blob: 2eb722d48773cb8a8de49d58b934eed830755da7 [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
Mike Snitzerd07335e2010-11-16 12:52:38 +010043EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020044EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070045EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060046EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010047EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010048
Tejun Heoa73f7302011-12-14 00:33:37 +010049DEFINE_IDA(blk_queue_ida);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * For the allocated request tables
53 */
Jens Axboe320ae512013-10-24 09:20:05 +010054struct kmem_cache *request_cachep = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
57 * For queue allocation
58 */
Jens Axboe6728cb02008-01-31 13:03:55 +010059struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * Controlling structure to kblockd
63 */
Jens Axboeff856ba2006-01-09 16:02:34 +010064static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Tejun Heod40f75a2015-05-22 17:13:42 -040066static void blk_clear_congested(struct request_list *rl, int sync)
67{
Tejun Heod40f75a2015-05-22 17:13:42 -040068#ifdef CONFIG_CGROUP_WRITEBACK
69 clear_wb_congested(rl->blkg->wb_congested, sync);
70#else
Tejun Heo482cf792015-05-22 17:13:43 -040071 /*
72 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
73 * flip its congestion state for events on other blkcgs.
74 */
75 if (rl == &rl->q->root_rl)
76 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
Tejun Heod40f75a2015-05-22 17:13:42 -040077#endif
78}
79
80static void blk_set_congested(struct request_list *rl, int sync)
81{
Tejun Heod40f75a2015-05-22 17:13:42 -040082#ifdef CONFIG_CGROUP_WRITEBACK
83 set_wb_congested(rl->blkg->wb_congested, sync);
84#else
Tejun Heo482cf792015-05-22 17:13:43 -040085 /* see blk_clear_congested() */
86 if (rl == &rl->q->root_rl)
87 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
Tejun Heod40f75a2015-05-22 17:13:42 -040088#endif
89}
90
Jens Axboe8324aa92008-01-29 14:51:59 +010091void blk_queue_congestion_threshold(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 int nr;
94
95 nr = q->nr_requests - (q->nr_requests / 8) + 1;
96 if (nr > q->nr_requests)
97 nr = q->nr_requests;
98 q->nr_congestion_on = nr;
99
100 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
101 if (nr < 1)
102 nr = 1;
103 q->nr_congestion_off = nr;
104}
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106/**
107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
108 * @bdev: device
109 *
110 * Locates the passed device's request queue and returns the address of its
Tejun Heoff9ea322014-09-08 08:03:56 +0900111 * backing_dev_info. This function can only be called if @bdev is opened
112 * and the return value is never NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 */
114struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
115{
Jens Axboe165125e2007-07-24 09:28:11 +0200116 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Tejun Heoff9ea322014-09-08 08:03:56 +0900118 return &q->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120EXPORT_SYMBOL(blk_get_backing_dev_info);
121
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200122void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200124 memset(rq, 0, sizeof(*rq));
125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700127 INIT_LIST_HEAD(&rq->timeout_list);
Jens Axboec7c22e42008-09-13 20:26:01 +0200128 rq->cpu = -1;
Jens Axboe63a71382008-02-08 12:41:03 +0100129 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900130 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
FUJITA Tomonorid7e3c322008-04-29 09:54:39 +0200133 rq->cmd = rq->__cmd;
Li Zefane2494e12009-04-02 13:43:26 +0800134 rq->cmd_len = BLK_MAX_CDB;
Jens Axboe63a71382008-02-08 12:41:03 +0100135 rq->tag = -1;
Tejun Heob243ddc2009-04-23 11:05:18 +0900136 rq->start_time = jiffies;
Divyesh Shah91952912010-04-01 15:01:41 -0700137 set_start_time_ns(rq);
Jerome Marchand09e099d2011-01-05 16:57:38 +0100138 rq->part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200140EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
NeilBrown5bb23a62007-09-27 12:46:13 +0200142static void req_bio_endio(struct request *rq, struct bio *bio,
143 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100144{
Mike Snitzer78d8e582015-06-26 10:01:13 -0400145 if (error)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200146 bio->bi_error = error;
Tejun Heo797e7db2006-01-06 09:51:03 +0100147
Tejun Heo143a87f2011-01-25 12:43:52 +0100148 if (unlikely(rq->cmd_flags & REQ_QUIET))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600149 bio_set_flag(bio, BIO_QUIET);
Tejun Heo143a87f2011-01-25 12:43:52 +0100150
Kent Overstreetf79ea412012-09-20 16:38:30 -0700151 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100152
153 /* don't actually finish bio if it's part of flush sequence */
Mike Snitzer78d8e582015-06-26 10:01:13 -0400154 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200155 bio_endio(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +0100156}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158void blk_dump_rq_flags(struct request *rq, char *msg)
159{
160 int bit;
161
Jens Axboe59533162013-05-23 12:25:08 +0200162 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
Jens Axboe4aff5e22006-08-10 08:44:47 +0200163 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
Jens Axboe59533162013-05-23 12:25:08 +0200164 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
Tejun Heo83096eb2009-05-07 22:24:39 +0900166 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
167 (unsigned long long)blk_rq_pos(rq),
168 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600169 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
170 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Christoph Hellwig33659eb2010-08-07 18:17:56 +0200172 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
Jens Axboe6728cb02008-01-31 13:03:55 +0100173 printk(KERN_INFO " cdb: ");
FUJITA Tomonorid34c87e2008-04-29 14:37:52 +0200174 for (bit = 0; bit < BLK_MAX_CDB; bit++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 printk("%02x ", rq->cmd[bit]);
176 printk("\n");
177 }
178}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179EXPORT_SYMBOL(blk_dump_rq_flags);
180
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500181static void blk_delay_work(struct work_struct *work)
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200182{
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500183 struct request_queue *q;
Jens Axboe6c5e0c42008-08-01 20:31:32 +0200184
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500185 q = container_of(work, struct request_queue, delay_work.work);
186 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200187 __blk_run_queue(q);
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500188 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
191/**
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500192 * blk_delay_queue - restart queueing after defined interval
193 * @q: The &struct request_queue in question
194 * @msecs: Delay in msecs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 *
196 * Description:
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500197 * Sometimes queueing needs to be postponed for a little while, to allow
198 * resources to come back. This function will make sure that queueing is
Bart Van Assche70460572012-11-28 13:45:56 +0100199 * restarted around the specified time. Queue lock must be held.
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500200 */
201void blk_delay_queue(struct request_queue *q, unsigned long msecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Bart Van Assche70460572012-11-28 13:45:56 +0100203 if (likely(!blk_queue_dead(q)))
204 queue_delayed_work(kblockd_workqueue, &q->delay_work,
205 msecs_to_jiffies(msecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500207EXPORT_SYMBOL(blk_delay_queue);
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209/**
210 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200211 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 *
213 * Description:
214 * blk_start_queue() will clear the stop flag on the queue, and call
215 * the request_fn for the queue if it was in a stopped state when
216 * entered. Also see blk_stop_queue(). Queue lock must be held.
217 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200218void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200220 WARN_ON(!irqs_disabled());
221
Nick Piggin75ad23b2008-04-29 14:48:33 +0200222 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200223 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225EXPORT_SYMBOL(blk_start_queue);
226
227/**
228 * blk_stop_queue - stop a 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 * The Linux block layer assumes that a block driver will consume all
233 * entries on the request queue when the request_fn strategy is called.
234 * Often this will not happen, because of hardware limitations (queue
235 * depth settings). If a device driver gets a 'queue full' response,
236 * or if it simply chooses not to queue more I/O at one point, it can
237 * call this function to prevent the request_fn from being called until
238 * the driver has signalled it's ready to go again. This happens by calling
239 * blk_start_queue() to restart queue operations. Queue lock must be held.
240 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200241void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Tejun Heo136b5722012-08-21 13:18:24 -0700243 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200244 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246EXPORT_SYMBOL(blk_stop_queue);
247
248/**
249 * blk_sync_queue - cancel any pending callbacks on a queue
250 * @q: the queue
251 *
252 * Description:
253 * The block layer may perform asynchronous callback activity
254 * on a queue, such as calling the unplug function after a timeout.
255 * A block device may call blk_sync_queue to ensure that any
256 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200257 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * that its ->make_request_fn will not re-add plugging prior to calling
259 * this function.
260 *
Vivek Goyalda527772011-03-02 19:05:33 -0500261 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900262 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800263 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500264 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 */
266void blk_sync_queue(struct request_queue *q)
267{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100268 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800269
270 if (q->mq_ops) {
271 struct blk_mq_hw_ctx *hctx;
272 int i;
273
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600274 queue_for_each_hw_ctx(q, hctx, i) {
275 cancel_delayed_work_sync(&hctx->run_work);
276 cancel_delayed_work_sync(&hctx->delay_work);
277 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800278 } else {
279 cancel_delayed_work_sync(&q->delay_work);
280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282EXPORT_SYMBOL(blk_sync_queue);
283
284/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100285 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
286 * @q: The queue to run
287 *
288 * Description:
289 * Invoke request handling on a queue if there are any pending requests.
290 * May be used to restart request handling after a request has completed.
291 * This variant runs the queue whether or not the queue has been
292 * stopped. Must be called with the queue lock held and interrupts
293 * disabled. See also @blk_run_queue.
294 */
295inline void __blk_run_queue_uncond(struct request_queue *q)
296{
297 if (unlikely(blk_queue_dead(q)))
298 return;
299
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100300 /*
301 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
302 * the queue lock internally. As a result multiple threads may be
303 * running such a request function concurrently. Keep track of the
304 * number of active request_fn invocations such that blk_drain_queue()
305 * can wait until all these request_fn calls have finished.
306 */
307 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100308 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100309 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100310}
Christoph Hellwiga7928c12015-04-17 22:37:20 +0200311EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
Bart Van Asschec246e802012-12-06 14:32:01 +0100312
313/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200314 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200316 *
317 * Description:
318 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200319 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200321void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
Tejun Heoa538cd02009-04-23 11:05:17 +0900323 if (unlikely(blk_queue_stopped(q)))
324 return;
325
Bart Van Asschec246e802012-12-06 14:32:01 +0100326 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200327}
328EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200329
Nick Piggin75ad23b2008-04-29 14:48:33 +0200330/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200331 * blk_run_queue_async - run a single device queue in workqueue context
332 * @q: The queue to run
333 *
334 * Description:
335 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100336 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200337 */
338void blk_run_queue_async(struct request_queue *q)
339{
Bart Van Assche70460572012-11-28 13:45:56 +0100340 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700341 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200342}
Jens Axboec21e6be2011-04-19 13:32:46 +0200343EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200344
345/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200346 * blk_run_queue - run a single device queue
347 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200348 *
349 * Description:
350 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900351 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200352 */
353void blk_run_queue(struct request_queue *q)
354{
355 unsigned long flags;
356
357 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200358 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 spin_unlock_irqrestore(q->queue_lock, flags);
360}
361EXPORT_SYMBOL(blk_run_queue);
362
Jens Axboe165125e2007-07-24 09:28:11 +0200363void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500364{
365 kobject_put(&q->kobj);
366}
Jens Axboed86e0e82011-05-27 07:44:43 +0200367EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500368
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200369/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100370 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200371 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200372 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200373 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200374 * Drain requests from @q. If @drain_all is set, all requests are drained.
375 * If not, only ELVPRIV requests are drained. The caller is responsible
376 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200377 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100378static void __blk_drain_queue(struct request_queue *q, bool drain_all)
379 __releases(q->queue_lock)
380 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200381{
Asias He458f27a2012-06-15 08:45:25 +0200382 int i;
383
Bart Van Assche807592a2012-11-28 13:43:38 +0100384 lockdep_assert_held(q->queue_lock);
385
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200386 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100387 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200388
Tejun Heob855b042012-03-06 21:24:55 +0100389 /*
390 * The caller might be trying to drain @q before its
391 * elevator is initialized.
392 */
393 if (q->elevator)
394 elv_drain_elevator(q);
395
Tejun Heo5efd6112012-03-05 13:15:12 -0800396 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200397
Tejun Heo4eabc942011-12-15 20:03:04 +0100398 /*
399 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100400 * driver init after queue creation or is not yet fully
401 * active yet. Some drivers (e.g. fd and loop) get unhappy
402 * in such cases. Kick queue iff dispatch queue has
403 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100404 */
Tejun Heob855b042012-03-06 21:24:55 +0100405 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100406 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200407
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700408 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100409 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100410
411 /*
412 * Unfortunately, requests are queued at and tracked from
413 * multiple places and there's no single counter which can
414 * be drained. Check all the queues and counters.
415 */
416 if (drain_all) {
Ming Leie97c2932014-09-25 23:23:46 +0800417 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
Tejun Heo481a7d62011-12-14 00:33:37 +0100418 drain |= !list_empty(&q->queue_head);
419 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700420 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100421 drain |= q->in_flight[i];
Ming Lei7c94e1c2014-09-25 23:23:43 +0800422 if (fq)
423 drain |= !list_empty(&fq->flush_queue[i]);
Tejun Heo481a7d62011-12-14 00:33:37 +0100424 }
425 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200426
Tejun Heo481a7d62011-12-14 00:33:37 +0100427 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200428 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100429
430 spin_unlock_irq(q->queue_lock);
431
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200432 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100433
434 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200435 }
Asias He458f27a2012-06-15 08:45:25 +0200436
437 /*
438 * With queue marked dead, any woken up waiter will fail the
439 * allocation path, so the wakeup chaining is lost and we're
440 * left with hung waiters. We need to wake up those waiters.
441 */
442 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700443 struct request_list *rl;
444
Tejun Heoa0516612012-06-26 15:05:44 -0700445 blk_queue_for_each_rl(rl, q)
446 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
447 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200448 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200449}
450
Tejun Heoc9a929d2011-10-19 14:42:16 +0200451/**
Tejun Heod7325802012-03-05 13:14:58 -0800452 * blk_queue_bypass_start - enter queue bypass mode
453 * @q: queue of interest
454 *
455 * In bypass mode, only the dispatch FIFO queue of @q is used. This
456 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800457 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700458 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
459 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800460 */
461void blk_queue_bypass_start(struct request_queue *q)
462{
463 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600464 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800465 queue_flag_set(QUEUE_FLAG_BYPASS, q);
466 spin_unlock_irq(q->queue_lock);
467
Tejun Heo776687b2014-07-01 10:29:17 -0600468 /*
469 * Queues start drained. Skip actual draining till init is
470 * complete. This avoids lenghty delays during queue init which
471 * can happen many times during boot.
472 */
473 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100474 spin_lock_irq(q->queue_lock);
475 __blk_drain_queue(q, false);
476 spin_unlock_irq(q->queue_lock);
477
Tejun Heob82d4b12012-04-13 13:11:31 -0700478 /* ensure blk_queue_bypass() is %true inside RCU read lock */
479 synchronize_rcu();
480 }
Tejun Heod7325802012-03-05 13:14:58 -0800481}
482EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
483
484/**
485 * blk_queue_bypass_end - leave queue bypass mode
486 * @q: queue of interest
487 *
488 * Leave bypass mode and restore the normal queueing behavior.
489 */
490void blk_queue_bypass_end(struct request_queue *q)
491{
492 spin_lock_irq(q->queue_lock);
493 if (!--q->bypass_depth)
494 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
495 WARN_ON_ONCE(q->bypass_depth < 0);
496 spin_unlock_irq(q->queue_lock);
497}
498EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
499
Jens Axboeaed3ea92014-12-22 14:04:42 -0700500void blk_set_queue_dying(struct request_queue *q)
501{
502 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
503
504 if (q->mq_ops)
505 blk_mq_wake_waiters(q);
506 else {
507 struct request_list *rl;
508
509 blk_queue_for_each_rl(rl, q) {
510 if (rl->rq_pool) {
511 wake_up(&rl->wait[BLK_RW_SYNC]);
512 wake_up(&rl->wait[BLK_RW_ASYNC]);
513 }
514 }
515 }
516}
517EXPORT_SYMBOL_GPL(blk_set_queue_dying);
518
Tejun Heod7325802012-03-05 13:14:58 -0800519/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200520 * blk_cleanup_queue - shutdown a request queue
521 * @q: request queue to shutdown
522 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100523 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
524 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500525 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100526void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500527{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200528 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700529
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100530 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500531 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700532 blk_set_queue_dying(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200533 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800534
Tejun Heo80fd9972012-04-13 14:50:53 -0700535 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100536 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700537 * that, unlike blk_queue_bypass_start(), we aren't performing
538 * synchronize_rcu() after entering bypass mode to avoid the delay
539 * as some drivers create and destroy a lot of queues while
540 * probing. This is still safe because blk_release_queue() will be
541 * called only after the queue refcnt drops to zero and nothing,
542 * RCU or not, would be traversing the queue by then.
543 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800544 q->bypass_depth++;
545 queue_flag_set(QUEUE_FLAG_BYPASS, q);
546
Tejun Heoc9a929d2011-10-19 14:42:16 +0200547 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
548 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100549 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200550 spin_unlock_irq(lock);
551 mutex_unlock(&q->sysfs_lock);
552
Bart Van Asschec246e802012-12-06 14:32:01 +0100553 /*
554 * Drain all requests queued before DYING marking. Set DEAD flag to
555 * prevent that q->request_fn() gets invoked after draining finished.
556 */
Ming Lei43a5e4e2013-12-26 21:31:35 +0800557 if (q->mq_ops) {
Tejun Heo780db202014-07-01 10:31:13 -0600558 blk_mq_freeze_queue(q);
Ming Lei43a5e4e2013-12-26 21:31:35 +0800559 spin_lock_irq(lock);
560 } else {
561 spin_lock_irq(lock);
562 __blk_drain_queue(q, true);
563 }
Bart Van Asschec246e802012-12-06 14:32:01 +0100564 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100565 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200566
567 /* @q won't process any more request, flush async actions */
568 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
569 blk_sync_queue(q);
570
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100571 if (q->mq_ops)
572 blk_mq_free_queue(q);
573
Asias He5e5cfac2012-05-24 23:28:52 +0800574 spin_lock_irq(lock);
575 if (q->queue_lock != &q->__queue_lock)
576 q->queue_lock = &q->__queue_lock;
577 spin_unlock_irq(lock);
578
NeilBrown6cd18e72015-04-27 14:12:22 +1000579 bdi_destroy(&q->backing_dev_info);
580
Tejun Heoc9a929d2011-10-19 14:42:16 +0200581 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500582 blk_put_queue(q);
583}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584EXPORT_SYMBOL(blk_cleanup_queue);
585
David Rientjes271508d2015-03-24 16:21:16 -0700586/* Allocate memory local to the request queue */
587static void *alloc_request_struct(gfp_t gfp_mask, void *data)
588{
589 int nid = (int)(long)data;
590 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
591}
592
593static void free_request_struct(void *element, void *unused)
594{
595 kmem_cache_free(request_cachep, element);
596}
597
Tejun Heo5b788ce2012-06-04 20:40:59 -0700598int blk_init_rl(struct request_list *rl, struct request_queue *q,
599 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400601 if (unlikely(rl->rq_pool))
602 return 0;
603
Tejun Heo5b788ce2012-06-04 20:40:59 -0700604 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200605 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
606 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200607 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
608 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
David Rientjes271508d2015-03-24 16:21:16 -0700610 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
611 free_request_struct,
612 (void *)(long)q->node, gfp_mask,
613 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (!rl->rq_pool)
615 return -ENOMEM;
616
617 return 0;
618}
619
Tejun Heo5b788ce2012-06-04 20:40:59 -0700620void blk_exit_rl(struct request_list *rl)
621{
622 if (rl->rq_pool)
623 mempool_destroy(rl->rq_pool);
624}
625
Jens Axboe165125e2007-07-24 09:28:11 +0200626struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100628 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700629}
630EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Jens Axboe165125e2007-07-24 09:28:11 +0200632struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700633{
Jens Axboe165125e2007-07-24 09:28:11 +0200634 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700635 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700636
Jens Axboe8324aa92008-01-29 14:51:59 +0100637 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700638 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (!q)
640 return NULL;
641
Dan Carpenter00380a42012-03-23 09:58:54 +0100642 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100643 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800644 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100645
Kent Overstreet54efd502015-04-23 22:37:18 -0700646 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
647 if (!q->bio_split)
648 goto fail_id;
649
Jens Axboe0989a022009-06-12 14:42:56 +0200650 q->backing_dev_info.ra_pages =
651 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
Tejun Heo89e9b9e2015-05-22 17:13:36 -0400652 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
Jens Axboed9938312009-06-12 14:45:52 +0200653 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100654 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200655
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700656 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100657 if (err)
Kent Overstreet54efd502015-04-23 22:37:18 -0700658 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700659
Matthew Garrett31373d02010-04-06 14:25:14 +0200660 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
661 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700662 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100663 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700664 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100665 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800666#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800667 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800668#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500669 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500670
Jens Axboe8324aa92008-01-29 14:51:59 +0100671 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Al Viro483f4af2006-03-18 18:34:37 -0500673 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700674 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500675
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500676 /*
677 * By default initialize queue_lock to internal lock and driver can
678 * override it later if need be.
679 */
680 q->queue_lock = &q->__queue_lock;
681
Tejun Heob82d4b12012-04-13 13:11:31 -0700682 /*
683 * A queue starts its life with bypass turned on to avoid
684 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700685 * init. The initial bypass will be finished when the queue is
686 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700687 */
688 q->bypass_depth = 1;
689 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
690
Jens Axboe320ae512013-10-24 09:20:05 +0100691 init_waitqueue_head(&q->mq_freeze_wq);
692
Tejun Heo5efd6112012-03-05 13:15:12 -0800693 if (blkcg_init_queue(q))
Mikulas Patockafff49962013-10-14 12:11:36 -0400694 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100697
Mikulas Patockafff49962013-10-14 12:11:36 -0400698fail_bdi:
699 bdi_destroy(&q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700700fail_split:
701 bioset_free(q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100702fail_id:
703 ida_simple_remove(&blk_queue_ida, q->id);
704fail_q:
705 kmem_cache_free(blk_requestq_cachep, q);
706 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707}
Christoph Lameter19460892005-06-23 00:08:19 -0700708EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
710/**
711 * blk_init_queue - prepare a request queue for use with a block device
712 * @rfn: The function to be called to process requests that have been
713 * placed on the queue.
714 * @lock: Request queue spin lock
715 *
716 * Description:
717 * If a block device wishes to use the standard request handling procedures,
718 * which sorts requests and coalesces adjacent requests, then it must
719 * call blk_init_queue(). The function @rfn will be called when there
720 * are requests on the queue that need to be processed. If the device
721 * supports plugging, then @rfn may not be called immediately when requests
722 * are available on the queue, but may be called at some time later instead.
723 * Plugged queues are generally unplugged when a buffer belonging to one
724 * of the requests on the queue is needed, or due to memory pressure.
725 *
726 * @rfn is not required, or even expected, to remove all requests off the
727 * queue, but only as many as it can handle at a time. If it does leave
728 * requests on the queue, it is responsible for arranging that the requests
729 * get dealt with eventually.
730 *
731 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200732 * request queue; this lock will be taken also from interrupt context, so irq
733 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200735 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 * it didn't succeed.
737 *
738 * Note:
739 * blk_init_queue() must be paired with a blk_cleanup_queue() call
740 * when the block device is deactivated (such as at module unload).
741 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700742
Jens Axboe165125e2007-07-24 09:28:11 +0200743struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100745 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700746}
747EXPORT_SYMBOL(blk_init_queue);
748
Jens Axboe165125e2007-07-24 09:28:11 +0200749struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700750blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
751{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600752 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600754 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
755 if (!uninit_q)
756 return NULL;
757
Mike Snitzer51514122011-11-23 10:59:13 +0100758 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600759 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700760 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700761
Mike Snitzer7982e902014-03-08 17:20:01 -0700762 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200763}
764EXPORT_SYMBOL(blk_init_queue_node);
765
Mike Snitzer336b7e12015-05-11 14:06:32 -0400766static void blk_queue_bio(struct request_queue *q, struct bio *bio);
767
Mike Snitzer01effb02010-05-11 08:57:42 +0200768struct request_queue *
769blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
770 spinlock_t *lock)
771{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 if (!q)
773 return NULL;
774
Ming Leif70ced02014-09-25 23:23:47 +0800775 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800776 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700777 return NULL;
778
Tejun Heoa0516612012-06-26 15:05:44 -0700779 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600780 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
782 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900784 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700785 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500786
787 /* Override internal queue lock with supplied lock pointer */
788 if (lock)
789 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Jens Axboef3b144a2009-03-06 08:48:33 +0100791 /*
792 * This also sets hw/phys segments, boundary and size
793 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200794 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Alan Stern44ec9542007-02-20 11:01:57 -0500796 q->sg_reserved_size = INT_MAX;
797
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600798 /* Protect q->elevator from elevator_change */
799 mutex_lock(&q->sysfs_lock);
800
Tejun Heob82d4b12012-04-13 13:11:31 -0700801 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600802 if (elevator_init(q, NULL)) {
803 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600804 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600805 }
806
807 mutex_unlock(&q->sysfs_lock);
808
Tejun Heob82d4b12012-04-13 13:11:31 -0700809 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600810
811fail:
Ming Leiba483382014-09-25 23:23:44 +0800812 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600813 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
Mike Snitzer51514122011-11-23 10:59:13 +0100815EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Tejun Heo09ac46c2011-12-14 00:33:38 +0100817bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100819 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100820 __blk_get_queue(q);
821 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
823
Tejun Heo09ac46c2011-12-14 00:33:38 +0100824 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
Jens Axboed86e0e82011-05-27 07:44:43 +0200826EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Tejun Heo5b788ce2012-06-04 20:40:59 -0700828static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100830 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700831 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100832 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100833 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100834 }
835
Tejun Heo5b788ce2012-06-04 20:40:59 -0700836 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839/*
840 * ioc_batching returns true if the ioc is a valid batching request and
841 * should be given priority access to a request.
842 */
Jens Axboe165125e2007-07-24 09:28:11 +0200843static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
845 if (!ioc)
846 return 0;
847
848 /*
849 * Make sure the process is able to allocate at least 1 request
850 * even if the batch times out, otherwise we could theoretically
851 * lose wakeups.
852 */
853 return ioc->nr_batch_requests == q->nr_batching ||
854 (ioc->nr_batch_requests > 0
855 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
856}
857
858/*
859 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
860 * will cause the process to be a "batcher" on all queues in the system. This
861 * is the behaviour we want though - once it gets a wakeup it should be given
862 * a nice run.
863 */
Jens Axboe165125e2007-07-24 09:28:11 +0200864static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
866 if (!ioc || ioc_batching(q, ioc))
867 return;
868
869 ioc->nr_batch_requests = q->nr_batching;
870 ioc->last_waited = jiffies;
871}
872
Tejun Heo5b788ce2012-06-04 20:40:59 -0700873static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700875 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Tejun Heod40f75a2015-05-22 17:13:42 -0400877 if (rl->count[sync] < queue_congestion_off_threshold(q))
878 blk_clear_congested(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Jens Axboe1faa16d2009-04-06 14:48:01 +0200880 if (rl->count[sync] + 1 <= q->nr_requests) {
881 if (waitqueue_active(&rl->wait[sync]))
882 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Tejun Heo5b788ce2012-06-04 20:40:59 -0700884 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886}
887
888/*
889 * A request has just been released. Account for it, update the full and
890 * congestion status, wake up any waiters. Called under q->queue_lock.
891 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700892static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700894 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200895 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700897 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200898 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200899 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700900 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Tejun Heo5b788ce2012-06-04 20:40:59 -0700902 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Jens Axboe1faa16d2009-04-06 14:48:01 +0200904 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700905 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906}
907
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600908int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
909{
910 struct request_list *rl;
Tejun Heod40f75a2015-05-22 17:13:42 -0400911 int on_thresh, off_thresh;
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600912
913 spin_lock_irq(q->queue_lock);
914 q->nr_requests = nr;
915 blk_queue_congestion_threshold(q);
Tejun Heod40f75a2015-05-22 17:13:42 -0400916 on_thresh = queue_congestion_on_threshold(q);
917 off_thresh = queue_congestion_off_threshold(q);
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600918
919 blk_queue_for_each_rl(rl, q) {
Tejun Heod40f75a2015-05-22 17:13:42 -0400920 if (rl->count[BLK_RW_SYNC] >= on_thresh)
921 blk_set_congested(rl, BLK_RW_SYNC);
922 else if (rl->count[BLK_RW_SYNC] < off_thresh)
923 blk_clear_congested(rl, BLK_RW_SYNC);
924
925 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
926 blk_set_congested(rl, BLK_RW_ASYNC);
927 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
928 blk_clear_congested(rl, BLK_RW_ASYNC);
929
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600930 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
931 blk_set_rl_full(rl, BLK_RW_SYNC);
932 } else {
933 blk_clear_rl_full(rl, BLK_RW_SYNC);
934 wake_up(&rl->wait[BLK_RW_SYNC]);
935 }
936
937 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
938 blk_set_rl_full(rl, BLK_RW_ASYNC);
939 } else {
940 blk_clear_rl_full(rl, BLK_RW_ASYNC);
941 wake_up(&rl->wait[BLK_RW_ASYNC]);
942 }
943 }
944
945 spin_unlock_irq(q->queue_lock);
946 return 0;
947}
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100950 * Determine if elevator data should be initialized when allocating the
951 * request associated with @bio.
952 */
953static bool blk_rq_should_init_elevator(struct bio *bio)
954{
955 if (!bio)
956 return true;
957
958 /*
959 * Flush requests do not use the elevator so skip initialization.
960 * This allows a request to share the flush and elevator data.
961 */
962 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
963 return false;
964
965 return true;
966}
967
Tejun Heoda8303c2011-10-19 14:33:05 +0200968/**
Tejun Heo852c7882012-03-05 13:15:27 -0800969 * rq_ioc - determine io_context for request allocation
970 * @bio: request being allocated is for this bio (can be %NULL)
971 *
972 * Determine io_context to use for request allocation for @bio. May return
973 * %NULL if %current->io_context doesn't exist.
974 */
975static struct io_context *rq_ioc(struct bio *bio)
976{
977#ifdef CONFIG_BLK_CGROUP
978 if (bio && bio->bi_ioc)
979 return bio->bi_ioc;
980#endif
981 return current->io_context;
982}
983
984/**
Tejun Heoa06e05e2012-06-04 20:40:55 -0700985 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -0700986 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +0200987 * @rw_flags: RW and SYNC flags
988 * @bio: bio to allocate request for (can be %NULL)
989 * @gfp_mask: allocation mask
990 *
991 * Get a free request from @q. This function may fail under memory
992 * pressure or if @q is dead.
993 *
Masanari Iidada3dae52014-09-09 01:27:23 +0900994 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -0600995 * Returns ERR_PTR on failure, with @q->queue_lock held.
996 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700998static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -0700999 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Tejun Heo5b788ce2012-06-04 20:40:59 -07001001 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -08001002 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001003 struct elevator_type *et = q->elevator->type;
1004 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +01001005 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001006 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +02001007 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001009 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -06001010 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +02001011
Jens Axboe7749a8d2006-12-13 13:02:26 +01001012 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001013 if (may_queue == ELV_MQUEUE_NO)
1014 goto rq_starved;
1015
Jens Axboe1faa16d2009-04-06 14:48:01 +02001016 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1017 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +01001018 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001019 * The queue will fill after this allocation, so set
1020 * it as full, and mark this process as "batching".
1021 * This process will be allowed to complete a batch of
1022 * requests, others will be blocked.
1023 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001024 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001025 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001026 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001027 } else {
1028 if (may_queue != ELV_MQUEUE_MUST
1029 && !ioc_batching(q, ioc)) {
1030 /*
1031 * The queue is full and the allocating
1032 * process is not a "batcher", and not
1033 * exempted by the IO scheduler
1034 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001035 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001036 }
1037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
Tejun Heod40f75a2015-05-22 17:13:42 -04001039 blk_set_congested(rl, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041
Jens Axboe082cf692005-06-28 16:35:11 +02001042 /*
1043 * Only allow batching queuers to allocate up to 50% over the defined
1044 * limit of requests, otherwise we could have thousands of requests
1045 * allocated with any setting of ->nr_requests
1046 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001047 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001048 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001049
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001050 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001051 rl->count[is_sync]++;
1052 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001053
Tejun Heof1f8cc92011-12-14 00:33:42 +01001054 /*
1055 * Decide whether the new request will be managed by elevator. If
1056 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1057 * prevent the current elevator from being destroyed until the new
1058 * request is freed. This guarantees icq's won't be destroyed and
1059 * makes creating new ones safe.
1060 *
1061 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1062 * it will be created after releasing queue_lock.
1063 */
Tejun Heod7325802012-03-05 13:14:58 -08001064 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001065 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001066 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001067 if (et->icq_cache && ioc)
1068 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001069 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001070
Jens Axboef253b862010-10-24 22:06:02 +02001071 if (blk_queue_io_stat(q))
1072 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 spin_unlock_irq(q->queue_lock);
1074
Tejun Heo29e2b092012-04-19 16:29:21 -07001075 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001076 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001077 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001078 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Tejun Heo29e2b092012-04-19 16:29:21 -07001080 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001081 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001082 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1083
Tejun Heoaaf7c682012-04-19 16:29:22 -07001084 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001085 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001086 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001087 if (ioc)
1088 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001089 if (!icq)
1090 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001091 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001092
1093 rq->elv.icq = icq;
1094 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1095 goto fail_elvpriv;
1096
1097 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001098 if (icq)
1099 get_io_context(icq->ioc);
1100 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001101out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001102 /*
1103 * ioc may be NULL here, and ioc_batching will be false. That's
1104 * OK, if the queue is under the request limit then requests need
1105 * not count toward the nr_batch_requests limit. There will always
1106 * be some limit enforced by BLK_BATCH_TIME.
1107 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 if (ioc_batching(q, ioc))
1109 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001110
Jens Axboe1faa16d2009-04-06 14:48:01 +02001111 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001113
Tejun Heoaaf7c682012-04-19 16:29:22 -07001114fail_elvpriv:
1115 /*
1116 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1117 * and may fail indefinitely under memory pressure and thus
1118 * shouldn't stall IO. Treat this request as !elvpriv. This will
1119 * disturb iosched and blkcg but weird is bettern than dead.
1120 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001121 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1122 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001123
1124 rq->cmd_flags &= ~REQ_ELVPRIV;
1125 rq->elv.icq = NULL;
1126
1127 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001128 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001129 spin_unlock_irq(q->queue_lock);
1130 goto out;
1131
Tejun Heob6792812012-03-05 13:15:23 -08001132fail_alloc:
1133 /*
1134 * Allocation failed presumably due to memory. Undo anything we
1135 * might have messed up.
1136 *
1137 * Allocating task should really be put onto the front of the wait
1138 * queue, but this is pretty rare.
1139 */
1140 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001141 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001142
1143 /*
1144 * in the very unlikely event that allocation failed and no
1145 * requests for this direction was pending, mark us starved so that
1146 * freeing of a request in the other direction will notice
1147 * us. another possible fix would be to split the rq mempool into
1148 * READ and WRITE
1149 */
1150rq_starved:
1151 if (unlikely(rl->count[is_sync] == 0))
1152 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001153 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154}
1155
Tejun Heoda8303c2011-10-19 14:33:05 +02001156/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001157 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001158 * @q: request_queue to allocate request from
1159 * @rw_flags: RW and SYNC flags
1160 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001161 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001162 *
Tejun Heoa06e05e2012-06-04 20:40:55 -07001163 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1164 * function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001165 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001166 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001167 * Returns ERR_PTR on failure, with @q->queue_lock held.
1168 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001170static struct request *get_request(struct request_queue *q, int rw_flags,
1171 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001173 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001174 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001175 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001177
1178 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001179retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001180 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001181 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001182 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001184 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001185 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001186 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Tejun Heoa06e05e2012-06-04 20:40:55 -07001189 /* wait on @rl and retry */
1190 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1191 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001192
Tejun Heoa06e05e2012-06-04 20:40:55 -07001193 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Tejun Heoa06e05e2012-06-04 20:40:55 -07001195 spin_unlock_irq(q->queue_lock);
1196 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Tejun Heoa06e05e2012-06-04 20:40:55 -07001198 /*
1199 * After sleeping, we become a "batching" process and will be able
1200 * to allocate at least one request, and up to a big batch of them
1201 * for a small period time. See ioc_batching, ioc_set_batching
1202 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001203 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Tejun Heoa06e05e2012-06-04 20:40:55 -07001205 spin_lock_irq(q->queue_lock);
1206 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001207
Tejun Heoa06e05e2012-06-04 20:40:55 -07001208 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209}
1210
Jens Axboe320ae512013-10-24 09:20:05 +01001211static struct request *blk_old_get_request(struct request_queue *q, int rw,
1212 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
1214 struct request *rq;
1215
1216 BUG_ON(rw != READ && rw != WRITE);
1217
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001218 /* create ioc upfront */
1219 create_io_context(gfp_mask, q->node);
1220
Nick Piggind6344532005-06-28 20:45:14 -07001221 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001222 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001223 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001224 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001225 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 return rq;
1228}
Jens Axboe320ae512013-10-24 09:20:05 +01001229
1230struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1231{
1232 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001233 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001234 else
1235 return blk_old_get_request(q, rw, gfp_mask);
1236}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237EXPORT_SYMBOL(blk_get_request);
1238
1239/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001240 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001241 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001242 * @bio: The bio describing the memory mappings that will be submitted for IO.
1243 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001244 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001245 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001246 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1247 * type commands. Where the struct request needs to be farther initialized by
1248 * the caller. It is passed a &struct bio, which describes the memory info of
1249 * the I/O transfer.
1250 *
1251 * The caller of blk_make_request must make sure that bi_io_vec
1252 * are set to describe the memory buffers. That bio_data_dir() will return
1253 * the needed direction of the request. (And all bio's in the passed bio-chain
1254 * are properly set accordingly)
1255 *
1256 * If called under none-sleepable conditions, mapped bio buffers must not
1257 * need bouncing, by calling the appropriate masked or flagged allocator,
1258 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1259 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001260 *
1261 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1262 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1263 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1264 * completion of a bio that hasn't been submitted yet, thus resulting in a
1265 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1266 * of bio_alloc(), as that avoids the mempool deadlock.
1267 * If possible a big IO should be split into smaller parts when allocation
1268 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001269 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001270struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1271 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001272{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001273 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1274
Joe Lawrencea492f072014-08-28 08:15:21 -06001275 if (IS_ERR(rq))
1276 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001277
Jens Axboef27b0872014-06-06 07:57:37 -06001278 blk_rq_set_block_pc(rq);
1279
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001280 for_each_bio(bio) {
1281 struct bio *bounce_bio = bio;
1282 int ret;
1283
1284 blk_queue_bounce(q, &bounce_bio);
1285 ret = blk_rq_append_bio(q, rq, bounce_bio);
1286 if (unlikely(ret)) {
1287 blk_put_request(rq);
1288 return ERR_PTR(ret);
1289 }
1290 }
1291
1292 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001293}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001294EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001295
1296/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001297 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001298 * @rq: request to be initialized
1299 *
1300 */
1301void blk_rq_set_block_pc(struct request *rq)
1302{
1303 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1304 rq->__data_len = 0;
1305 rq->__sector = (sector_t) -1;
1306 rq->bio = rq->biotail = NULL;
1307 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001308}
1309EXPORT_SYMBOL(blk_rq_set_block_pc);
1310
1311/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 * blk_requeue_request - put a request back on queue
1313 * @q: request queue where request should be inserted
1314 * @rq: request to be inserted
1315 *
1316 * Description:
1317 * Drivers often keep queueing requests until the hardware cannot accept
1318 * more, when that condition happens we need to put the request back
1319 * on the queue. Must be called with queue lock held.
1320 */
Jens Axboe165125e2007-07-24 09:28:11 +02001321void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001323 blk_delete_timer(rq);
1324 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001325 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001326
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001327 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 blk_queue_end_tag(q, rq);
1329
James Bottomleyba396a62009-05-27 14:17:08 +02001330 BUG_ON(blk_queued_rq(rq));
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 elv_requeue_request(q, rq);
1333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334EXPORT_SYMBOL(blk_requeue_request);
1335
Jens Axboe73c10102011-03-08 13:19:51 +01001336static void add_acct_request(struct request_queue *q, struct request *rq,
1337 int where)
1338{
Jens Axboe320ae512013-10-24 09:20:05 +01001339 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001340 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001341}
1342
Tejun Heo074a7ac2008-08-25 19:56:14 +09001343static void part_round_stats_single(int cpu, struct hd_struct *part,
1344 unsigned long now)
1345{
Jens Axboe7276d022014-05-09 15:48:23 -06001346 int inflight;
1347
Tejun Heo074a7ac2008-08-25 19:56:14 +09001348 if (now == part->stamp)
1349 return;
1350
Jens Axboe7276d022014-05-09 15:48:23 -06001351 inflight = part_in_flight(part);
1352 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001353 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001354 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001355 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1356 }
1357 part->stamp = now;
1358}
1359
1360/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001361 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1362 * @cpu: cpu number for stats access
1363 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 *
1365 * The average IO queue length and utilisation statistics are maintained
1366 * by observing the current state of the queue length and the amount of
1367 * time it has been in this state for.
1368 *
1369 * Normally, that accounting is done on IO completion, but that can result
1370 * in more than a second's worth of IO being accounted for within any one
1371 * second, leading to >100% utilisation. To deal with that, we call this
1372 * function to do a round-off before returning the results when reading
1373 * /proc/diskstats. This accounts immediately for all queue usage up to
1374 * the current jiffies and restarts the counters again.
1375 */
Tejun Heoc9959052008-08-25 19:47:21 +09001376void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001377{
1378 unsigned long now = jiffies;
1379
Tejun Heo074a7ac2008-08-25 19:56:14 +09001380 if (part->partno)
1381 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1382 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001383}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001384EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001385
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001386#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001387static void blk_pm_put_request(struct request *rq)
1388{
1389 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1390 pm_runtime_mark_last_busy(rq->q->dev);
1391}
1392#else
1393static inline void blk_pm_put_request(struct request *rq) {}
1394#endif
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396/*
1397 * queue lock must be held
1398 */
Jens Axboe165125e2007-07-24 09:28:11 +02001399void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (unlikely(!q))
1402 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001404 if (q->mq_ops) {
1405 blk_mq_free_request(req);
1406 return;
1407 }
1408
Lin Mingc8158812013-03-23 11:42:27 +08001409 blk_pm_put_request(req);
1410
Tejun Heo8922e162005-10-20 16:23:44 +02001411 elv_completed_request(q, req);
1412
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001413 /* this is a bio leak */
1414 WARN_ON(req->bio != NULL);
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 /*
1417 * Request may not have originated from ll_rw_blk. if not,
1418 * it didn't come out of our reserved rq pools
1419 */
Jens Axboe49171e52006-08-10 08:59:11 +02001420 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001421 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001422 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001425 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Tejun Heoa0516612012-06-26 15:05:44 -07001427 blk_free_request(rl, req);
1428 freed_request(rl, flags);
1429 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 }
1431}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001432EXPORT_SYMBOL_GPL(__blk_put_request);
1433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434void blk_put_request(struct request *req)
1435{
Jens Axboe165125e2007-07-24 09:28:11 +02001436 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Jens Axboe320ae512013-10-24 09:20:05 +01001438 if (q->mq_ops)
1439 blk_mq_free_request(req);
1440 else {
1441 unsigned long flags;
1442
1443 spin_lock_irqsave(q->queue_lock, flags);
1444 __blk_put_request(q, req);
1445 spin_unlock_irqrestore(q->queue_lock, flags);
1446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448EXPORT_SYMBOL(blk_put_request);
1449
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001450/**
1451 * blk_add_request_payload - add a payload to a request
1452 * @rq: request to update
1453 * @page: page backing the payload
1454 * @len: length of the payload.
1455 *
1456 * This allows to later add a payload to an already submitted request by
1457 * a block driver. The driver needs to take care of freeing the payload
1458 * itself.
1459 *
1460 * Note that this is a quite horrible hack and nothing but handling of
1461 * discard requests should ever use it.
1462 */
1463void blk_add_request_payload(struct request *rq, struct page *page,
1464 unsigned int len)
1465{
1466 struct bio *bio = rq->bio;
1467
1468 bio->bi_io_vec->bv_page = page;
1469 bio->bi_io_vec->bv_offset = 0;
1470 bio->bi_io_vec->bv_len = len;
1471
Kent Overstreet4f024f32013-10-11 15:44:27 -07001472 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001473 bio->bi_vcnt = 1;
1474 bio->bi_phys_segments = 1;
1475
1476 rq->__data_len = rq->resid_len = len;
1477 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001478}
1479EXPORT_SYMBOL_GPL(blk_add_request_payload);
1480
Jens Axboe320ae512013-10-24 09:20:05 +01001481bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1482 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001483{
1484 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1485
Jens Axboe73c10102011-03-08 13:19:51 +01001486 if (!ll_back_merge_fn(q, req, bio))
1487 return false;
1488
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001489 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001490
1491 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1492 blk_rq_set_mixed_merge(req);
1493
1494 req->biotail->bi_next = bio;
1495 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001496 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001497 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1498
Jens Axboe320ae512013-10-24 09:20:05 +01001499 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001500 return true;
1501}
1502
Jens Axboe320ae512013-10-24 09:20:05 +01001503bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1504 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001505{
1506 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001507
Jens Axboe73c10102011-03-08 13:19:51 +01001508 if (!ll_front_merge_fn(q, req, bio))
1509 return false;
1510
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001511 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001512
1513 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1514 blk_rq_set_mixed_merge(req);
1515
Jens Axboe73c10102011-03-08 13:19:51 +01001516 bio->bi_next = req->bio;
1517 req->bio = bio;
1518
Kent Overstreet4f024f32013-10-11 15:44:27 -07001519 req->__sector = bio->bi_iter.bi_sector;
1520 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001521 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1522
Jens Axboe320ae512013-10-24 09:20:05 +01001523 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001524 return true;
1525}
1526
Tejun Heobd87b582011-10-19 14:33:08 +02001527/**
Jens Axboe320ae512013-10-24 09:20:05 +01001528 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001529 * @q: request_queue new bio is being queued at
1530 * @bio: new bio being queued
1531 * @request_count: out parameter for number of traversed plugged requests
1532 *
1533 * Determine whether @bio being queued on @q can be merged with a request
1534 * on %current's plugged list. Returns %true if merge was successful,
1535 * otherwise %false.
1536 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001537 * Plugging coalesces IOs from the same issuer for the same purpose without
1538 * going through @q->queue_lock. As such it's more of an issuing mechanism
1539 * than scheduling, and the request, while may have elvpriv data, is not
1540 * added on the elevator at this point. In addition, we don't have
1541 * reliable access to the elevator outside queue lock. Only check basic
1542 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001543 *
1544 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001545 */
Jens Axboe320ae512013-10-24 09:20:05 +01001546bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Shaohua Li5b3f3412015-05-08 10:51:33 -07001547 unsigned int *request_count,
1548 struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +01001549{
1550 struct blk_plug *plug;
1551 struct request *rq;
1552 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001553 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001554
Tejun Heobd87b582011-10-19 14:33:08 +02001555 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001556 if (!plug)
1557 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001558 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001559
Shaohua Li92f399c2013-10-29 12:01:03 -06001560 if (q->mq_ops)
1561 plug_list = &plug->mq_list;
1562 else
1563 plug_list = &plug->list;
1564
1565 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001566 int el_ret;
1567
Shaohua Li5b3f3412015-05-08 10:51:33 -07001568 if (rq->q == q) {
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001569 (*request_count)++;
Shaohua Li5b3f3412015-05-08 10:51:33 -07001570 /*
1571 * Only blk-mq multiple hardware queues case checks the
1572 * rq in the same queue, there should be only one such
1573 * rq in a queue
1574 **/
1575 if (same_queue_rq)
1576 *same_queue_rq = rq;
1577 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001578
Tejun Heo07c2bd32012-02-08 09:19:42 +01001579 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001580 continue;
1581
Tejun Heo050c8ea2012-02-08 09:19:38 +01001582 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001583 if (el_ret == ELEVATOR_BACK_MERGE) {
1584 ret = bio_attempt_back_merge(q, rq, bio);
1585 if (ret)
1586 break;
1587 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1588 ret = bio_attempt_front_merge(q, rq, bio);
1589 if (ret)
1590 break;
1591 }
1592 }
1593out:
1594 return ret;
1595}
1596
Jens Axboe86db1e22008-01-29 14:53:40 +01001597void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001598{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001599 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001600
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001601 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1602 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001603 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001604
Tejun Heo52d9e672006-01-06 09:49:58 +01001605 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001606 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001607 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001608 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001609}
1610
Mike Snitzer336b7e12015-05-11 14:06:32 -04001611static void blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001613 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001614 struct blk_plug *plug;
1615 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1616 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001617 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Kent Overstreet54efd502015-04-23 22:37:18 -07001619 blk_queue_split(q, &bio, q->bio_split);
1620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 /*
1622 * low level driver can indicate that it wants pages above a
1623 * certain limit bounced to low memory (ie for highmem, or even
1624 * ISA dma in theory)
1625 */
1626 blk_queue_bounce(q, &bio);
1627
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001628 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001629 bio->bi_error = -EIO;
1630 bio_endio(bio);
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001631 return;
1632 }
1633
Tejun Heo4fed9472010-09-03 11:56:17 +02001634 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001635 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001636 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001637 goto get_rq;
1638 }
1639
Jens Axboe73c10102011-03-08 13:19:51 +01001640 /*
1641 * Check if we can merge with the plugged list before grabbing
1642 * any locks.
1643 */
Robert Elliottda41a582014-05-20 16:46:26 -05001644 if (!blk_queue_nomerges(q) &&
Shaohua Li5b3f3412015-05-08 10:51:33 -07001645 blk_attempt_plug_merge(q, bio, &request_count, NULL))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001646 return;
Jens Axboe73c10102011-03-08 13:19:51 +01001647
1648 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
1650 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001651 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001652 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001653 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001654 if (!attempt_back_merge(q, req))
1655 elv_merged_request(q, req, el_ret);
1656 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001657 }
Jens Axboe73c10102011-03-08 13:19:51 +01001658 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001659 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001660 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001661 if (!attempt_front_merge(q, req))
1662 elv_merged_request(q, req, el_ret);
1663 goto out_unlock;
1664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
1666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001668 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001669 * This sync check and mask will be re-done in init_request_from_bio(),
1670 * but we need to set it earlier to expose the sync flag to the
1671 * rq allocator and io schedulers.
1672 */
1673 rw_flags = bio_data_dir(bio);
1674 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001675 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001676
1677 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001678 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001679 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001680 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001681 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001682 if (IS_ERR(req)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001683 bio->bi_error = PTR_ERR(req);
1684 bio_endio(bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001685 goto out_unlock;
1686 }
Nick Piggind6344532005-06-28 20:45:14 -07001687
Nick Piggin450991b2005-06-28 20:45:13 -07001688 /*
1689 * After dropping the lock and possibly sleeping here, our request
1690 * may now be mergeable after it had proven unmergeable (above).
1691 * We don't worry about that case for efficiency. It won't happen
1692 * often, and the elevators are able to handle it.
1693 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001694 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
Tao Ma9562ad92011-10-24 16:11:30 +02001696 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001697 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001698
Jens Axboe73c10102011-03-08 13:19:51 +01001699 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001700 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001701 /*
1702 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001703 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001704 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001705 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001706 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001707 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001708 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001709 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001710 trace_block_plug(q);
1711 }
Jens Axboe73c10102011-03-08 13:19:51 +01001712 }
Shaohua Lia6327162011-08-24 16:04:32 +02001713 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001714 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001715 } else {
1716 spin_lock_irq(q->queue_lock);
1717 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001718 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001719out_unlock:
1720 spin_unlock_irq(q->queue_lock);
1721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722}
1723
1724/*
1725 * If bio->bi_dev is a partition, remap the location
1726 */
1727static inline void blk_partition_remap(struct bio *bio)
1728{
1729 struct block_device *bdev = bio->bi_bdev;
1730
Jens Axboebf2de6f2007-09-27 13:01:25 +02001731 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001733
Kent Overstreet4f024f32013-10-11 15:44:27 -07001734 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001736
Mike Snitzerd07335e2010-11-16 12:52:38 +01001737 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1738 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001739 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 }
1741}
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743static void handle_bad_sector(struct bio *bio)
1744{
1745 char b[BDEVNAME_SIZE];
1746
1747 printk(KERN_INFO "attempt to access beyond end of device\n");
1748 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1749 bdevname(bio->bi_bdev, b),
1750 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001751 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001752 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753}
1754
Akinobu Mitac17bb492006-12-08 02:39:46 -08001755#ifdef CONFIG_FAIL_MAKE_REQUEST
1756
1757static DECLARE_FAULT_ATTR(fail_make_request);
1758
1759static int __init setup_fail_make_request(char *str)
1760{
1761 return setup_fault_attr(&fail_make_request, str);
1762}
1763__setup("fail_make_request=", setup_fail_make_request);
1764
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001765static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001766{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001767 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001768}
1769
1770static int __init fail_make_request_debugfs(void)
1771{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001772 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1773 NULL, &fail_make_request);
1774
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001775 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001776}
1777
1778late_initcall(fail_make_request_debugfs);
1779
1780#else /* CONFIG_FAIL_MAKE_REQUEST */
1781
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001782static inline bool should_fail_request(struct hd_struct *part,
1783 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001784{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001785 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001786}
1787
1788#endif /* CONFIG_FAIL_MAKE_REQUEST */
1789
Jens Axboec07e2b42007-07-18 13:27:58 +02001790/*
1791 * Check whether this bio extends beyond the end of the device.
1792 */
1793static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1794{
1795 sector_t maxsector;
1796
1797 if (!nr_sectors)
1798 return 0;
1799
1800 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001801 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001802 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001803 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001804
1805 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1806 /*
1807 * This may well happen - the kernel calls bread()
1808 * without checking the size of the device, e.g., when
1809 * mounting a device.
1810 */
1811 handle_bad_sector(bio);
1812 return 1;
1813 }
1814 }
1815
1816 return 0;
1817}
1818
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001819static noinline_for_stack bool
1820generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821{
Jens Axboe165125e2007-07-24 09:28:11 +02001822 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001823 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001824 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001825 char b[BDEVNAME_SIZE];
1826 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Jens Axboec07e2b42007-07-18 13:27:58 +02001830 if (bio_check_eod(bio, nr_sectors))
1831 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001833 q = bdev_get_queue(bio->bi_bdev);
1834 if (unlikely(!q)) {
1835 printk(KERN_ERR
1836 "generic_make_request: Trying to access "
1837 "nonexistent block-device %s (%Lu)\n",
1838 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001839 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001840 goto end_io;
1841 }
1842
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001843 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001844 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001845 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001846 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001847 goto end_io;
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001850 * If this device has partitions, remap block n
1851 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001853 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001855 if (bio_check_eod(bio, nr_sectors))
1856 goto end_io;
1857
1858 /*
1859 * Filter flush bio's early so that make_request based
1860 * drivers without flush support don't have to worry
1861 * about them.
1862 */
1863 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1864 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1865 if (!nr_sectors) {
1866 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001867 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001871 if ((bio->bi_rw & REQ_DISCARD) &&
1872 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001873 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001874 err = -EOPNOTSUPP;
1875 goto end_io;
1876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001878 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 err = -EOPNOTSUPP;
1880 goto end_io;
1881 }
1882
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001883 /*
1884 * Various block parts want %current->io_context and lazy ioc
1885 * allocation ends up trading a lot of pain for a small amount of
1886 * memory. Just allocate it upfront. This may fail and block
1887 * layer knows how to live with it.
1888 */
1889 create_io_context(GFP_ATOMIC, q->node);
1890
Tejun Heoae118892015-08-18 14:55:20 -07001891 if (!blkcg_bio_issue_check(q, bio))
1892 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001893
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001894 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001895 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001896
1897end_io:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001898 bio->bi_error = err;
1899 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001900 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901}
1902
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001903/**
1904 * generic_make_request - hand a buffer to its device driver for I/O
1905 * @bio: The bio describing the location in memory and on the device.
1906 *
1907 * generic_make_request() is used to make I/O requests of block
1908 * devices. It is passed a &struct bio, which describes the I/O that needs
1909 * to be done.
1910 *
1911 * generic_make_request() does not return any status. The
1912 * success/failure status of the request, along with notification of
1913 * completion, is delivered asynchronously through the bio->bi_end_io
1914 * function described (one day) else where.
1915 *
1916 * The caller of generic_make_request must make sure that bi_io_vec
1917 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1918 * set to describe the device address, and the
1919 * bi_end_io and optionally bi_private are set to describe how
1920 * completion notification should be signaled.
1921 *
1922 * generic_make_request and the drivers it calls may use bi_next if this
1923 * bio happens to be merged with someone else, and may resubmit the bio to
1924 * a lower device by calling into generic_make_request recursively, which
1925 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001926 */
1927void generic_make_request(struct bio *bio)
1928{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001929 struct bio_list bio_list_on_stack;
1930
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001931 if (!generic_make_request_checks(bio))
1932 return;
1933
1934 /*
1935 * We only want one ->make_request_fn to be active at a time, else
1936 * stack usage with stacked devices could be a problem. So use
1937 * current->bio_list to keep a list of requests submited by a
1938 * make_request_fn function. current->bio_list is also used as a
1939 * flag to say if generic_make_request is currently active in this
1940 * task or not. If it is NULL, then no make_request is active. If
1941 * it is non-NULL, then a make_request is active, and new requests
1942 * should be added at the tail
1943 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001944 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001945 bio_list_add(current->bio_list, bio);
Neil Brownd89d8792007-05-01 09:53:42 +02001946 return;
1947 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001948
Neil Brownd89d8792007-05-01 09:53:42 +02001949 /* following loop may be a bit non-obvious, and so deserves some
1950 * explanation.
1951 * Before entering the loop, bio->bi_next is NULL (as all callers
1952 * ensure that) so we have a list with a single bio.
1953 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001954 * we assign bio_list to a pointer to the bio_list_on_stack,
1955 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001956 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001957 * through a recursive call to generic_make_request. If it
1958 * did, we find a non-NULL value in bio_list and re-enter the loop
1959 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001960 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001961 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001962 */
1963 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001964 bio_list_init(&bio_list_on_stack);
1965 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001966 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001967 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1968
1969 q->make_request_fn(q, bio);
1970
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001971 bio = bio_list_pop(current->bio_list);
Neil Brownd89d8792007-05-01 09:53:42 +02001972 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001973 current->bio_list = NULL; /* deactivate */
Neil Brownd89d8792007-05-01 09:53:42 +02001974}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975EXPORT_SYMBOL(generic_make_request);
1976
1977/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001978 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1980 * @bio: The &struct bio which describes the I/O
1981 *
1982 * submit_bio() is very similar in purpose to generic_make_request(), and
1983 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001984 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 *
1986 */
1987void submit_bio(int rw, struct bio *bio)
1988{
Jens Axboe22e2c502005-06-27 10:55:12 +02001989 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
Jens Axboebf2de6f2007-09-27 13:01:25 +02001991 /*
1992 * If it's a regular read/write or a barrier with data attached,
1993 * go through the normal accounting stuff before submission.
1994 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001995 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001996 unsigned int count;
1997
1998 if (unlikely(rw & REQ_WRITE_SAME))
1999 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2000 else
2001 count = bio_sectors(bio);
2002
Jens Axboebf2de6f2007-09-27 13:01:25 +02002003 if (rw & WRITE) {
2004 count_vm_events(PGPGOUT, count);
2005 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07002006 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002007 count_vm_events(PGPGIN, count);
2008 }
2009
2010 if (unlikely(block_dump)) {
2011 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02002012 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07002013 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02002014 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002015 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02002016 bdevname(bio->bi_bdev, b),
2017 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
2020
2021 generic_make_request(bio);
2022}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023EXPORT_SYMBOL(submit_bio);
2024
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002025/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002026 * blk_rq_check_limits - Helper function to check a request for the queue limit
2027 * @q: the queue
2028 * @rq: the request being checked
2029 *
2030 * Description:
2031 * @rq may have been made based on weaker limitations of upper-level queues
2032 * in request stacking drivers, and it may violate the limitation of @q.
2033 * Since the block layer and the underlying device driver trust @rq
2034 * after it is inserted to @q, it should be checked against @q before
2035 * the insertion using this generic function.
2036 *
2037 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02002038 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002039 * Request stacking drivers like request-based dm may change the queue
2040 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09002041 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002042 * the new queue limits again when they dispatch those requests,
2043 * although such checkings are also done against the old queue limits
2044 * when submitting requests.
2045 */
2046int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2047{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002048 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04002049 return 0;
2050
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04002051 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002052 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2053 return -EIO;
2054 }
2055
2056 /*
2057 * queue's settings related to segment counting like q->bounce_pfn
2058 * may differ from that of other stacking queues.
2059 * Recalculate it to check the request correctly on this queue's
2060 * limitation.
2061 */
2062 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002063 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002064 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2065 return -EIO;
2066 }
2067
2068 return 0;
2069}
2070EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2071
2072/**
2073 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2074 * @q: the queue to submit the request
2075 * @rq: the request being queued
2076 */
2077int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2078{
2079 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002080 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002081
2082 if (blk_rq_check_limits(q, rq))
2083 return -EIO;
2084
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002085 if (rq->rq_disk &&
2086 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002087 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002088
Keith Busch7fb48982014-10-17 17:46:38 -06002089 if (q->mq_ops) {
2090 if (blk_queue_io_stat(q))
2091 blk_account_io_start(rq, true);
2092 blk_mq_insert_request(rq, false, true, true);
2093 return 0;
2094 }
2095
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002096 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002097 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002098 spin_unlock_irqrestore(q->queue_lock, flags);
2099 return -ENODEV;
2100 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002101
2102 /*
2103 * Submitting request must be dequeued before calling this function
2104 * because it will be linked to another request_queue
2105 */
2106 BUG_ON(blk_queued_rq(rq));
2107
Jeff Moyer4853aba2011-08-15 21:37:25 +02002108 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2109 where = ELEVATOR_INSERT_FLUSH;
2110
2111 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002112 if (where == ELEVATOR_INSERT_FLUSH)
2113 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002114 spin_unlock_irqrestore(q->queue_lock, flags);
2115
2116 return 0;
2117}
2118EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2119
Tejun Heo80a761f2009-07-03 17:48:17 +09002120/**
2121 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2122 * @rq: request to examine
2123 *
2124 * Description:
2125 * A request could be merge of IOs which require different failure
2126 * handling. This function determines the number of bytes which
2127 * can be failed from the beginning of the request without
2128 * crossing into area which need to be retried further.
2129 *
2130 * Return:
2131 * The number of bytes to fail.
2132 *
2133 * Context:
2134 * queue_lock must be held.
2135 */
2136unsigned int blk_rq_err_bytes(const struct request *rq)
2137{
2138 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2139 unsigned int bytes = 0;
2140 struct bio *bio;
2141
2142 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2143 return blk_rq_bytes(rq);
2144
2145 /*
2146 * Currently the only 'mixing' which can happen is between
2147 * different fastfail types. We can safely fail portions
2148 * which have all the failfast bits that the first one has -
2149 * the ones which are at least as eager to fail as the first
2150 * one.
2151 */
2152 for (bio = rq->bio; bio; bio = bio->bi_next) {
2153 if ((bio->bi_rw & ff) != ff)
2154 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002155 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002156 }
2157
2158 /* this could lead to infinite loop */
2159 BUG_ON(blk_rq_bytes(rq) && !bytes);
2160 return bytes;
2161}
2162EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2163
Jens Axboe320ae512013-10-24 09:20:05 +01002164void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002165{
Jens Axboec2553b52009-04-24 08:10:11 +02002166 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002167 const int rw = rq_data_dir(req);
2168 struct hd_struct *part;
2169 int cpu;
2170
2171 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002172 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002173 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2174 part_stat_unlock();
2175 }
2176}
2177
Jens Axboe320ae512013-10-24 09:20:05 +01002178void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002179{
Jens Axboebc58ba92009-01-23 10:54:44 +01002180 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002181 * Account IO completion. flush_rq isn't accounted as a
2182 * normal IO on queueing nor completion. Accounting the
2183 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002184 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002185 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002186 unsigned long duration = jiffies - req->start_time;
2187 const int rw = rq_data_dir(req);
2188 struct hd_struct *part;
2189 int cpu;
2190
2191 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002192 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002193
2194 part_stat_inc(cpu, part, ios[rw]);
2195 part_stat_add(cpu, part, ticks[rw], duration);
2196 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002197 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002198
Jens Axboe6c23a962011-01-07 08:43:37 +01002199 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002200 part_stat_unlock();
2201 }
2202}
2203
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002204#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002205/*
2206 * Don't process normal requests when queue is suspended
2207 * or in the process of suspending/resuming
2208 */
2209static struct request *blk_pm_peek_request(struct request_queue *q,
2210 struct request *rq)
2211{
2212 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2213 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2214 return NULL;
2215 else
2216 return rq;
2217}
2218#else
2219static inline struct request *blk_pm_peek_request(struct request_queue *q,
2220 struct request *rq)
2221{
2222 return rq;
2223}
2224#endif
2225
Jens Axboe320ae512013-10-24 09:20:05 +01002226void blk_account_io_start(struct request *rq, bool new_io)
2227{
2228 struct hd_struct *part;
2229 int rw = rq_data_dir(rq);
2230 int cpu;
2231
2232 if (!blk_do_io_stat(rq))
2233 return;
2234
2235 cpu = part_stat_lock();
2236
2237 if (!new_io) {
2238 part = rq->part;
2239 part_stat_inc(cpu, part, merges[rw]);
2240 } else {
2241 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2242 if (!hd_struct_try_get(part)) {
2243 /*
2244 * The partition is already being removed,
2245 * the request will be accounted on the disk only
2246 *
2247 * We take a reference on disk->part0 although that
2248 * partition will never be deleted, so we can treat
2249 * it as any other partition.
2250 */
2251 part = &rq->rq_disk->part0;
2252 hd_struct_get(part);
2253 }
2254 part_round_stats(cpu, part);
2255 part_inc_in_flight(part, rw);
2256 rq->part = part;
2257 }
2258
2259 part_stat_unlock();
2260}
2261
Tejun Heo53a08802008-12-03 12:41:26 +01002262/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002263 * blk_peek_request - peek at the top of a request queue
2264 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002265 *
2266 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002267 * Return the request at the top of @q. The returned request
2268 * should be started using blk_start_request() before LLD starts
2269 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002270 *
2271 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002272 * Pointer to the request at the top of @q if available. Null
2273 * otherwise.
2274 *
2275 * Context:
2276 * queue_lock must be held.
2277 */
2278struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002279{
2280 struct request *rq;
2281 int ret;
2282
2283 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002284
2285 rq = blk_pm_peek_request(q, rq);
2286 if (!rq)
2287 break;
2288
Tejun Heo158dbda2009-04-23 11:05:18 +09002289 if (!(rq->cmd_flags & REQ_STARTED)) {
2290 /*
2291 * This is the first time the device driver
2292 * sees this request (possibly after
2293 * requeueing). Notify IO scheduler.
2294 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002295 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002296 elv_activate_rq(q, rq);
2297
2298 /*
2299 * just mark as started even if we don't start
2300 * it, a request that has been delayed should
2301 * not be passed by new incoming requests
2302 */
2303 rq->cmd_flags |= REQ_STARTED;
2304 trace_block_rq_issue(q, rq);
2305 }
2306
2307 if (!q->boundary_rq || q->boundary_rq == rq) {
2308 q->end_sector = rq_end_sector(rq);
2309 q->boundary_rq = NULL;
2310 }
2311
2312 if (rq->cmd_flags & REQ_DONTPREP)
2313 break;
2314
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002315 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002316 /*
2317 * make sure space for the drain appears we
2318 * know we can do this because max_hw_segments
2319 * has been adjusted to be one fewer than the
2320 * device can handle
2321 */
2322 rq->nr_phys_segments++;
2323 }
2324
2325 if (!q->prep_rq_fn)
2326 break;
2327
2328 ret = q->prep_rq_fn(q, rq);
2329 if (ret == BLKPREP_OK) {
2330 break;
2331 } else if (ret == BLKPREP_DEFER) {
2332 /*
2333 * the request may have been (partially) prepped.
2334 * we need to keep this request in the front to
2335 * avoid resource deadlock. REQ_STARTED will
2336 * prevent other fs requests from passing this one.
2337 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002338 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002339 !(rq->cmd_flags & REQ_DONTPREP)) {
2340 /*
2341 * remove the space for the drain we added
2342 * so that we don't add it again
2343 */
2344 --rq->nr_phys_segments;
2345 }
2346
2347 rq = NULL;
2348 break;
2349 } else if (ret == BLKPREP_KILL) {
2350 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002351 /*
2352 * Mark this request as started so we don't trigger
2353 * any debug logic in the end I/O path.
2354 */
2355 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002356 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002357 } else {
2358 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2359 break;
2360 }
2361 }
2362
2363 return rq;
2364}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002365EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002366
Tejun Heo9934c8c2009-05-08 11:54:16 +09002367void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002368{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002369 struct request_queue *q = rq->q;
2370
Tejun Heo158dbda2009-04-23 11:05:18 +09002371 BUG_ON(list_empty(&rq->queuelist));
2372 BUG_ON(ELV_ON_HASH(rq));
2373
2374 list_del_init(&rq->queuelist);
2375
2376 /*
2377 * the time frame between a request being removed from the lists
2378 * and to it is freed is accounted as io that is in progress at
2379 * the driver side.
2380 */
Divyesh Shah91952912010-04-01 15:01:41 -07002381 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002382 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002383 set_io_start_time_ns(rq);
2384 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002385}
2386
Tejun Heo5efccd12009-04-23 11:05:18 +09002387/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002388 * blk_start_request - start request processing on the driver
2389 * @req: request to dequeue
2390 *
2391 * Description:
2392 * Dequeue @req and start timeout timer on it. This hands off the
2393 * request to the driver.
2394 *
2395 * Block internal functions which don't want to start timer should
2396 * call blk_dequeue_request().
2397 *
2398 * Context:
2399 * queue_lock must be held.
2400 */
2401void blk_start_request(struct request *req)
2402{
2403 blk_dequeue_request(req);
2404
2405 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002406 * We are now handing the request to the hardware, initialize
2407 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002408 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002409 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002410 if (unlikely(blk_bidi_rq(req)))
2411 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2412
Jeff Moyer4912aa62013-10-08 14:36:41 -04002413 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002414 blk_add_timer(req);
2415}
2416EXPORT_SYMBOL(blk_start_request);
2417
2418/**
2419 * blk_fetch_request - fetch a request from a request queue
2420 * @q: request queue to fetch a request from
2421 *
2422 * Description:
2423 * Return the request at the top of @q. The request is started on
2424 * return and LLD can start processing it immediately.
2425 *
2426 * Return:
2427 * Pointer to the request at the top of @q if available. Null
2428 * otherwise.
2429 *
2430 * Context:
2431 * queue_lock must be held.
2432 */
2433struct request *blk_fetch_request(struct request_queue *q)
2434{
2435 struct request *rq;
2436
2437 rq = blk_peek_request(q);
2438 if (rq)
2439 blk_start_request(rq);
2440 return rq;
2441}
2442EXPORT_SYMBOL(blk_fetch_request);
2443
2444/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002445 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002446 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002447 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002448 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002449 *
2450 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002451 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2452 * the request structure even if @req doesn't have leftover.
2453 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002454 *
2455 * This special helper function is only for request stacking drivers
2456 * (e.g. request-based dm) so that they can handle partial completion.
2457 * Actual device drivers should use blk_end_request instead.
2458 *
2459 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2460 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002461 *
2462 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002463 * %false - this request doesn't have any more data
2464 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002465 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002466bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002468 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002470 trace_block_rq_complete(req->q, req, nr_bytes);
2471
Tejun Heo2e60e022009-04-23 11:05:18 +09002472 if (!req->bio)
2473 return false;
2474
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002476 * For fs requests, rq is just carrier of independent bio's
2477 * and each partial completion should be handled separately.
2478 * Reset per-request error on each partial completion.
2479 *
2480 * TODO: tj: This is too subtle. It would be better to let
2481 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002483 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 req->errors = 0;
2485
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002486 if (error && req->cmd_type == REQ_TYPE_FS &&
2487 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002488 char *error_type;
2489
2490 switch (error) {
2491 case -ENOLINK:
2492 error_type = "recoverable transport";
2493 break;
2494 case -EREMOTEIO:
2495 error_type = "critical target";
2496 break;
2497 case -EBADE:
2498 error_type = "critical nexus";
2499 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002500 case -ETIMEDOUT:
2501 error_type = "timeout";
2502 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002503 case -ENOSPC:
2504 error_type = "critical space allocation";
2505 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002506 case -ENODATA:
2507 error_type = "critical medium";
2508 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002509 case -EIO:
2510 default:
2511 error_type = "I/O";
2512 break;
2513 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002514 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2515 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002516 req->rq_disk->disk_name : "?",
2517 (unsigned long long)blk_rq_pos(req));
2518
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 }
2520
Jens Axboebc58ba92009-01-23 10:54:44 +01002521 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002522
Kent Overstreetf79ea412012-09-20 16:38:30 -07002523 total_bytes = 0;
2524 while (req->bio) {
2525 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002526 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
Kent Overstreet4f024f32013-10-11 15:44:27 -07002528 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Kent Overstreetf79ea412012-09-20 16:38:30 -07002531 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Kent Overstreetf79ea412012-09-20 16:38:30 -07002533 total_bytes += bio_bytes;
2534 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
Kent Overstreetf79ea412012-09-20 16:38:30 -07002536 if (!nr_bytes)
2537 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 }
2539
2540 /*
2541 * completely done
2542 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002543 if (!req->bio) {
2544 /*
2545 * Reset counters so that the request stacking driver
2546 * can find how many bytes remain in the request
2547 * later.
2548 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002549 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002550 return false;
2551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002553 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002554
2555 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002556 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002557 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002558
Tejun Heo80a761f2009-07-03 17:48:17 +09002559 /* mixed attributes always follow the first bio */
2560 if (req->cmd_flags & REQ_MIXED_MERGE) {
2561 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2562 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2563 }
2564
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002565 /*
2566 * If total number of sectors is less than the first segment
2567 * size, something has gone terribly wrong.
2568 */
2569 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002570 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002571 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002572 }
2573
2574 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002576
Tejun Heo2e60e022009-04-23 11:05:18 +09002577 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578}
Tejun Heo2e60e022009-04-23 11:05:18 +09002579EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Tejun Heo2e60e022009-04-23 11:05:18 +09002581static bool blk_update_bidi_request(struct request *rq, int error,
2582 unsigned int nr_bytes,
2583 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002584{
Tejun Heo2e60e022009-04-23 11:05:18 +09002585 if (blk_update_request(rq, error, nr_bytes))
2586 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002587
Tejun Heo2e60e022009-04-23 11:05:18 +09002588 /* Bidi request must be completed as a whole */
2589 if (unlikely(blk_bidi_rq(rq)) &&
2590 blk_update_request(rq->next_rq, error, bidi_bytes))
2591 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002592
Jens Axboee2e1a142010-06-09 10:42:09 +02002593 if (blk_queue_add_random(rq->q))
2594 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002595
2596 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597}
2598
James Bottomley28018c22010-07-01 19:49:17 +09002599/**
2600 * blk_unprep_request - unprepare a request
2601 * @req: the request
2602 *
2603 * This function makes a request ready for complete resubmission (or
2604 * completion). It happens only after all error handling is complete,
2605 * so represents the appropriate moment to deallocate any resources
2606 * that were allocated to the request in the prep_rq_fn. The queue
2607 * lock is held when calling this.
2608 */
2609void blk_unprep_request(struct request *req)
2610{
2611 struct request_queue *q = req->q;
2612
2613 req->cmd_flags &= ~REQ_DONTPREP;
2614 if (q->unprep_rq_fn)
2615 q->unprep_rq_fn(q, req);
2616}
2617EXPORT_SYMBOL_GPL(blk_unprep_request);
2618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619/*
2620 * queue lock must be held
2621 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002622void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002624 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002625 blk_queue_end_tag(req->q, req);
2626
James Bottomleyba396a62009-05-27 14:17:08 +02002627 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002629 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002630 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
Mike Andersone78042e2008-10-30 02:16:20 -07002632 blk_delete_timer(req);
2633
James Bottomley28018c22010-07-01 19:49:17 +09002634 if (req->cmd_flags & REQ_DONTPREP)
2635 blk_unprep_request(req);
2636
Jens Axboebc58ba92009-01-23 10:54:44 +01002637 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002640 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002641 else {
2642 if (blk_bidi_rq(req))
2643 __blk_put_request(req->next_rq->q, req->next_rq);
2644
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 }
2647}
Christoph Hellwig12120072014-04-16 09:44:59 +02002648EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002650/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002651 * blk_end_bidi_request - Complete a bidi request
2652 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002653 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002654 * @nr_bytes: number of bytes to complete @rq
2655 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002656 *
2657 * Description:
2658 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002659 * Drivers that supports bidi can safely call this member for any
2660 * type of request, bidi or uni. In the later case @bidi_bytes is
2661 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002662 *
2663 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002664 * %false - we are done with this request
2665 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002666 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002667static bool blk_end_bidi_request(struct request *rq, int error,
2668 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002669{
2670 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002671 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002672
Tejun Heo2e60e022009-04-23 11:05:18 +09002673 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2674 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002675
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002676 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002677 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002678 spin_unlock_irqrestore(q->queue_lock, flags);
2679
Tejun Heo2e60e022009-04-23 11:05:18 +09002680 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002681}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002682
2683/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002684 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2685 * @rq: the request to complete
2686 * @error: %0 for success, < %0 for error
2687 * @nr_bytes: number of bytes to complete @rq
2688 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002689 *
2690 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002691 * Identical to blk_end_bidi_request() except that queue lock is
2692 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002693 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002694 * Return:
2695 * %false - we are done with this request
2696 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002697 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002698bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002699 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002700{
Tejun Heo2e60e022009-04-23 11:05:18 +09002701 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2702 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002703
Tejun Heo2e60e022009-04-23 11:05:18 +09002704 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002705
Tejun Heo2e60e022009-04-23 11:05:18 +09002706 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002707}
2708
2709/**
2710 * blk_end_request - Helper function for drivers to complete the request.
2711 * @rq: the request being processed
2712 * @error: %0 for success, < %0 for error
2713 * @nr_bytes: number of bytes to complete
2714 *
2715 * Description:
2716 * Ends I/O on a number of bytes attached to @rq.
2717 * If @rq has leftover, sets it up for the next range of segments.
2718 *
2719 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002720 * %false - we are done with this request
2721 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002722 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002723bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002724{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002725 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002726}
Jens Axboe56ad1742009-07-28 22:11:24 +02002727EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002728
2729/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002730 * blk_end_request_all - Helper function for drives to finish the request.
2731 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002732 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002733 *
2734 * Description:
2735 * Completely finish @rq.
2736 */
2737void blk_end_request_all(struct request *rq, int error)
2738{
2739 bool pending;
2740 unsigned int bidi_bytes = 0;
2741
2742 if (unlikely(blk_bidi_rq(rq)))
2743 bidi_bytes = blk_rq_bytes(rq->next_rq);
2744
2745 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2746 BUG_ON(pending);
2747}
Jens Axboe56ad1742009-07-28 22:11:24 +02002748EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002749
2750/**
2751 * blk_end_request_cur - Helper function to finish the current request chunk.
2752 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002753 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002754 *
2755 * Description:
2756 * Complete the current consecutively mapped chunk from @rq.
2757 *
2758 * Return:
2759 * %false - we are done with this request
2760 * %true - still buffers pending for this request
2761 */
2762bool blk_end_request_cur(struct request *rq, int error)
2763{
2764 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2765}
Jens Axboe56ad1742009-07-28 22:11:24 +02002766EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002767
2768/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002769 * blk_end_request_err - Finish a request till the next failure boundary.
2770 * @rq: the request to finish till the next failure boundary for
2771 * @error: must be negative errno
2772 *
2773 * Description:
2774 * Complete @rq till the next failure boundary.
2775 *
2776 * Return:
2777 * %false - we are done with this request
2778 * %true - still buffers pending for this request
2779 */
2780bool blk_end_request_err(struct request *rq, int error)
2781{
2782 WARN_ON(error >= 0);
2783 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2784}
2785EXPORT_SYMBOL_GPL(blk_end_request_err);
2786
2787/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002788 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002789 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002790 * @error: %0 for success, < %0 for error
2791 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002792 *
2793 * Description:
2794 * Must be called with queue lock held unlike blk_end_request().
2795 *
2796 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002797 * %false - we are done with this request
2798 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002799 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002800bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002801{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002802 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002803}
Jens Axboe56ad1742009-07-28 22:11:24 +02002804EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002805
2806/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002807 * __blk_end_request_all - Helper function for drives to finish the request.
2808 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002809 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002810 *
2811 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002812 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002813 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002814void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002815{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002816 bool pending;
2817 unsigned int bidi_bytes = 0;
2818
2819 if (unlikely(blk_bidi_rq(rq)))
2820 bidi_bytes = blk_rq_bytes(rq->next_rq);
2821
2822 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2823 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002824}
Jens Axboe56ad1742009-07-28 22:11:24 +02002825EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002826
2827/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002828 * __blk_end_request_cur - Helper function to finish the current request chunk.
2829 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002830 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002831 *
2832 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002833 * Complete the current consecutively mapped chunk from @rq. Must
2834 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002835 *
2836 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002837 * %false - we are done with this request
2838 * %true - still buffers pending for this request
2839 */
2840bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002841{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002842 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002843}
Jens Axboe56ad1742009-07-28 22:11:24 +02002844EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002845
Tejun Heo80a761f2009-07-03 17:48:17 +09002846/**
2847 * __blk_end_request_err - Finish a request till the next failure boundary.
2848 * @rq: the request to finish till the next failure boundary for
2849 * @error: must be negative errno
2850 *
2851 * Description:
2852 * Complete @rq till the next failure boundary. Must be called
2853 * with queue lock held.
2854 *
2855 * Return:
2856 * %false - we are done with this request
2857 * %true - still buffers pending for this request
2858 */
2859bool __blk_end_request_err(struct request *rq, int error)
2860{
2861 WARN_ON(error >= 0);
2862 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2863}
2864EXPORT_SYMBOL_GPL(__blk_end_request_err);
2865
Jens Axboe86db1e22008-01-29 14:53:40 +01002866void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2867 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002869 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002870 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871
Jens Axboeb4f42e22014-04-10 09:46:28 -06002872 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002873 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002874
Kent Overstreet4f024f32013-10-11 15:44:27 -07002875 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877
NeilBrown66846572007-08-16 13:31:28 +02002878 if (bio->bi_bdev)
2879 rq->rq_disk = bio->bi_bdev->bd_disk;
2880}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002882#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2883/**
2884 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2885 * @rq: the request to be flushed
2886 *
2887 * Description:
2888 * Flush all pages in @rq.
2889 */
2890void rq_flush_dcache_pages(struct request *rq)
2891{
2892 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002893 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002894
2895 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002896 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002897}
2898EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2899#endif
2900
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002901/**
2902 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2903 * @q : the queue of the device being checked
2904 *
2905 * Description:
2906 * Check if underlying low-level drivers of a device are busy.
2907 * If the drivers want to export their busy state, they must set own
2908 * exporting function using blk_queue_lld_busy() first.
2909 *
2910 * Basically, this function is used only by request stacking drivers
2911 * to stop dispatching requests to underlying devices when underlying
2912 * devices are busy. This behavior helps more I/O merging on the queue
2913 * of the request stacking driver and prevents I/O throughput regression
2914 * on burst I/O load.
2915 *
2916 * Return:
2917 * 0 - Not busy (The request stacking driver should dispatch request)
2918 * 1 - Busy (The request stacking driver should stop dispatching request)
2919 */
2920int blk_lld_busy(struct request_queue *q)
2921{
2922 if (q->lld_busy_fn)
2923 return q->lld_busy_fn(q);
2924
2925 return 0;
2926}
2927EXPORT_SYMBOL_GPL(blk_lld_busy);
2928
Mike Snitzer78d8e582015-06-26 10:01:13 -04002929/**
2930 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2931 * @rq: the clone request to be cleaned up
2932 *
2933 * Description:
2934 * Free all bios in @rq for a cloned request.
2935 */
2936void blk_rq_unprep_clone(struct request *rq)
2937{
2938 struct bio *bio;
2939
2940 while ((bio = rq->bio) != NULL) {
2941 rq->bio = bio->bi_next;
2942
2943 bio_put(bio);
2944 }
2945}
2946EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2947
2948/*
2949 * Copy attributes of the original request to the clone request.
2950 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2951 */
2952static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002953{
2954 dst->cpu = src->cpu;
Mike Snitzer78d8e582015-06-26 10:01:13 -04002955 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02002956 dst->cmd_type = src->cmd_type;
2957 dst->__sector = blk_rq_pos(src);
2958 dst->__data_len = blk_rq_bytes(src);
2959 dst->nr_phys_segments = src->nr_phys_segments;
2960 dst->ioprio = src->ioprio;
2961 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04002962}
2963
2964/**
2965 * blk_rq_prep_clone - Helper function to setup clone request
2966 * @rq: the request to be setup
2967 * @rq_src: original request to be cloned
2968 * @bs: bio_set that bios for clone are allocated from
2969 * @gfp_mask: memory allocation mask for bio
2970 * @bio_ctr: setup function to be called for each clone bio.
2971 * Returns %0 for success, non %0 for failure.
2972 * @data: private data to be passed to @bio_ctr
2973 *
2974 * Description:
2975 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2976 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2977 * are not copied, and copying such parts is the caller's responsibility.
2978 * Also, pages which the original bios are pointing to are not copied
2979 * and the cloned bios just point same pages.
2980 * So cloned bios must be completed before original bios, which means
2981 * the caller must complete @rq before @rq_src.
2982 */
2983int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2984 struct bio_set *bs, gfp_t gfp_mask,
2985 int (*bio_ctr)(struct bio *, struct bio *, void *),
2986 void *data)
2987{
2988 struct bio *bio, *bio_src;
2989
2990 if (!bs)
2991 bs = fs_bio_set;
2992
2993 __rq_for_each_bio(bio_src, rq_src) {
2994 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2995 if (!bio)
2996 goto free_and_out;
2997
2998 if (bio_ctr && bio_ctr(bio, bio_src, data))
2999 goto free_and_out;
3000
3001 if (rq->bio) {
3002 rq->biotail->bi_next = bio;
3003 rq->biotail = bio;
3004 } else
3005 rq->bio = rq->biotail = bio;
3006 }
3007
3008 __blk_rq_prep_clone(rq, rq_src);
3009
3010 return 0;
3011
3012free_and_out:
3013 if (bio)
3014 bio_put(bio);
3015 blk_rq_unprep_clone(rq);
3016
3017 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003018}
3019EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3020
Jens Axboe59c3d452014-04-08 09:15:35 -06003021int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022{
3023 return queue_work(kblockd_workqueue, work);
3024}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025EXPORT_SYMBOL(kblockd_schedule_work);
3026
Jens Axboe59c3d452014-04-08 09:15:35 -06003027int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3028 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003029{
3030 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3031}
3032EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3033
Jens Axboe8ab14592014-04-08 09:17:40 -06003034int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3035 unsigned long delay)
3036{
3037 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3038}
3039EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3040
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003041/**
3042 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3043 * @plug: The &struct blk_plug that needs to be initialized
3044 *
3045 * Description:
3046 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3047 * pending I/O should the task end up blocking between blk_start_plug() and
3048 * blk_finish_plug(). This is important from a performance perspective, but
3049 * also ensures that we don't deadlock. For instance, if the task is blocking
3050 * for a memory allocation, memory reclaim could end up wanting to free a
3051 * page belonging to that request that is currently residing in our private
3052 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3053 * this kind of deadlock.
3054 */
Jens Axboe73c10102011-03-08 13:19:51 +01003055void blk_start_plug(struct blk_plug *plug)
3056{
3057 struct task_struct *tsk = current;
3058
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003059 /*
3060 * If this is a nested plug, don't actually assign it.
3061 */
3062 if (tsk->plug)
3063 return;
3064
Jens Axboe73c10102011-03-08 13:19:51 +01003065 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003066 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003067 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003068 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003069 * Store ordering should not be needed here, since a potential
3070 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01003071 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003072 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01003073}
3074EXPORT_SYMBOL(blk_start_plug);
3075
3076static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3077{
3078 struct request *rqa = container_of(a, struct request, queuelist);
3079 struct request *rqb = container_of(b, struct request, queuelist);
3080
Jianpeng Ma975927b2012-10-25 21:58:17 +02003081 return !(rqa->q < rqb->q ||
3082 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003083}
3084
Jens Axboe49cac012011-04-16 13:51:05 +02003085/*
3086 * If 'from_schedule' is true, then postpone the dispatch of requests
3087 * until a safe kblockd context. We due this to avoid accidental big
3088 * additional stack usage in driver dispatch, in places where the originally
3089 * plugger did not intend it.
3090 */
Jens Axboef6603782011-04-15 15:49:07 +02003091static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003092 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003093 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003094{
Jens Axboe49cac012011-04-16 13:51:05 +02003095 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003096
Bart Van Assche70460572012-11-28 13:45:56 +01003097 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003098 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003099 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003100 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003101 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003102}
3103
NeilBrown74018dc2012-07-31 09:08:15 +02003104static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003105{
3106 LIST_HEAD(callbacks);
3107
Shaohua Li2a7d5552012-07-31 09:08:15 +02003108 while (!list_empty(&plug->cb_list)) {
3109 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003110
Shaohua Li2a7d5552012-07-31 09:08:15 +02003111 while (!list_empty(&callbacks)) {
3112 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003113 struct blk_plug_cb,
3114 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003115 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003116 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003117 }
NeilBrown048c9372011-04-18 09:52:22 +02003118 }
3119}
3120
NeilBrown9cbb1752012-07-31 09:08:14 +02003121struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3122 int size)
3123{
3124 struct blk_plug *plug = current->plug;
3125 struct blk_plug_cb *cb;
3126
3127 if (!plug)
3128 return NULL;
3129
3130 list_for_each_entry(cb, &plug->cb_list, list)
3131 if (cb->callback == unplug && cb->data == data)
3132 return cb;
3133
3134 /* Not currently on the callback list */
3135 BUG_ON(size < sizeof(*cb));
3136 cb = kzalloc(size, GFP_ATOMIC);
3137 if (cb) {
3138 cb->data = data;
3139 cb->callback = unplug;
3140 list_add(&cb->list, &plug->cb_list);
3141 }
3142 return cb;
3143}
3144EXPORT_SYMBOL(blk_check_plugged);
3145
Jens Axboe49cac012011-04-16 13:51:05 +02003146void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003147{
3148 struct request_queue *q;
3149 unsigned long flags;
3150 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003151 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003152 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003153
NeilBrown74018dc2012-07-31 09:08:15 +02003154 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003155
3156 if (!list_empty(&plug->mq_list))
3157 blk_mq_flush_plug_list(plug, from_schedule);
3158
Jens Axboe73c10102011-03-08 13:19:51 +01003159 if (list_empty(&plug->list))
3160 return;
3161
NeilBrown109b8122011-04-11 14:13:10 +02003162 list_splice_init(&plug->list, &list);
3163
Jianpeng Ma422765c2013-01-11 14:46:09 +01003164 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003165
3166 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003167 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003168
3169 /*
3170 * Save and disable interrupts here, to avoid doing it for every
3171 * queue lock we have to take.
3172 */
Jens Axboe73c10102011-03-08 13:19:51 +01003173 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003174 while (!list_empty(&list)) {
3175 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003176 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003177 BUG_ON(!rq->q);
3178 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003179 /*
3180 * This drops the queue lock
3181 */
3182 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003183 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003184 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003185 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003186 spin_lock(q->queue_lock);
3187 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003188
3189 /*
3190 * Short-circuit if @q is dead
3191 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003192 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003193 __blk_end_request_all(rq, -ENODEV);
3194 continue;
3195 }
3196
Jens Axboe73c10102011-03-08 13:19:51 +01003197 /*
3198 * rq is already accounted, so use raw insert
3199 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003200 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3201 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3202 else
3203 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003204
3205 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003206 }
3207
Jens Axboe99e22592011-04-18 09:59:55 +02003208 /*
3209 * This drops the queue lock
3210 */
3211 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003212 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003213
Jens Axboe73c10102011-03-08 13:19:51 +01003214 local_irq_restore(flags);
3215}
Jens Axboe73c10102011-03-08 13:19:51 +01003216
3217void blk_finish_plug(struct blk_plug *plug)
3218{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003219 if (plug != current->plug)
3220 return;
Jens Axboef6603782011-04-15 15:49:07 +02003221 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003222
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003223 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003224}
3225EXPORT_SYMBOL(blk_finish_plug);
3226
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003227#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003228/**
3229 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3230 * @q: the queue of the device
3231 * @dev: the device the queue belongs to
3232 *
3233 * Description:
3234 * Initialize runtime-PM-related fields for @q and start auto suspend for
3235 * @dev. Drivers that want to take advantage of request-based runtime PM
3236 * should call this function after @dev has been initialized, and its
3237 * request queue @q has been allocated, and runtime PM for it can not happen
3238 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3239 * cases, driver should call this function before any I/O has taken place.
3240 *
3241 * This function takes care of setting up using auto suspend for the device,
3242 * the autosuspend delay is set to -1 to make runtime suspend impossible
3243 * until an updated value is either set by user or by driver. Drivers do
3244 * not need to touch other autosuspend settings.
3245 *
3246 * The block layer runtime PM is request based, so only works for drivers
3247 * that use request as their IO unit instead of those directly use bio's.
3248 */
3249void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3250{
3251 q->dev = dev;
3252 q->rpm_status = RPM_ACTIVE;
3253 pm_runtime_set_autosuspend_delay(q->dev, -1);
3254 pm_runtime_use_autosuspend(q->dev);
3255}
3256EXPORT_SYMBOL(blk_pm_runtime_init);
3257
3258/**
3259 * blk_pre_runtime_suspend - Pre runtime suspend check
3260 * @q: the queue of the device
3261 *
3262 * Description:
3263 * This function will check if runtime suspend is allowed for the device
3264 * by examining if there are any requests pending in the queue. If there
3265 * are requests pending, the device can not be runtime suspended; otherwise,
3266 * the queue's status will be updated to SUSPENDING and the driver can
3267 * proceed to suspend the device.
3268 *
3269 * For the not allowed case, we mark last busy for the device so that
3270 * runtime PM core will try to autosuspend it some time later.
3271 *
3272 * This function should be called near the start of the device's
3273 * runtime_suspend callback.
3274 *
3275 * Return:
3276 * 0 - OK to runtime suspend the device
3277 * -EBUSY - Device should not be runtime suspended
3278 */
3279int blk_pre_runtime_suspend(struct request_queue *q)
3280{
3281 int ret = 0;
3282
3283 spin_lock_irq(q->queue_lock);
3284 if (q->nr_pending) {
3285 ret = -EBUSY;
3286 pm_runtime_mark_last_busy(q->dev);
3287 } else {
3288 q->rpm_status = RPM_SUSPENDING;
3289 }
3290 spin_unlock_irq(q->queue_lock);
3291 return ret;
3292}
3293EXPORT_SYMBOL(blk_pre_runtime_suspend);
3294
3295/**
3296 * blk_post_runtime_suspend - Post runtime suspend processing
3297 * @q: the queue of the device
3298 * @err: return value of the device's runtime_suspend function
3299 *
3300 * Description:
3301 * Update the queue's runtime status according to the return value of the
3302 * device's runtime suspend function and mark last busy for the device so
3303 * that PM core will try to auto suspend the device at a later time.
3304 *
3305 * This function should be called near the end of the device's
3306 * runtime_suspend callback.
3307 */
3308void blk_post_runtime_suspend(struct request_queue *q, int err)
3309{
3310 spin_lock_irq(q->queue_lock);
3311 if (!err) {
3312 q->rpm_status = RPM_SUSPENDED;
3313 } else {
3314 q->rpm_status = RPM_ACTIVE;
3315 pm_runtime_mark_last_busy(q->dev);
3316 }
3317 spin_unlock_irq(q->queue_lock);
3318}
3319EXPORT_SYMBOL(blk_post_runtime_suspend);
3320
3321/**
3322 * blk_pre_runtime_resume - Pre runtime resume processing
3323 * @q: the queue of the device
3324 *
3325 * Description:
3326 * Update the queue's runtime status to RESUMING in preparation for the
3327 * runtime resume of the device.
3328 *
3329 * This function should be called near the start of the device's
3330 * runtime_resume callback.
3331 */
3332void blk_pre_runtime_resume(struct request_queue *q)
3333{
3334 spin_lock_irq(q->queue_lock);
3335 q->rpm_status = RPM_RESUMING;
3336 spin_unlock_irq(q->queue_lock);
3337}
3338EXPORT_SYMBOL(blk_pre_runtime_resume);
3339
3340/**
3341 * blk_post_runtime_resume - Post runtime resume processing
3342 * @q: the queue of the device
3343 * @err: return value of the device's runtime_resume function
3344 *
3345 * Description:
3346 * Update the queue's runtime status according to the return value of the
3347 * device's runtime_resume function. If it is successfully resumed, process
3348 * the requests that are queued into the device's queue when it is resuming
3349 * and then mark last busy and initiate autosuspend for it.
3350 *
3351 * This function should be called near the end of the device's
3352 * runtime_resume callback.
3353 */
3354void blk_post_runtime_resume(struct request_queue *q, int err)
3355{
3356 spin_lock_irq(q->queue_lock);
3357 if (!err) {
3358 q->rpm_status = RPM_ACTIVE;
3359 __blk_run_queue(q);
3360 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003361 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003362 } else {
3363 q->rpm_status = RPM_SUSPENDED;
3364 }
3365 spin_unlock_irq(q->queue_lock);
3366}
3367EXPORT_SYMBOL(blk_post_runtime_resume);
3368#endif
3369
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370int __init blk_dev_init(void)
3371{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003372 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05303373 FIELD_SIZEOF(struct request, cmd_flags));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003374
Tejun Heo89b90be2011-01-03 15:01:47 +01003375 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3376 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003377 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 if (!kblockd_workqueue)
3379 panic("Failed to create kblockd\n");
3380
3381 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003382 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
Jens Axboe8324aa92008-01-29 14:51:59 +01003384 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003385 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 return 0;
3388}