runtime resume processing
3491 * @q: the queue of the device
3492 *
3493 * Description:
3494 * Update the queue's runtime status to RESUMING in preparation for the
3495 * runtime resume of the device.
3496 *
3497 * This function should be called near the start of the device's
3498 * runtime_resume callback.
3499 */
3500void blk_pre_runtime_resume(struct request_queue *q)
3501{
Ken Xue4fd41a852015-12-01 14:45:46 +08003502 if (!q->dev)
3503 return;
3504
Lin Ming6c954662013-03-23 11:42:26 +08003505 spin_lock_irq(q->queue_lock);
3506 q->rpm_status = RPM_RESUMING;
3507 spin_unlock_irq(q->queue_lock);
3508}
3509EXPORT_SYMBOL(blk_pre_runtime_resume);
3510
3511/**
3512 * blk_post_runtime_resume - Post runtime resume processing
3513 * @q: the queue of the device
3514 * @err: return value of the device's runtime_resume function
3515 *
3516 * Description:
3517 * Update the queue's runtime status according to the return value of the
3518 * device's runtime_resume function. If it is successfully resumed, process
3519 * the requests that are queued into the device's queue when it is resuming
3520 * and then mark last busy and initiate autosuspend for it.
3521 *
3522 * This function should be called near the end of the device's
3523 * runtime_resume callback.
3524 */
3525void blk_post_runtime_resume(struct request_queue *q, int err)
3526{
Ken Xue4fd41a852015-12-01 14:45:46 +08003527 if (!q->dev)
3528 return;
3529
Lin Ming6c954662013-03-23 11:42:26 +08003530 spin_lock_irq(q->queue_lock);
3531 if (!err) {
3532 q->rpm_status = RPM_ACTIVE;
3533 __blk_run_queue(q);
3534 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003535 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003536 } else {
3537 q->rpm_status = RPM_SUSPENDED;
3538 }
3539 spin_unlock_irq(q->queue_lock);
3540}
3541EXPORT_SYMBOL(blk_post_runtime_resume);
Mika Westerbergd07ab6d2016-02-18 10:54:11 +02003542
3543/**
3544 * blk_set_runtime_active - Force runtime status of the queue to be active
3545 * @q: the queue of the device
3546 *
3547 * If the device is left runtime suspended during system suspend the resume
3548 * hook typically resumes the device and corrects runtime status
3549 * accordingly. However, that does not affect the queue runtime PM status
3550 * which is still "suspended". This prevents processing requests from the
3551 * queue.
3552 *
3553 * This function can be used in driver's resume hook to correct queue
3554 * runtime PM status and re-enable peeking requests from the queue. It
3555 * should be called before first request is added to the queue.
3556 */
3557void blk_set_runtime_active(struct request_queue *q)
3558{
3559 spin_lock_irq(q->queue_lock);
3560 q->rpm_status = RPM_ACTIVE;
3561 pm_runtime_mark_last_busy(q->dev);
3562 pm_request_autosuspend(q->dev);
3563 spin_unlock_irq(q->queue_lock);
3564}
3565EXPORT_SYMBOL(blk_set_runtime_active);
Lin Ming6c954662013-03-23 11:42:26 +08003566#endif
3567
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568int __init blk_dev_init(void)
3569{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003570 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05303571 FIELD_SIZEOF(struct request, cmd_flags));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003572
Tejun Heo89b90be2011-01-03 15:01:47 +01003573 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3574 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003575 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 if (!kblockd_workqueue)
3577 panic("Failed to create kblockd\n");
3578
3579 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003580 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581
Ilya Dryomovc2789bd2015-11-20 22:16:46 +01003582 blk_requestq_cachep = kmem_cache_create("request_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003583 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585 return 0;
3586}
Mohan Srinivasane2d88782016-12-14 15:55:36 -08003587
3588/*
3589 * Blk IO latency support. We want this to be as cheap as possible, so doing
3590 * this lockless (and avoiding atomics), a few off by a few errors in this
3591 * code is not harmful, and we don't want to do anything that is
3592 * perf-impactful.
3593 * TODO : If necessary, we can make the histograms per-cpu and aggregate
3594 * them when printing them out.
3595 */
3596void
3597blk_zero_latency_hist(struct io_latency_state *s)
3598{
3599 memset(s->latency_y_axis_read, 0,
3600 sizeof(s->latency_y_axis_read));
3601 memset(s->latency_y_axis_write, 0,
3602 sizeof(s->latency_y_axis_write));
3603 s->latency_reads_elems = 0;
3604 s->latency_writes_elems = 0;
3605}
3606EXPORT_SYMBOL(blk_zero_latency_hist);
3607
3608ssize_t
3609blk_latency_hist_show(struct io_latency_state *s, char *buf)
3610{
3611 int i;
3612 int bytes_written = 0;
3613 u_int64_t num_elem, elem;
3614 int pct;
3615
3616 num_elem = s->latency_reads_elems;
3617 if (num_elem > 0) {
3618 bytes_written += scnprintf(buf + bytes_written,
3619 PAGE_SIZE - bytes_written,
3620 "IO svc_time Read Latency Histogram (n = %llu):\n",
3621 num_elem);
3622 for (i = 0;
3623 i < ARRAY_SIZE(latency_x_axis_us);
3624 i++) {
3625 elem = s->latency_y_axis_read[i];
3626 pct = div64_u64(elem * 100, num_elem);
3627 bytes_written += scnprintf(buf + bytes_written,
3628 PAGE_SIZE - bytes_written,
3629 "\t< %5lluus%15llu%15d%%\n",
3630 latency_x_axis_us[i],
3631 elem, pct);
3632 }
3633 /* Last element in y-axis table is overflow */
3634 elem = s->latency_y_axis_read[i];
3635 pct = div64_u64(elem * 100, num_elem);
3636 bytes_written += scnprintf(buf + bytes_written,
3637 PAGE_SIZE - bytes_written,
3638 "\t> %5dms%15llu%15d%%\n", 10,
3639 elem, pct);
3640 }
3641 num_elem = s->latency_writes_elems;
3642 if (num_elem > 0) {
3643 bytes_written += scnprintf(buf + bytes_written,
3644 PAGE_SIZE - bytes_written,
3645 "IO svc_time Write Latency Histogram (n = %llu):\n",
3646 num_elem);
3647 for (i = 0;
3648 i < ARRAY_SIZE(latency_x_axis_us);
3649 i++) {
3650 elem = s->latency_y_axis_write[i];
3651 pct = div64_u64(elem * 100, num_elem);
3652 bytes_written += scnprintf(buf + bytes_written,
3653 PAGE_SIZE - bytes_written,
3654 "\t< %5lluus%15llu%15d%%\n",
3655 latency_x_axis_us[i],
3656 elem, pct);
3657 }
3658 /* Last element in y-axis table is overflow */
3659 elem = s->latency_y_axis_write[i];
3660 pct = div64_u64(elem * 100, num_elem);
3661 bytes_written += scnprintf(buf + bytes_written,
3662 PAGE_SIZE - bytes_written,
3663 "\t> %5dms%15llu%15d%%\n", 10,
3664 elem, pct);
3665 }
3666 return bytes_written;
3667}
3668EXPORT_SYMBOL(blk_latency_hist_show);