blob: 5131993b23a1a2b35670adcbd4ab7d569a29a3d2 [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 */
Dan Williams3ef28e82015-10-21 13:20:12 -0400557 blk_freeze_queue(q);
558 spin_lock_irq(lock);
559 if (!q->mq_ops)
Ming Lei43a5e4e2013-12-26 21:31:35 +0800560 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100561 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100562 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200563
Dan Williams5a48fc12015-10-21 13:20:23 -0400564 /* for synchronous bio-based driver finish in-flight integrity i/o */
565 blk_flush_integrity();
566
Tejun Heoc9a929d2011-10-19 14:42:16 +0200567 /* @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);
Dan Williams3ef28e82015-10-21 13:20:12 -0400573 percpu_ref_exit(&q->q_usage_counter);
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100574
Asias He5e5cfac2012-05-24 23:28:52 +0800575 spin_lock_irq(lock);
576 if (q->queue_lock != &q->__queue_lock)
577 q->queue_lock = &q->__queue_lock;
578 spin_unlock_irq(lock);
579
Tejun Heob02176f2015-09-08 12:20:22 -0400580 bdi_unregister(&q->backing_dev_info);
NeilBrown6cd18e72015-04-27 14:12:22 +1000581
Tejun Heoc9a929d2011-10-19 14:42:16 +0200582 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500583 blk_put_queue(q);
584}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585EXPORT_SYMBOL(blk_cleanup_queue);
586
David Rientjes271508d2015-03-24 16:21:16 -0700587/* Allocate memory local to the request queue */
588static void *alloc_request_struct(gfp_t gfp_mask, void *data)
589{
590 int nid = (int)(long)data;
591 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
592}
593
594static void free_request_struct(void *element, void *unused)
595{
596 kmem_cache_free(request_cachep, element);
597}
598
Tejun Heo5b788ce2012-06-04 20:40:59 -0700599int blk_init_rl(struct request_list *rl, struct request_queue *q,
600 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400602 if (unlikely(rl->rq_pool))
603 return 0;
604
Tejun Heo5b788ce2012-06-04 20:40:59 -0700605 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200606 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
607 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200608 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
609 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
David Rientjes271508d2015-03-24 16:21:16 -0700611 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
612 free_request_struct,
613 (void *)(long)q->node, gfp_mask,
614 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (!rl->rq_pool)
616 return -ENOMEM;
617
618 return 0;
619}
620
Tejun Heo5b788ce2012-06-04 20:40:59 -0700621void blk_exit_rl(struct request_list *rl)
622{
623 if (rl->rq_pool)
624 mempool_destroy(rl->rq_pool);
625}
626
Jens Axboe165125e2007-07-24 09:28:11 +0200627struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100629 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700630}
631EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
Dan Williams3ef28e82015-10-21 13:20:12 -0400633int blk_queue_enter(struct request_queue *q, gfp_t gfp)
634{
635 while (true) {
636 int ret;
637
638 if (percpu_ref_tryget_live(&q->q_usage_counter))
639 return 0;
640
Mel Gorman71baba42015-11-06 16:28:28 -0800641 if (!gfpflags_allow_blocking(gfp))
Dan Williams3ef28e82015-10-21 13:20:12 -0400642 return -EBUSY;
643
644 ret = wait_event_interruptible(q->mq_freeze_wq,
645 !atomic_read(&q->mq_freeze_depth) ||
646 blk_queue_dying(q));
647 if (blk_queue_dying(q))
648 return -ENODEV;
649 if (ret)
650 return ret;
651 }
652}
653
654void blk_queue_exit(struct request_queue *q)
655{
656 percpu_ref_put(&q->q_usage_counter);
657}
658
659static void blk_queue_usage_counter_release(struct percpu_ref *ref)
660{
661 struct request_queue *q =
662 container_of(ref, struct request_queue, q_usage_counter);
663
664 wake_up_all(&q->mq_freeze_wq);
665}
666
Jens Axboe165125e2007-07-24 09:28:11 +0200667struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700668{
Jens Axboe165125e2007-07-24 09:28:11 +0200669 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700670 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700671
Jens Axboe8324aa92008-01-29 14:51:59 +0100672 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700673 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (!q)
675 return NULL;
676
Dan Carpenter00380a42012-03-23 09:58:54 +0100677 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100678 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800679 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100680
Kent Overstreet54efd502015-04-23 22:37:18 -0700681 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
682 if (!q->bio_split)
683 goto fail_id;
684
Jens Axboe0989a022009-06-12 14:42:56 +0200685 q->backing_dev_info.ra_pages =
686 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
Tejun Heo89e9b9e2015-05-22 17:13:36 -0400687 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
Jens Axboed9938312009-06-12 14:45:52 +0200688 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100689 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200690
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700691 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100692 if (err)
Kent Overstreet54efd502015-04-23 22:37:18 -0700693 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700694
Matthew Garrett31373d02010-04-06 14:25:14 +0200695 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
696 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700697 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100698 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700699 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100700 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800701#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800702 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800703#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500704 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500705
Jens Axboe8324aa92008-01-29 14:51:59 +0100706 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Al Viro483f4af2006-03-18 18:34:37 -0500708 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700709 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500710
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500711 /*
712 * By default initialize queue_lock to internal lock and driver can
713 * override it later if need be.
714 */
715 q->queue_lock = &q->__queue_lock;
716
Tejun Heob82d4b12012-04-13 13:11:31 -0700717 /*
718 * A queue starts its life with bypass turned on to avoid
719 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700720 * init. The initial bypass will be finished when the queue is
721 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700722 */
723 q->bypass_depth = 1;
724 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
725
Jens Axboe320ae512013-10-24 09:20:05 +0100726 init_waitqueue_head(&q->mq_freeze_wq);
727
Dan Williams3ef28e82015-10-21 13:20:12 -0400728 /*
729 * Init percpu_ref in atomic mode so that it's faster to shutdown.
730 * See blk_register_queue() for details.
731 */
732 if (percpu_ref_init(&q->q_usage_counter,
733 blk_queue_usage_counter_release,
734 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
Mikulas Patockafff49962013-10-14 12:11:36 -0400735 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800736
Dan Williams3ef28e82015-10-21 13:20:12 -0400737 if (blkcg_init_queue(q))
738 goto fail_ref;
739
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100741
Dan Williams3ef28e82015-10-21 13:20:12 -0400742fail_ref:
743 percpu_ref_exit(&q->q_usage_counter);
Mikulas Patockafff49962013-10-14 12:11:36 -0400744fail_bdi:
745 bdi_destroy(&q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700746fail_split:
747 bioset_free(q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100748fail_id:
749 ida_simple_remove(&blk_queue_ida, q->id);
750fail_q:
751 kmem_cache_free(blk_requestq_cachep, q);
752 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753}
Christoph Lameter19460892005-06-23 00:08:19 -0700754EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756/**
757 * blk_init_queue - prepare a request queue for use with a block device
758 * @rfn: The function to be called to process requests that have been
759 * placed on the queue.
760 * @lock: Request queue spin lock
761 *
762 * Description:
763 * If a block device wishes to use the standard request handling procedures,
764 * which sorts requests and coalesces adjacent requests, then it must
765 * call blk_init_queue(). The function @rfn will be called when there
766 * are requests on the queue that need to be processed. If the device
767 * supports plugging, then @rfn may not be called immediately when requests
768 * are available on the queue, but may be called at some time later instead.
769 * Plugged queues are generally unplugged when a buffer belonging to one
770 * of the requests on the queue is needed, or due to memory pressure.
771 *
772 * @rfn is not required, or even expected, to remove all requests off the
773 * queue, but only as many as it can handle at a time. If it does leave
774 * requests on the queue, it is responsible for arranging that the requests
775 * get dealt with eventually.
776 *
777 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200778 * request queue; this lock will be taken also from interrupt context, so irq
779 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200781 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * it didn't succeed.
783 *
784 * Note:
785 * blk_init_queue() must be paired with a blk_cleanup_queue() call
786 * when the block device is deactivated (such as at module unload).
787 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700788
Jens Axboe165125e2007-07-24 09:28:11 +0200789struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100791 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700792}
793EXPORT_SYMBOL(blk_init_queue);
794
Jens Axboe165125e2007-07-24 09:28:11 +0200795struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700796blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
797{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600798 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600800 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
801 if (!uninit_q)
802 return NULL;
803
Mike Snitzer51514122011-11-23 10:59:13 +0100804 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600805 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700806 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700807
Mike Snitzer7982e902014-03-08 17:20:01 -0700808 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200809}
810EXPORT_SYMBOL(blk_init_queue_node);
811
Jens Axboedece1632015-11-05 10:41:16 -0700812static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
Mike Snitzer336b7e12015-05-11 14:06:32 -0400813
Mike Snitzer01effb02010-05-11 08:57:42 +0200814struct request_queue *
815blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
816 spinlock_t *lock)
817{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (!q)
819 return NULL;
820
Ming Leif70ced02014-09-25 23:23:47 +0800821 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800822 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700823 return NULL;
824
Tejun Heoa0516612012-06-26 15:05:44 -0700825 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600826 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900830 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700831 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500832
833 /* Override internal queue lock with supplied lock pointer */
834 if (lock)
835 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Jens Axboef3b144a2009-03-06 08:48:33 +0100837 /*
838 * This also sets hw/phys segments, boundary and size
839 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200840 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Alan Stern44ec9542007-02-20 11:01:57 -0500842 q->sg_reserved_size = INT_MAX;
843
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600844 /* Protect q->elevator from elevator_change */
845 mutex_lock(&q->sysfs_lock);
846
Tejun Heob82d4b12012-04-13 13:11:31 -0700847 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600848 if (elevator_init(q, NULL)) {
849 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600850 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600851 }
852
853 mutex_unlock(&q->sysfs_lock);
854
Tejun Heob82d4b12012-04-13 13:11:31 -0700855 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600856
857fail:
Ming Leiba483382014-09-25 23:23:44 +0800858 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600859 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
Mike Snitzer51514122011-11-23 10:59:13 +0100861EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Tejun Heo09ac46c2011-12-14 00:33:38 +0100863bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100865 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100866 __blk_get_queue(q);
867 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 }
869
Tejun Heo09ac46c2011-12-14 00:33:38 +0100870 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871}
Jens Axboed86e0e82011-05-27 07:44:43 +0200872EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Tejun Heo5b788ce2012-06-04 20:40:59 -0700874static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100876 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700877 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100878 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100879 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100880 }
881
Tejun Heo5b788ce2012-06-04 20:40:59 -0700882 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883}
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
886 * ioc_batching returns true if the ioc is a valid batching request and
887 * should be given priority access to a request.
888 */
Jens Axboe165125e2007-07-24 09:28:11 +0200889static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
891 if (!ioc)
892 return 0;
893
894 /*
895 * Make sure the process is able to allocate at least 1 request
896 * even if the batch times out, otherwise we could theoretically
897 * lose wakeups.
898 */
899 return ioc->nr_batch_requests == q->nr_batching ||
900 (ioc->nr_batch_requests > 0
901 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
902}
903
904/*
905 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
906 * will cause the process to be a "batcher" on all queues in the system. This
907 * is the behaviour we want though - once it gets a wakeup it should be given
908 * a nice run.
909 */
Jens Axboe165125e2007-07-24 09:28:11 +0200910static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
912 if (!ioc || ioc_batching(q, ioc))
913 return;
914
915 ioc->nr_batch_requests = q->nr_batching;
916 ioc->last_waited = jiffies;
917}
918
Tejun Heo5b788ce2012-06-04 20:40:59 -0700919static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700921 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Tejun Heod40f75a2015-05-22 17:13:42 -0400923 if (rl->count[sync] < queue_congestion_off_threshold(q))
924 blk_clear_congested(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
Jens Axboe1faa16d2009-04-06 14:48:01 +0200926 if (rl->count[sync] + 1 <= q->nr_requests) {
927 if (waitqueue_active(&rl->wait[sync]))
928 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Tejun Heo5b788ce2012-06-04 20:40:59 -0700930 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 }
932}
933
934/*
935 * A request has just been released. Account for it, update the full and
936 * congestion status, wake up any waiters. Called under q->queue_lock.
937 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700938static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700940 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200941 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700943 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200944 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200945 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700946 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Tejun Heo5b788ce2012-06-04 20:40:59 -0700948 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Jens Axboe1faa16d2009-04-06 14:48:01 +0200950 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700951 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952}
953
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600954int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
955{
956 struct request_list *rl;
Tejun Heod40f75a2015-05-22 17:13:42 -0400957 int on_thresh, off_thresh;
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600958
959 spin_lock_irq(q->queue_lock);
960 q->nr_requests = nr;
961 blk_queue_congestion_threshold(q);
Tejun Heod40f75a2015-05-22 17:13:42 -0400962 on_thresh = queue_congestion_on_threshold(q);
963 off_thresh = queue_congestion_off_threshold(q);
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600964
965 blk_queue_for_each_rl(rl, q) {
Tejun Heod40f75a2015-05-22 17:13:42 -0400966 if (rl->count[BLK_RW_SYNC] >= on_thresh)
967 blk_set_congested(rl, BLK_RW_SYNC);
968 else if (rl->count[BLK_RW_SYNC] < off_thresh)
969 blk_clear_congested(rl, BLK_RW_SYNC);
970
971 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
972 blk_set_congested(rl, BLK_RW_ASYNC);
973 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
974 blk_clear_congested(rl, BLK_RW_ASYNC);
975
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600976 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
977 blk_set_rl_full(rl, BLK_RW_SYNC);
978 } else {
979 blk_clear_rl_full(rl, BLK_RW_SYNC);
980 wake_up(&rl->wait[BLK_RW_SYNC]);
981 }
982
983 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
984 blk_set_rl_full(rl, BLK_RW_ASYNC);
985 } else {
986 blk_clear_rl_full(rl, BLK_RW_ASYNC);
987 wake_up(&rl->wait[BLK_RW_ASYNC]);
988 }
989 }
990
991 spin_unlock_irq(q->queue_lock);
992 return 0;
993}
994
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +0100996 * Determine if elevator data should be initialized when allocating the
997 * request associated with @bio.
998 */
999static bool blk_rq_should_init_elevator(struct bio *bio)
1000{
1001 if (!bio)
1002 return true;
1003
1004 /*
1005 * Flush requests do not use the elevator so skip initialization.
1006 * This allows a request to share the flush and elevator data.
1007 */
1008 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1009 return false;
1010
1011 return true;
1012}
1013
Tejun Heoda8303c2011-10-19 14:33:05 +02001014/**
Tejun Heo852c7882012-03-05 13:15:27 -08001015 * rq_ioc - determine io_context for request allocation
1016 * @bio: request being allocated is for this bio (can be %NULL)
1017 *
1018 * Determine io_context to use for request allocation for @bio. May return
1019 * %NULL if %current->io_context doesn't exist.
1020 */
1021static struct io_context *rq_ioc(struct bio *bio)
1022{
1023#ifdef CONFIG_BLK_CGROUP
1024 if (bio && bio->bi_ioc)
1025 return bio->bi_ioc;
1026#endif
1027 return current->io_context;
1028}
1029
1030/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001031 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -07001032 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +02001033 * @rw_flags: RW and SYNC flags
1034 * @bio: bio to allocate request for (can be %NULL)
1035 * @gfp_mask: allocation mask
1036 *
1037 * Get a free request from @q. This function may fail under memory
1038 * pressure or if @q is dead.
1039 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001040 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001041 * Returns ERR_PTR on failure, with @q->queue_lock held.
1042 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001044static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -07001045 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Tejun Heo5b788ce2012-06-04 20:40:59 -07001047 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -08001048 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001049 struct elevator_type *et = q->elevator->type;
1050 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +01001051 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001052 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +02001053 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001055 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -06001056 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +02001057
Jens Axboe7749a8d2006-12-13 13:02:26 +01001058 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001059 if (may_queue == ELV_MQUEUE_NO)
1060 goto rq_starved;
1061
Jens Axboe1faa16d2009-04-06 14:48:01 +02001062 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1063 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +01001064 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001065 * The queue will fill after this allocation, so set
1066 * it as full, and mark this process as "batching".
1067 * This process will be allowed to complete a batch of
1068 * requests, others will be blocked.
1069 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001070 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001071 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001072 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001073 } else {
1074 if (may_queue != ELV_MQUEUE_MUST
1075 && !ioc_batching(q, ioc)) {
1076 /*
1077 * The queue is full and the allocating
1078 * process is not a "batcher", and not
1079 * exempted by the IO scheduler
1080 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001081 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001082 }
1083 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Tejun Heod40f75a2015-05-22 17:13:42 -04001085 blk_set_congested(rl, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
1087
Jens Axboe082cf692005-06-28 16:35:11 +02001088 /*
1089 * Only allow batching queuers to allocate up to 50% over the defined
1090 * limit of requests, otherwise we could have thousands of requests
1091 * allocated with any setting of ->nr_requests
1092 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001093 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001094 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001095
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001096 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001097 rl->count[is_sync]++;
1098 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001099
Tejun Heof1f8cc92011-12-14 00:33:42 +01001100 /*
1101 * Decide whether the new request will be managed by elevator. If
1102 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1103 * prevent the current elevator from being destroyed until the new
1104 * request is freed. This guarantees icq's won't be destroyed and
1105 * makes creating new ones safe.
1106 *
1107 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1108 * it will be created after releasing queue_lock.
1109 */
Tejun Heod7325802012-03-05 13:14:58 -08001110 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001111 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001112 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001113 if (et->icq_cache && ioc)
1114 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001115 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001116
Jens Axboef253b862010-10-24 22:06:02 +02001117 if (blk_queue_io_stat(q))
1118 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 spin_unlock_irq(q->queue_lock);
1120
Tejun Heo29e2b092012-04-19 16:29:21 -07001121 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001122 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001123 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001124 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Tejun Heo29e2b092012-04-19 16:29:21 -07001126 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001127 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001128 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1129
Tejun Heoaaf7c682012-04-19 16:29:22 -07001130 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001131 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001132 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001133 if (ioc)
1134 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001135 if (!icq)
1136 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001137 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001138
1139 rq->elv.icq = icq;
1140 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1141 goto fail_elvpriv;
1142
1143 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001144 if (icq)
1145 get_io_context(icq->ioc);
1146 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001147out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001148 /*
1149 * ioc may be NULL here, and ioc_batching will be false. That's
1150 * OK, if the queue is under the request limit then requests need
1151 * not count toward the nr_batch_requests limit. There will always
1152 * be some limit enforced by BLK_BATCH_TIME.
1153 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (ioc_batching(q, ioc))
1155 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001156
Jens Axboe1faa16d2009-04-06 14:48:01 +02001157 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001159
Tejun Heoaaf7c682012-04-19 16:29:22 -07001160fail_elvpriv:
1161 /*
1162 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1163 * and may fail indefinitely under memory pressure and thus
1164 * shouldn't stall IO. Treat this request as !elvpriv. This will
1165 * disturb iosched and blkcg but weird is bettern than dead.
1166 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001167 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1168 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001169
1170 rq->cmd_flags &= ~REQ_ELVPRIV;
1171 rq->elv.icq = NULL;
1172
1173 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001174 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001175 spin_unlock_irq(q->queue_lock);
1176 goto out;
1177
Tejun Heob6792812012-03-05 13:15:23 -08001178fail_alloc:
1179 /*
1180 * Allocation failed presumably due to memory. Undo anything we
1181 * might have messed up.
1182 *
1183 * Allocating task should really be put onto the front of the wait
1184 * queue, but this is pretty rare.
1185 */
1186 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001187 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001188
1189 /*
1190 * in the very unlikely event that allocation failed and no
1191 * requests for this direction was pending, mark us starved so that
1192 * freeing of a request in the other direction will notice
1193 * us. another possible fix would be to split the rq mempool into
1194 * READ and WRITE
1195 */
1196rq_starved:
1197 if (unlikely(rl->count[is_sync] == 0))
1198 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001199 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
1201
Tejun Heoda8303c2011-10-19 14:33:05 +02001202/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001203 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001204 * @q: request_queue to allocate request from
1205 * @rw_flags: RW and SYNC flags
1206 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001207 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001208 *
Mel Gormand0164ad2015-11-06 16:28:21 -08001209 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1210 * this function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001211 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001212 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001213 * Returns ERR_PTR on failure, with @q->queue_lock held.
1214 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001216static struct request *get_request(struct request_queue *q, int rw_flags,
1217 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001219 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001220 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001221 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001223
1224 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001225retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001226 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001227 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001228 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Mel Gormand0164ad2015-11-06 16:28:21 -08001230 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001231 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001232 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
Tejun Heoa06e05e2012-06-04 20:40:55 -07001235 /* wait on @rl and retry */
1236 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1237 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001238
Tejun Heoa06e05e2012-06-04 20:40:55 -07001239 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Tejun Heoa06e05e2012-06-04 20:40:55 -07001241 spin_unlock_irq(q->queue_lock);
1242 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Tejun Heoa06e05e2012-06-04 20:40:55 -07001244 /*
1245 * After sleeping, we become a "batching" process and will be able
1246 * to allocate at least one request, and up to a big batch of them
1247 * for a small period time. See ioc_batching, ioc_set_batching
1248 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001249 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Tejun Heoa06e05e2012-06-04 20:40:55 -07001251 spin_lock_irq(q->queue_lock);
1252 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001253
Tejun Heoa06e05e2012-06-04 20:40:55 -07001254 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
1256
Jens Axboe320ae512013-10-24 09:20:05 +01001257static struct request *blk_old_get_request(struct request_queue *q, int rw,
1258 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259{
1260 struct request *rq;
1261
1262 BUG_ON(rw != READ && rw != WRITE);
1263
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001264 /* create ioc upfront */
1265 create_io_context(gfp_mask, q->node);
1266
Nick Piggind6344532005-06-28 20:45:14 -07001267 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001268 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001269 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001270 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001271 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
1273 return rq;
1274}
Jens Axboe320ae512013-10-24 09:20:05 +01001275
1276struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1277{
1278 if (q->mq_ops)
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +02001279 return blk_mq_alloc_request(q, rw, gfp_mask, false);
Jens Axboe320ae512013-10-24 09:20:05 +01001280 else
1281 return blk_old_get_request(q, rw, gfp_mask);
1282}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283EXPORT_SYMBOL(blk_get_request);
1284
1285/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001286 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001287 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001288 * @bio: The bio describing the memory mappings that will be submitted for IO.
1289 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001290 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001291 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001292 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1293 * type commands. Where the struct request needs to be farther initialized by
1294 * the caller. It is passed a &struct bio, which describes the memory info of
1295 * the I/O transfer.
1296 *
1297 * The caller of blk_make_request must make sure that bi_io_vec
1298 * are set to describe the memory buffers. That bio_data_dir() will return
1299 * the needed direction of the request. (And all bio's in the passed bio-chain
1300 * are properly set accordingly)
1301 *
1302 * If called under none-sleepable conditions, mapped bio buffers must not
1303 * need bouncing, by calling the appropriate masked or flagged allocator,
1304 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1305 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001306 *
1307 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
Mel Gormand0164ad2015-11-06 16:28:21 -08001308 * given to how you allocate bios. In particular, you cannot use
1309 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1310 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1311 * thus resulting in a deadlock. Alternatively bios should be allocated using
1312 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
Jens Axboe53674ac2009-05-19 19:52:35 +02001313 * If possible a big IO should be split into smaller parts when allocation
1314 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001315 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001316struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1317 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001318{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001319 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1320
Joe Lawrencea492f072014-08-28 08:15:21 -06001321 if (IS_ERR(rq))
1322 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001323
Jens Axboef27b0872014-06-06 07:57:37 -06001324 blk_rq_set_block_pc(rq);
1325
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001326 for_each_bio(bio) {
1327 struct bio *bounce_bio = bio;
1328 int ret;
1329
1330 blk_queue_bounce(q, &bounce_bio);
1331 ret = blk_rq_append_bio(q, rq, bounce_bio);
1332 if (unlikely(ret)) {
1333 blk_put_request(rq);
1334 return ERR_PTR(ret);
1335 }
1336 }
1337
1338 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001339}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001340EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001341
1342/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001343 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001344 * @rq: request to be initialized
1345 *
1346 */
1347void blk_rq_set_block_pc(struct request *rq)
1348{
1349 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1350 rq->__data_len = 0;
1351 rq->__sector = (sector_t) -1;
1352 rq->bio = rq->biotail = NULL;
1353 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001354}
1355EXPORT_SYMBOL(blk_rq_set_block_pc);
1356
1357/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 * blk_requeue_request - put a request back on queue
1359 * @q: request queue where request should be inserted
1360 * @rq: request to be inserted
1361 *
1362 * Description:
1363 * Drivers often keep queueing requests until the hardware cannot accept
1364 * more, when that condition happens we need to put the request back
1365 * on the queue. Must be called with queue lock held.
1366 */
Jens Axboe165125e2007-07-24 09:28:11 +02001367void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001369 blk_delete_timer(rq);
1370 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001371 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001372
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001373 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 blk_queue_end_tag(q, rq);
1375
James Bottomleyba396a62009-05-27 14:17:08 +02001376 BUG_ON(blk_queued_rq(rq));
1377
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 elv_requeue_request(q, rq);
1379}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380EXPORT_SYMBOL(blk_requeue_request);
1381
Jens Axboe73c10102011-03-08 13:19:51 +01001382static void add_acct_request(struct request_queue *q, struct request *rq,
1383 int where)
1384{
Jens Axboe320ae512013-10-24 09:20:05 +01001385 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001386 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001387}
1388
Tejun Heo074a7ac2008-08-25 19:56:14 +09001389static void part_round_stats_single(int cpu, struct hd_struct *part,
1390 unsigned long now)
1391{
Jens Axboe7276d022014-05-09 15:48:23 -06001392 int inflight;
1393
Tejun Heo074a7ac2008-08-25 19:56:14 +09001394 if (now == part->stamp)
1395 return;
1396
Jens Axboe7276d022014-05-09 15:48:23 -06001397 inflight = part_in_flight(part);
1398 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001399 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001400 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001401 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1402 }
1403 part->stamp = now;
1404}
1405
1406/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001407 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1408 * @cpu: cpu number for stats access
1409 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 *
1411 * The average IO queue length and utilisation statistics are maintained
1412 * by observing the current state of the queue length and the amount of
1413 * time it has been in this state for.
1414 *
1415 * Normally, that accounting is done on IO completion, but that can result
1416 * in more than a second's worth of IO being accounted for within any one
1417 * second, leading to >100% utilisation. To deal with that, we call this
1418 * function to do a round-off before returning the results when reading
1419 * /proc/diskstats. This accounts immediately for all queue usage up to
1420 * the current jiffies and restarts the counters again.
1421 */
Tejun Heoc9959052008-08-25 19:47:21 +09001422void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001423{
1424 unsigned long now = jiffies;
1425
Tejun Heo074a7ac2008-08-25 19:56:14 +09001426 if (part->partno)
1427 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1428 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001429}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001430EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001431
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001432#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001433static void blk_pm_put_request(struct request *rq)
1434{
1435 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1436 pm_runtime_mark_last_busy(rq->q->dev);
1437}
1438#else
1439static inline void blk_pm_put_request(struct request *rq) {}
1440#endif
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442/*
1443 * queue lock must be held
1444 */
Jens Axboe165125e2007-07-24 09:28:11 +02001445void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (unlikely(!q))
1448 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001450 if (q->mq_ops) {
1451 blk_mq_free_request(req);
1452 return;
1453 }
1454
Lin Mingc8158812013-03-23 11:42:27 +08001455 blk_pm_put_request(req);
1456
Tejun Heo8922e162005-10-20 16:23:44 +02001457 elv_completed_request(q, req);
1458
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001459 /* this is a bio leak */
1460 WARN_ON(req->bio != NULL);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 /*
1463 * Request may not have originated from ll_rw_blk. if not,
1464 * it didn't come out of our reserved rq pools
1465 */
Jens Axboe49171e52006-08-10 08:59:11 +02001466 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001467 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001468 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001471 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Tejun Heoa0516612012-06-26 15:05:44 -07001473 blk_free_request(rl, req);
1474 freed_request(rl, flags);
1475 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 }
1477}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001478EXPORT_SYMBOL_GPL(__blk_put_request);
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480void blk_put_request(struct request *req)
1481{
Jens Axboe165125e2007-07-24 09:28:11 +02001482 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483
Jens Axboe320ae512013-10-24 09:20:05 +01001484 if (q->mq_ops)
1485 blk_mq_free_request(req);
1486 else {
1487 unsigned long flags;
1488
1489 spin_lock_irqsave(q->queue_lock, flags);
1490 __blk_put_request(q, req);
1491 spin_unlock_irqrestore(q->queue_lock, flags);
1492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494EXPORT_SYMBOL(blk_put_request);
1495
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001496/**
1497 * blk_add_request_payload - add a payload to a request
1498 * @rq: request to update
1499 * @page: page backing the payload
1500 * @len: length of the payload.
1501 *
1502 * This allows to later add a payload to an already submitted request by
1503 * a block driver. The driver needs to take care of freeing the payload
1504 * itself.
1505 *
1506 * Note that this is a quite horrible hack and nothing but handling of
1507 * discard requests should ever use it.
1508 */
1509void blk_add_request_payload(struct request *rq, struct page *page,
1510 unsigned int len)
1511{
1512 struct bio *bio = rq->bio;
1513
1514 bio->bi_io_vec->bv_page = page;
1515 bio->bi_io_vec->bv_offset = 0;
1516 bio->bi_io_vec->bv_len = len;
1517
Kent Overstreet4f024f32013-10-11 15:44:27 -07001518 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001519 bio->bi_vcnt = 1;
1520 bio->bi_phys_segments = 1;
1521
1522 rq->__data_len = rq->resid_len = len;
1523 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001524}
1525EXPORT_SYMBOL_GPL(blk_add_request_payload);
1526
Jens Axboe320ae512013-10-24 09:20:05 +01001527bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1528 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001529{
1530 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1531
Jens Axboe73c10102011-03-08 13:19:51 +01001532 if (!ll_back_merge_fn(q, req, bio))
1533 return false;
1534
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001535 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001536
1537 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1538 blk_rq_set_mixed_merge(req);
1539
1540 req->biotail->bi_next = bio;
1541 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001542 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001543 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1544
Jens Axboe320ae512013-10-24 09:20:05 +01001545 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001546 return true;
1547}
1548
Jens Axboe320ae512013-10-24 09:20:05 +01001549bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1550 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001551{
1552 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001553
Jens Axboe73c10102011-03-08 13:19:51 +01001554 if (!ll_front_merge_fn(q, req, bio))
1555 return false;
1556
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001557 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001558
1559 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1560 blk_rq_set_mixed_merge(req);
1561
Jens Axboe73c10102011-03-08 13:19:51 +01001562 bio->bi_next = req->bio;
1563 req->bio = bio;
1564
Kent Overstreet4f024f32013-10-11 15:44:27 -07001565 req->__sector = bio->bi_iter.bi_sector;
1566 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001567 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1568
Jens Axboe320ae512013-10-24 09:20:05 +01001569 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001570 return true;
1571}
1572
Tejun Heobd87b582011-10-19 14:33:08 +02001573/**
Jens Axboe320ae512013-10-24 09:20:05 +01001574 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001575 * @q: request_queue new bio is being queued at
1576 * @bio: new bio being queued
1577 * @request_count: out parameter for number of traversed plugged requests
Randy Dunlapccc26002015-10-30 18:36:16 -07001578 * @same_queue_rq: pointer to &struct request that gets filled in when
1579 * another request associated with @q is found on the plug list
1580 * (optional, may be %NULL)
Tejun Heobd87b582011-10-19 14:33:08 +02001581 *
1582 * Determine whether @bio being queued on @q can be merged with a request
1583 * on %current's plugged list. Returns %true if merge was successful,
1584 * otherwise %false.
1585 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001586 * Plugging coalesces IOs from the same issuer for the same purpose without
1587 * going through @q->queue_lock. As such it's more of an issuing mechanism
1588 * than scheduling, and the request, while may have elvpriv data, is not
1589 * added on the elevator at this point. In addition, we don't have
1590 * reliable access to the elevator outside queue lock. Only check basic
1591 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001592 *
1593 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001594 */
Jens Axboe320ae512013-10-24 09:20:05 +01001595bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Shaohua Li5b3f3412015-05-08 10:51:33 -07001596 unsigned int *request_count,
1597 struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +01001598{
1599 struct blk_plug *plug;
1600 struct request *rq;
1601 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001602 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001603
Tejun Heobd87b582011-10-19 14:33:08 +02001604 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001605 if (!plug)
1606 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001607 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001608
Shaohua Li92f399c2013-10-29 12:01:03 -06001609 if (q->mq_ops)
1610 plug_list = &plug->mq_list;
1611 else
1612 plug_list = &plug->list;
1613
1614 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001615 int el_ret;
1616
Shaohua Li5b3f3412015-05-08 10:51:33 -07001617 if (rq->q == q) {
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001618 (*request_count)++;
Shaohua Li5b3f3412015-05-08 10:51:33 -07001619 /*
1620 * Only blk-mq multiple hardware queues case checks the
1621 * rq in the same queue, there should be only one such
1622 * rq in a queue
1623 **/
1624 if (same_queue_rq)
1625 *same_queue_rq = rq;
1626 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001627
Tejun Heo07c2bd32012-02-08 09:19:42 +01001628 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001629 continue;
1630
Tejun Heo050c8ea2012-02-08 09:19:38 +01001631 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001632 if (el_ret == ELEVATOR_BACK_MERGE) {
1633 ret = bio_attempt_back_merge(q, rq, bio);
1634 if (ret)
1635 break;
1636 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1637 ret = bio_attempt_front_merge(q, rq, bio);
1638 if (ret)
1639 break;
1640 }
1641 }
1642out:
1643 return ret;
1644}
1645
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001646unsigned int blk_plug_queued_count(struct request_queue *q)
1647{
1648 struct blk_plug *plug;
1649 struct request *rq;
1650 struct list_head *plug_list;
1651 unsigned int ret = 0;
1652
1653 plug = current->plug;
1654 if (!plug)
1655 goto out;
1656
1657 if (q->mq_ops)
1658 plug_list = &plug->mq_list;
1659 else
1660 plug_list = &plug->list;
1661
1662 list_for_each_entry(rq, plug_list, queuelist) {
1663 if (rq->q == q)
1664 ret++;
1665 }
1666out:
1667 return ret;
1668}
1669
Jens Axboe86db1e22008-01-29 14:53:40 +01001670void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001671{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001672 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001673
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001674 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1675 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001676 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001677
Tejun Heo52d9e672006-01-06 09:49:58 +01001678 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001679 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001680 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001681 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001682}
1683
Jens Axboedece1632015-11-05 10:41:16 -07001684static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001686 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001687 struct blk_plug *plug;
1688 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1689 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001690 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
Kent Overstreet54efd502015-04-23 22:37:18 -07001692 blk_queue_split(q, &bio, q->bio_split);
1693
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 /*
1695 * low level driver can indicate that it wants pages above a
1696 * certain limit bounced to low memory (ie for highmem, or even
1697 * ISA dma in theory)
1698 */
1699 blk_queue_bounce(q, &bio);
1700
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001701 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001702 bio->bi_error = -EIO;
1703 bio_endio(bio);
Jens Axboedece1632015-11-05 10:41:16 -07001704 return BLK_QC_T_NONE;
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001705 }
1706
Tejun Heo4fed9472010-09-03 11:56:17 +02001707 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001708 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001709 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001710 goto get_rq;
1711 }
1712
Jens Axboe73c10102011-03-08 13:19:51 +01001713 /*
1714 * Check if we can merge with the plugged list before grabbing
1715 * any locks.
1716 */
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001717 if (!blk_queue_nomerges(q)) {
1718 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
Jens Axboedece1632015-11-05 10:41:16 -07001719 return BLK_QC_T_NONE;
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001720 } else
1721 request_count = blk_plug_queued_count(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001722
1723 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001726 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001727 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001728 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001729 if (!attempt_back_merge(q, req))
1730 elv_merged_request(q, req, el_ret);
1731 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001732 }
Jens Axboe73c10102011-03-08 13:19:51 +01001733 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001734 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001735 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001736 if (!attempt_front_merge(q, req))
1737 elv_merged_request(q, req, el_ret);
1738 goto out_unlock;
1739 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 }
1741
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001743 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001744 * This sync check and mask will be re-done in init_request_from_bio(),
1745 * but we need to set it earlier to expose the sync flag to the
1746 * rq allocator and io schedulers.
1747 */
1748 rw_flags = bio_data_dir(bio);
1749 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001750 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001751
1752 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001753 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001754 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001755 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001756 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001757 if (IS_ERR(req)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001758 bio->bi_error = PTR_ERR(req);
1759 bio_endio(bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001760 goto out_unlock;
1761 }
Nick Piggind6344532005-06-28 20:45:14 -07001762
Nick Piggin450991b2005-06-28 20:45:13 -07001763 /*
1764 * After dropping the lock and possibly sleeping here, our request
1765 * may now be mergeable after it had proven unmergeable (above).
1766 * We don't worry about that case for efficiency. It won't happen
1767 * often, and the elevators are able to handle it.
1768 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001769 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Tao Ma9562ad92011-10-24 16:11:30 +02001771 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001772 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001773
Jens Axboe73c10102011-03-08 13:19:51 +01001774 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001775 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001776 /*
1777 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001778 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001779 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001780 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001781 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001782 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001783 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001784 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001785 trace_block_plug(q);
1786 }
Jens Axboe73c10102011-03-08 13:19:51 +01001787 }
Shaohua Lia6327162011-08-24 16:04:32 +02001788 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001789 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001790 } else {
1791 spin_lock_irq(q->queue_lock);
1792 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001793 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001794out_unlock:
1795 spin_unlock_irq(q->queue_lock);
1796 }
Jens Axboedece1632015-11-05 10:41:16 -07001797
1798 return BLK_QC_T_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
1800
1801/*
1802 * If bio->bi_dev is a partition, remap the location
1803 */
1804static inline void blk_partition_remap(struct bio *bio)
1805{
1806 struct block_device *bdev = bio->bi_bdev;
1807
Jens Axboebf2de6f2007-09-27 13:01:25 +02001808 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001810
Kent Overstreet4f024f32013-10-11 15:44:27 -07001811 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001813
Mike Snitzerd07335e2010-11-16 12:52:38 +01001814 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1815 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001816 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 }
1818}
1819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820static void handle_bad_sector(struct bio *bio)
1821{
1822 char b[BDEVNAME_SIZE];
1823
1824 printk(KERN_INFO "attempt to access beyond end of device\n");
1825 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1826 bdevname(bio->bi_bdev, b),
1827 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001828 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001829 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830}
1831
Akinobu Mitac17bb492006-12-08 02:39:46 -08001832#ifdef CONFIG_FAIL_MAKE_REQUEST
1833
1834static DECLARE_FAULT_ATTR(fail_make_request);
1835
1836static int __init setup_fail_make_request(char *str)
1837{
1838 return setup_fault_attr(&fail_make_request, str);
1839}
1840__setup("fail_make_request=", setup_fail_make_request);
1841
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001842static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001843{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001844 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001845}
1846
1847static int __init fail_make_request_debugfs(void)
1848{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001849 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1850 NULL, &fail_make_request);
1851
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001852 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001853}
1854
1855late_initcall(fail_make_request_debugfs);
1856
1857#else /* CONFIG_FAIL_MAKE_REQUEST */
1858
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001859static inline bool should_fail_request(struct hd_struct *part,
1860 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001861{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001862 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001863}
1864
1865#endif /* CONFIG_FAIL_MAKE_REQUEST */
1866
Jens Axboec07e2b42007-07-18 13:27:58 +02001867/*
1868 * Check whether this bio extends beyond the end of the device.
1869 */
1870static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1871{
1872 sector_t maxsector;
1873
1874 if (!nr_sectors)
1875 return 0;
1876
1877 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001878 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001879 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001880 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001881
1882 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1883 /*
1884 * This may well happen - the kernel calls bread()
1885 * without checking the size of the device, e.g., when
1886 * mounting a device.
1887 */
1888 handle_bad_sector(bio);
1889 return 1;
1890 }
1891 }
1892
1893 return 0;
1894}
1895
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001896static noinline_for_stack bool
1897generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Jens Axboe165125e2007-07-24 09:28:11 +02001899 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001900 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001901 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001902 char b[BDEVNAME_SIZE];
1903 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Jens Axboec07e2b42007-07-18 13:27:58 +02001907 if (bio_check_eod(bio, nr_sectors))
1908 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001910 q = bdev_get_queue(bio->bi_bdev);
1911 if (unlikely(!q)) {
1912 printk(KERN_ERR
1913 "generic_make_request: Trying to access "
1914 "nonexistent block-device %s (%Lu)\n",
1915 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001916 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001917 goto end_io;
1918 }
1919
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001920 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001921 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001922 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001923 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001924 goto end_io;
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001927 * If this device has partitions, remap block n
1928 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001930 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001932 if (bio_check_eod(bio, nr_sectors))
1933 goto end_io;
1934
1935 /*
1936 * Filter flush bio's early so that make_request based
1937 * drivers without flush support don't have to worry
1938 * about them.
1939 */
1940 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1941 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1942 if (!nr_sectors) {
1943 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001944 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001948 if ((bio->bi_rw & REQ_DISCARD) &&
1949 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001950 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001951 err = -EOPNOTSUPP;
1952 goto end_io;
1953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001955 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 err = -EOPNOTSUPP;
1957 goto end_io;
1958 }
1959
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001960 /*
1961 * Various block parts want %current->io_context and lazy ioc
1962 * allocation ends up trading a lot of pain for a small amount of
1963 * memory. Just allocate it upfront. This may fail and block
1964 * layer knows how to live with it.
1965 */
1966 create_io_context(GFP_ATOMIC, q->node);
1967
Tejun Heoae118892015-08-18 14:55:20 -07001968 if (!blkcg_bio_issue_check(q, bio))
1969 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001970
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001971 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001972 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001973
1974end_io:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001975 bio->bi_error = err;
1976 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001977 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978}
1979
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001980/**
1981 * generic_make_request - hand a buffer to its device driver for I/O
1982 * @bio: The bio describing the location in memory and on the device.
1983 *
1984 * generic_make_request() is used to make I/O requests of block
1985 * devices. It is passed a &struct bio, which describes the I/O that needs
1986 * to be done.
1987 *
1988 * generic_make_request() does not return any status. The
1989 * success/failure status of the request, along with notification of
1990 * completion, is delivered asynchronously through the bio->bi_end_io
1991 * function described (one day) else where.
1992 *
1993 * The caller of generic_make_request must make sure that bi_io_vec
1994 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1995 * set to describe the device address, and the
1996 * bi_end_io and optionally bi_private are set to describe how
1997 * completion notification should be signaled.
1998 *
1999 * generic_make_request and the drivers it calls may use bi_next if this
2000 * bio happens to be merged with someone else, and may resubmit the bio to
2001 * a lower device by calling into generic_make_request recursively, which
2002 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02002003 */
Jens Axboedece1632015-11-05 10:41:16 -07002004blk_qc_t generic_make_request(struct bio *bio)
Neil Brownd89d8792007-05-01 09:53:42 +02002005{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002006 struct bio_list bio_list_on_stack;
Jens Axboedece1632015-11-05 10:41:16 -07002007 blk_qc_t ret = BLK_QC_T_NONE;
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002008
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002009 if (!generic_make_request_checks(bio))
Jens Axboedece1632015-11-05 10:41:16 -07002010 goto out;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002011
2012 /*
2013 * We only want one ->make_request_fn to be active at a time, else
2014 * stack usage with stacked devices could be a problem. So use
2015 * current->bio_list to keep a list of requests submited by a
2016 * make_request_fn function. current->bio_list is also used as a
2017 * flag to say if generic_make_request is currently active in this
2018 * task or not. If it is NULL, then no make_request is active. If
2019 * it is non-NULL, then a make_request is active, and new requests
2020 * should be added at the tail
2021 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002022 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002023 bio_list_add(current->bio_list, bio);
Jens Axboedece1632015-11-05 10:41:16 -07002024 goto out;
Neil Brownd89d8792007-05-01 09:53:42 +02002025 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002026
Neil Brownd89d8792007-05-01 09:53:42 +02002027 /* following loop may be a bit non-obvious, and so deserves some
2028 * explanation.
2029 * Before entering the loop, bio->bi_next is NULL (as all callers
2030 * ensure that) so we have a list with a single bio.
2031 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002032 * we assign bio_list to a pointer to the bio_list_on_stack,
2033 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002034 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02002035 * through a recursive call to generic_make_request. If it
2036 * did, we find a non-NULL value in bio_list and re-enter the loop
2037 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002038 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002039 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02002040 */
2041 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002042 bio_list_init(&bio_list_on_stack);
2043 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02002044 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002045 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2046
Mel Gorman71baba42015-11-06 16:28:28 -08002047 if (likely(blk_queue_enter(q, __GFP_DIRECT_RECLAIM) == 0)) {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002048
Jens Axboedece1632015-11-05 10:41:16 -07002049 ret = q->make_request_fn(q, bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002050
2051 blk_queue_exit(q);
2052
2053 bio = bio_list_pop(current->bio_list);
2054 } else {
2055 struct bio *bio_next = bio_list_pop(current->bio_list);
2056
2057 bio_io_error(bio);
2058 bio = bio_next;
2059 }
Neil Brownd89d8792007-05-01 09:53:42 +02002060 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002061 current->bio_list = NULL; /* deactivate */
Jens Axboedece1632015-11-05 10:41:16 -07002062
2063out:
2064 return ret;
Neil Brownd89d8792007-05-01 09:53:42 +02002065}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066EXPORT_SYMBOL(generic_make_request);
2067
2068/**
Randy Dunlap710027a2008-08-19 20:13:11 +02002069 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2071 * @bio: The &struct bio which describes the I/O
2072 *
2073 * submit_bio() is very similar in purpose to generic_make_request(), and
2074 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02002075 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 *
2077 */
Jens Axboedece1632015-11-05 10:41:16 -07002078blk_qc_t submit_bio(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079{
Jens Axboe22e2c502005-06-27 10:55:12 +02002080 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Jens Axboebf2de6f2007-09-27 13:01:25 +02002082 /*
2083 * If it's a regular read/write or a barrier with data attached,
2084 * go through the normal accounting stuff before submission.
2085 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002086 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002087 unsigned int count;
2088
2089 if (unlikely(rw & REQ_WRITE_SAME))
2090 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2091 else
2092 count = bio_sectors(bio);
2093
Jens Axboebf2de6f2007-09-27 13:01:25 +02002094 if (rw & WRITE) {
2095 count_vm_events(PGPGOUT, count);
2096 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07002097 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002098 count_vm_events(PGPGIN, count);
2099 }
2100
2101 if (unlikely(block_dump)) {
2102 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02002103 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07002104 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02002105 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002106 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02002107 bdevname(bio->bi_bdev, b),
2108 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
2111
Jens Axboedece1632015-11-05 10:41:16 -07002112 return generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114EXPORT_SYMBOL(submit_bio);
2115
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002116/**
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002117 * blk_rq_check_limits - Helper function to check a request for the queue limit
2118 * @q: the queue
2119 * @rq: the request being checked
2120 *
2121 * Description:
2122 * @rq may have been made based on weaker limitations of upper-level queues
2123 * in request stacking drivers, and it may violate the limitation of @q.
2124 * Since the block layer and the underlying device driver trust @rq
2125 * after it is inserted to @q, it should be checked against @q before
2126 * the insertion using this generic function.
2127 *
2128 * This function should also be useful for request stacking drivers
Stefan Weileef35c22010-08-06 21:11:15 +02002129 * in some cases below, so export this function.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002130 * Request stacking drivers like request-based dm may change the queue
2131 * limits while requests are in the queue (e.g. dm's table swapping).
Masanari Iidae2278672014-02-18 22:54:36 +09002132 * Such request stacking drivers should check those requests against
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002133 * the new queue limits again when they dispatch those requests,
2134 * although such checkings are also done against the old queue limits
2135 * when submitting requests.
2136 */
2137int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2138{
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002139 if (!rq_mergeable(rq))
ike Snitzer33839772010-08-08 12:11:33 -04002140 return 0;
2141
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04002142 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002143 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2144 return -EIO;
2145 }
2146
2147 /*
2148 * queue's settings related to segment counting like q->bounce_pfn
2149 * may differ from that of other stacking queues.
2150 * Recalculate it to check the request correctly on this queue's
2151 * limitation.
2152 */
2153 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002154 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002155 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2156 return -EIO;
2157 }
2158
2159 return 0;
2160}
2161EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2162
2163/**
2164 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2165 * @q: the queue to submit the request
2166 * @rq: the request being queued
2167 */
2168int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2169{
2170 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002171 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002172
2173 if (blk_rq_check_limits(q, rq))
2174 return -EIO;
2175
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002176 if (rq->rq_disk &&
2177 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002178 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002179
Keith Busch7fb48982014-10-17 17:46:38 -06002180 if (q->mq_ops) {
2181 if (blk_queue_io_stat(q))
2182 blk_account_io_start(rq, true);
2183 blk_mq_insert_request(rq, false, true, true);
2184 return 0;
2185 }
2186
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002187 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002188 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002189 spin_unlock_irqrestore(q->queue_lock, flags);
2190 return -ENODEV;
2191 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002192
2193 /*
2194 * Submitting request must be dequeued before calling this function
2195 * because it will be linked to another request_queue
2196 */
2197 BUG_ON(blk_queued_rq(rq));
2198
Jeff Moyer4853aba2011-08-15 21:37:25 +02002199 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2200 where = ELEVATOR_INSERT_FLUSH;
2201
2202 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002203 if (where == ELEVATOR_INSERT_FLUSH)
2204 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002205 spin_unlock_irqrestore(q->queue_lock, flags);
2206
2207 return 0;
2208}
2209EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2210
Tejun Heo80a761f2009-07-03 17:48:17 +09002211/**
2212 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2213 * @rq: request to examine
2214 *
2215 * Description:
2216 * A request could be merge of IOs which require different failure
2217 * handling. This function determines the number of bytes which
2218 * can be failed from the beginning of the request without
2219 * crossing into area which need to be retried further.
2220 *
2221 * Return:
2222 * The number of bytes to fail.
2223 *
2224 * Context:
2225 * queue_lock must be held.
2226 */
2227unsigned int blk_rq_err_bytes(const struct request *rq)
2228{
2229 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2230 unsigned int bytes = 0;
2231 struct bio *bio;
2232
2233 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2234 return blk_rq_bytes(rq);
2235
2236 /*
2237 * Currently the only 'mixing' which can happen is between
2238 * different fastfail types. We can safely fail portions
2239 * which have all the failfast bits that the first one has -
2240 * the ones which are at least as eager to fail as the first
2241 * one.
2242 */
2243 for (bio = rq->bio; bio; bio = bio->bi_next) {
2244 if ((bio->bi_rw & ff) != ff)
2245 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002246 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002247 }
2248
2249 /* this could lead to infinite loop */
2250 BUG_ON(blk_rq_bytes(rq) && !bytes);
2251 return bytes;
2252}
2253EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2254
Jens Axboe320ae512013-10-24 09:20:05 +01002255void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002256{
Jens Axboec2553b52009-04-24 08:10:11 +02002257 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002258 const int rw = rq_data_dir(req);
2259 struct hd_struct *part;
2260 int cpu;
2261
2262 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002263 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002264 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2265 part_stat_unlock();
2266 }
2267}
2268
Jens Axboe320ae512013-10-24 09:20:05 +01002269void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002270{
Jens Axboebc58ba92009-01-23 10:54:44 +01002271 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002272 * Account IO completion. flush_rq isn't accounted as a
2273 * normal IO on queueing nor completion. Accounting the
2274 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002275 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002276 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002277 unsigned long duration = jiffies - req->start_time;
2278 const int rw = rq_data_dir(req);
2279 struct hd_struct *part;
2280 int cpu;
2281
2282 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002283 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002284
2285 part_stat_inc(cpu, part, ios[rw]);
2286 part_stat_add(cpu, part, ticks[rw], duration);
2287 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002288 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002289
Jens Axboe6c23a962011-01-07 08:43:37 +01002290 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002291 part_stat_unlock();
2292 }
2293}
2294
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002295#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002296/*
2297 * Don't process normal requests when queue is suspended
2298 * or in the process of suspending/resuming
2299 */
2300static struct request *blk_pm_peek_request(struct request_queue *q,
2301 struct request *rq)
2302{
2303 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2304 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2305 return NULL;
2306 else
2307 return rq;
2308}
2309#else
2310static inline struct request *blk_pm_peek_request(struct request_queue *q,
2311 struct request *rq)
2312{
2313 return rq;
2314}
2315#endif
2316
Jens Axboe320ae512013-10-24 09:20:05 +01002317void blk_account_io_start(struct request *rq, bool new_io)
2318{
2319 struct hd_struct *part;
2320 int rw = rq_data_dir(rq);
2321 int cpu;
2322
2323 if (!blk_do_io_stat(rq))
2324 return;
2325
2326 cpu = part_stat_lock();
2327
2328 if (!new_io) {
2329 part = rq->part;
2330 part_stat_inc(cpu, part, merges[rw]);
2331 } else {
2332 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2333 if (!hd_struct_try_get(part)) {
2334 /*
2335 * The partition is already being removed,
2336 * the request will be accounted on the disk only
2337 *
2338 * We take a reference on disk->part0 although that
2339 * partition will never be deleted, so we can treat
2340 * it as any other partition.
2341 */
2342 part = &rq->rq_disk->part0;
2343 hd_struct_get(part);
2344 }
2345 part_round_stats(cpu, part);
2346 part_inc_in_flight(part, rw);
2347 rq->part = part;
2348 }
2349
2350 part_stat_unlock();
2351}
2352
Tejun Heo53a08802008-12-03 12:41:26 +01002353/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002354 * blk_peek_request - peek at the top of a request queue
2355 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002356 *
2357 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002358 * Return the request at the top of @q. The returned request
2359 * should be started using blk_start_request() before LLD starts
2360 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002361 *
2362 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002363 * Pointer to the request at the top of @q if available. Null
2364 * otherwise.
2365 *
2366 * Context:
2367 * queue_lock must be held.
2368 */
2369struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002370{
2371 struct request *rq;
2372 int ret;
2373
2374 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002375
2376 rq = blk_pm_peek_request(q, rq);
2377 if (!rq)
2378 break;
2379
Tejun Heo158dbda2009-04-23 11:05:18 +09002380 if (!(rq->cmd_flags & REQ_STARTED)) {
2381 /*
2382 * This is the first time the device driver
2383 * sees this request (possibly after
2384 * requeueing). Notify IO scheduler.
2385 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002386 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002387 elv_activate_rq(q, rq);
2388
2389 /*
2390 * just mark as started even if we don't start
2391 * it, a request that has been delayed should
2392 * not be passed by new incoming requests
2393 */
2394 rq->cmd_flags |= REQ_STARTED;
2395 trace_block_rq_issue(q, rq);
2396 }
2397
2398 if (!q->boundary_rq || q->boundary_rq == rq) {
2399 q->end_sector = rq_end_sector(rq);
2400 q->boundary_rq = NULL;
2401 }
2402
2403 if (rq->cmd_flags & REQ_DONTPREP)
2404 break;
2405
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002406 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002407 /*
2408 * make sure space for the drain appears we
2409 * know we can do this because max_hw_segments
2410 * has been adjusted to be one fewer than the
2411 * device can handle
2412 */
2413 rq->nr_phys_segments++;
2414 }
2415
2416 if (!q->prep_rq_fn)
2417 break;
2418
2419 ret = q->prep_rq_fn(q, rq);
2420 if (ret == BLKPREP_OK) {
2421 break;
2422 } else if (ret == BLKPREP_DEFER) {
2423 /*
2424 * the request may have been (partially) prepped.
2425 * we need to keep this request in the front to
2426 * avoid resource deadlock. REQ_STARTED will
2427 * prevent other fs requests from passing this one.
2428 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002429 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002430 !(rq->cmd_flags & REQ_DONTPREP)) {
2431 /*
2432 * remove the space for the drain we added
2433 * so that we don't add it again
2434 */
2435 --rq->nr_phys_segments;
2436 }
2437
2438 rq = NULL;
2439 break;
2440 } else if (ret == BLKPREP_KILL) {
2441 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002442 /*
2443 * Mark this request as started so we don't trigger
2444 * any debug logic in the end I/O path.
2445 */
2446 blk_start_request(rq);
Tejun Heo40cbbb72009-04-23 11:05:19 +09002447 __blk_end_request_all(rq, -EIO);
Tejun Heo158dbda2009-04-23 11:05:18 +09002448 } else {
2449 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2450 break;
2451 }
2452 }
2453
2454 return rq;
2455}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002456EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002457
Tejun Heo9934c8c2009-05-08 11:54:16 +09002458void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002459{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002460 struct request_queue *q = rq->q;
2461
Tejun Heo158dbda2009-04-23 11:05:18 +09002462 BUG_ON(list_empty(&rq->queuelist));
2463 BUG_ON(ELV_ON_HASH(rq));
2464
2465 list_del_init(&rq->queuelist);
2466
2467 /*
2468 * the time frame between a request being removed from the lists
2469 * and to it is freed is accounted as io that is in progress at
2470 * the driver side.
2471 */
Divyesh Shah91952912010-04-01 15:01:41 -07002472 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002473 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002474 set_io_start_time_ns(rq);
2475 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002476}
2477
Tejun Heo5efccd12009-04-23 11:05:18 +09002478/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002479 * blk_start_request - start request processing on the driver
2480 * @req: request to dequeue
2481 *
2482 * Description:
2483 * Dequeue @req and start timeout timer on it. This hands off the
2484 * request to the driver.
2485 *
2486 * Block internal functions which don't want to start timer should
2487 * call blk_dequeue_request().
2488 *
2489 * Context:
2490 * queue_lock must be held.
2491 */
2492void blk_start_request(struct request *req)
2493{
2494 blk_dequeue_request(req);
2495
2496 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002497 * We are now handing the request to the hardware, initialize
2498 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002499 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002500 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002501 if (unlikely(blk_bidi_rq(req)))
2502 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2503
Jeff Moyer4912aa62013-10-08 14:36:41 -04002504 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002505 blk_add_timer(req);
2506}
2507EXPORT_SYMBOL(blk_start_request);
2508
2509/**
2510 * blk_fetch_request - fetch a request from a request queue
2511 * @q: request queue to fetch a request from
2512 *
2513 * Description:
2514 * Return the request at the top of @q. The request is started on
2515 * return and LLD can start processing it immediately.
2516 *
2517 * Return:
2518 * Pointer to the request at the top of @q if available. Null
2519 * otherwise.
2520 *
2521 * Context:
2522 * queue_lock must be held.
2523 */
2524struct request *blk_fetch_request(struct request_queue *q)
2525{
2526 struct request *rq;
2527
2528 rq = blk_peek_request(q);
2529 if (rq)
2530 blk_start_request(rq);
2531 return rq;
2532}
2533EXPORT_SYMBOL(blk_fetch_request);
2534
2535/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002536 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002537 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002538 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002539 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002540 *
2541 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002542 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2543 * the request structure even if @req doesn't have leftover.
2544 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002545 *
2546 * This special helper function is only for request stacking drivers
2547 * (e.g. request-based dm) so that they can handle partial completion.
2548 * Actual device drivers should use blk_end_request instead.
2549 *
2550 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2551 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002552 *
2553 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002554 * %false - this request doesn't have any more data
2555 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002556 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002557bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002559 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002561 trace_block_rq_complete(req->q, req, nr_bytes);
2562
Tejun Heo2e60e022009-04-23 11:05:18 +09002563 if (!req->bio)
2564 return false;
2565
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002567 * For fs requests, rq is just carrier of independent bio's
2568 * and each partial completion should be handled separately.
2569 * Reset per-request error on each partial completion.
2570 *
2571 * TODO: tj: This is too subtle. It would be better to let
2572 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002574 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 req->errors = 0;
2576
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002577 if (error && req->cmd_type == REQ_TYPE_FS &&
2578 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002579 char *error_type;
2580
2581 switch (error) {
2582 case -ENOLINK:
2583 error_type = "recoverable transport";
2584 break;
2585 case -EREMOTEIO:
2586 error_type = "critical target";
2587 break;
2588 case -EBADE:
2589 error_type = "critical nexus";
2590 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002591 case -ETIMEDOUT:
2592 error_type = "timeout";
2593 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002594 case -ENOSPC:
2595 error_type = "critical space allocation";
2596 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002597 case -ENODATA:
2598 error_type = "critical medium";
2599 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002600 case -EIO:
2601 default:
2602 error_type = "I/O";
2603 break;
2604 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002605 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2606 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002607 req->rq_disk->disk_name : "?",
2608 (unsigned long long)blk_rq_pos(req));
2609
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 }
2611
Jens Axboebc58ba92009-01-23 10:54:44 +01002612 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002613
Kent Overstreetf79ea412012-09-20 16:38:30 -07002614 total_bytes = 0;
2615 while (req->bio) {
2616 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002617 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Kent Overstreet4f024f32013-10-11 15:44:27 -07002619 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Kent Overstreetf79ea412012-09-20 16:38:30 -07002622 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Kent Overstreetf79ea412012-09-20 16:38:30 -07002624 total_bytes += bio_bytes;
2625 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Kent Overstreetf79ea412012-09-20 16:38:30 -07002627 if (!nr_bytes)
2628 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 }
2630
2631 /*
2632 * completely done
2633 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002634 if (!req->bio) {
2635 /*
2636 * Reset counters so that the request stacking driver
2637 * can find how many bytes remain in the request
2638 * later.
2639 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002640 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002641 return false;
2642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002644 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002645
2646 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002647 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002648 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002649
Tejun Heo80a761f2009-07-03 17:48:17 +09002650 /* mixed attributes always follow the first bio */
2651 if (req->cmd_flags & REQ_MIXED_MERGE) {
2652 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2653 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2654 }
2655
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002656 /*
2657 * If total number of sectors is less than the first segment
2658 * size, something has gone terribly wrong.
2659 */
2660 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002661 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002662 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002663 }
2664
2665 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002667
Tejun Heo2e60e022009-04-23 11:05:18 +09002668 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669}
Tejun Heo2e60e022009-04-23 11:05:18 +09002670EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Tejun Heo2e60e022009-04-23 11:05:18 +09002672static bool blk_update_bidi_request(struct request *rq, int error,
2673 unsigned int nr_bytes,
2674 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002675{
Tejun Heo2e60e022009-04-23 11:05:18 +09002676 if (blk_update_request(rq, error, nr_bytes))
2677 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002678
Tejun Heo2e60e022009-04-23 11:05:18 +09002679 /* Bidi request must be completed as a whole */
2680 if (unlikely(blk_bidi_rq(rq)) &&
2681 blk_update_request(rq->next_rq, error, bidi_bytes))
2682 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002683
Jens Axboee2e1a142010-06-09 10:42:09 +02002684 if (blk_queue_add_random(rq->q))
2685 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002686
2687 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688}
2689
James Bottomley28018c22010-07-01 19:49:17 +09002690/**
2691 * blk_unprep_request - unprepare a request
2692 * @req: the request
2693 *
2694 * This function makes a request ready for complete resubmission (or
2695 * completion). It happens only after all error handling is complete,
2696 * so represents the appropriate moment to deallocate any resources
2697 * that were allocated to the request in the prep_rq_fn. The queue
2698 * lock is held when calling this.
2699 */
2700void blk_unprep_request(struct request *req)
2701{
2702 struct request_queue *q = req->q;
2703
2704 req->cmd_flags &= ~REQ_DONTPREP;
2705 if (q->unprep_rq_fn)
2706 q->unprep_rq_fn(q, req);
2707}
2708EXPORT_SYMBOL_GPL(blk_unprep_request);
2709
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710/*
2711 * queue lock must be held
2712 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002713void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002715 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002716 blk_queue_end_tag(req->q, req);
2717
James Bottomleyba396a62009-05-27 14:17:08 +02002718 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002720 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002721 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Mike Andersone78042e2008-10-30 02:16:20 -07002723 blk_delete_timer(req);
2724
James Bottomley28018c22010-07-01 19:49:17 +09002725 if (req->cmd_flags & REQ_DONTPREP)
2726 blk_unprep_request(req);
2727
Jens Axboebc58ba92009-01-23 10:54:44 +01002728 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002731 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002732 else {
2733 if (blk_bidi_rq(req))
2734 __blk_put_request(req->next_rq->q, req->next_rq);
2735
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 }
2738}
Christoph Hellwig12120072014-04-16 09:44:59 +02002739EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002741/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002742 * blk_end_bidi_request - Complete a bidi request
2743 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002744 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002745 * @nr_bytes: number of bytes to complete @rq
2746 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002747 *
2748 * Description:
2749 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002750 * Drivers that supports bidi can safely call this member for any
2751 * type of request, bidi or uni. In the later case @bidi_bytes is
2752 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002753 *
2754 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002755 * %false - we are done with this request
2756 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002757 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002758static bool blk_end_bidi_request(struct request *rq, int error,
2759 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002760{
2761 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002762 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002763
Tejun Heo2e60e022009-04-23 11:05:18 +09002764 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2765 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002766
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002767 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002768 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002769 spin_unlock_irqrestore(q->queue_lock, flags);
2770
Tejun Heo2e60e022009-04-23 11:05:18 +09002771 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002772}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002773
2774/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002775 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2776 * @rq: the request to complete
2777 * @error: %0 for success, < %0 for error
2778 * @nr_bytes: number of bytes to complete @rq
2779 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002780 *
2781 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002782 * Identical to blk_end_bidi_request() except that queue lock is
2783 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002784 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002785 * Return:
2786 * %false - we are done with this request
2787 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002788 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002789bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002790 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002791{
Tejun Heo2e60e022009-04-23 11:05:18 +09002792 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2793 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002794
Tejun Heo2e60e022009-04-23 11:05:18 +09002795 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002796
Tejun Heo2e60e022009-04-23 11:05:18 +09002797 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002798}
2799
2800/**
2801 * blk_end_request - Helper function for drivers to complete the request.
2802 * @rq: the request being processed
2803 * @error: %0 for success, < %0 for error
2804 * @nr_bytes: number of bytes to complete
2805 *
2806 * Description:
2807 * Ends I/O on a number of bytes attached to @rq.
2808 * If @rq has leftover, sets it up for the next range of segments.
2809 *
2810 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002811 * %false - we are done with this request
2812 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002813 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002814bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002815{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002816 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002817}
Jens Axboe56ad1742009-07-28 22:11:24 +02002818EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002819
2820/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002821 * blk_end_request_all - Helper function for drives to finish the request.
2822 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002823 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002824 *
2825 * Description:
2826 * Completely finish @rq.
2827 */
2828void blk_end_request_all(struct request *rq, int error)
2829{
2830 bool pending;
2831 unsigned int bidi_bytes = 0;
2832
2833 if (unlikely(blk_bidi_rq(rq)))
2834 bidi_bytes = blk_rq_bytes(rq->next_rq);
2835
2836 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2837 BUG_ON(pending);
2838}
Jens Axboe56ad1742009-07-28 22:11:24 +02002839EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002840
2841/**
2842 * blk_end_request_cur - Helper function to finish the current request chunk.
2843 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002844 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002845 *
2846 * Description:
2847 * Complete the current consecutively mapped chunk from @rq.
2848 *
2849 * Return:
2850 * %false - we are done with this request
2851 * %true - still buffers pending for this request
2852 */
2853bool blk_end_request_cur(struct request *rq, int error)
2854{
2855 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2856}
Jens Axboe56ad1742009-07-28 22:11:24 +02002857EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002858
2859/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002860 * blk_end_request_err - Finish a request till the next failure boundary.
2861 * @rq: the request to finish till the next failure boundary for
2862 * @error: must be negative errno
2863 *
2864 * Description:
2865 * Complete @rq till the next failure boundary.
2866 *
2867 * Return:
2868 * %false - we are done with this request
2869 * %true - still buffers pending for this request
2870 */
2871bool blk_end_request_err(struct request *rq, int error)
2872{
2873 WARN_ON(error >= 0);
2874 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2875}
2876EXPORT_SYMBOL_GPL(blk_end_request_err);
2877
2878/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002879 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002880 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002881 * @error: %0 for success, < %0 for error
2882 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002883 *
2884 * Description:
2885 * Must be called with queue lock held unlike blk_end_request().
2886 *
2887 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002888 * %false - we are done with this request
2889 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002890 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002891bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002892{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002893 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002894}
Jens Axboe56ad1742009-07-28 22:11:24 +02002895EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002896
2897/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002898 * __blk_end_request_all - Helper function for drives to finish the request.
2899 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002900 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002901 *
2902 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002903 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002904 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002905void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002906{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002907 bool pending;
2908 unsigned int bidi_bytes = 0;
2909
2910 if (unlikely(blk_bidi_rq(rq)))
2911 bidi_bytes = blk_rq_bytes(rq->next_rq);
2912
2913 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2914 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002915}
Jens Axboe56ad1742009-07-28 22:11:24 +02002916EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002917
2918/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002919 * __blk_end_request_cur - Helper function to finish the current request chunk.
2920 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002921 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002922 *
2923 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002924 * Complete the current consecutively mapped chunk from @rq. Must
2925 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002926 *
2927 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002928 * %false - we are done with this request
2929 * %true - still buffers pending for this request
2930 */
2931bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002932{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002933 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002934}
Jens Axboe56ad1742009-07-28 22:11:24 +02002935EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002936
Tejun Heo80a761f2009-07-03 17:48:17 +09002937/**
2938 * __blk_end_request_err - Finish a request till the next failure boundary.
2939 * @rq: the request to finish till the next failure boundary for
2940 * @error: must be negative errno
2941 *
2942 * Description:
2943 * Complete @rq till the next failure boundary. Must be called
2944 * with queue lock held.
2945 *
2946 * Return:
2947 * %false - we are done with this request
2948 * %true - still buffers pending for this request
2949 */
2950bool __blk_end_request_err(struct request *rq, int error)
2951{
2952 WARN_ON(error >= 0);
2953 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2954}
2955EXPORT_SYMBOL_GPL(__blk_end_request_err);
2956
Jens Axboe86db1e22008-01-29 14:53:40 +01002957void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2958 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002960 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002961 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962
Jens Axboeb4f42e22014-04-10 09:46:28 -06002963 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002964 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002965
Kent Overstreet4f024f32013-10-11 15:44:27 -07002966 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968
NeilBrown66846572007-08-16 13:31:28 +02002969 if (bio->bi_bdev)
2970 rq->rq_disk = bio->bi_bdev->bd_disk;
2971}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002973#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2974/**
2975 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2976 * @rq: the request to be flushed
2977 *
2978 * Description:
2979 * Flush all pages in @rq.
2980 */
2981void rq_flush_dcache_pages(struct request *rq)
2982{
2983 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002984 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002985
2986 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002987 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002988}
2989EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2990#endif
2991
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002992/**
2993 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2994 * @q : the queue of the device being checked
2995 *
2996 * Description:
2997 * Check if underlying low-level drivers of a device are busy.
2998 * If the drivers want to export their busy state, they must set own
2999 * exporting function using blk_queue_lld_busy() first.
3000 *
3001 * Basically, this function is used only by request stacking drivers
3002 * to stop dispatching requests to underlying devices when underlying
3003 * devices are busy. This behavior helps more I/O merging on the queue
3004 * of the request stacking driver and prevents I/O throughput regression
3005 * on burst I/O load.
3006 *
3007 * Return:
3008 * 0 - Not busy (The request stacking driver should dispatch request)
3009 * 1 - Busy (The request stacking driver should stop dispatching request)
3010 */
3011int blk_lld_busy(struct request_queue *q)
3012{
3013 if (q->lld_busy_fn)
3014 return q->lld_busy_fn(q);
3015
3016 return 0;
3017}
3018EXPORT_SYMBOL_GPL(blk_lld_busy);
3019
Mike Snitzer78d8e582015-06-26 10:01:13 -04003020/**
3021 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3022 * @rq: the clone request to be cleaned up
3023 *
3024 * Description:
3025 * Free all bios in @rq for a cloned request.
3026 */
3027void blk_rq_unprep_clone(struct request *rq)
3028{
3029 struct bio *bio;
3030
3031 while ((bio = rq->bio) != NULL) {
3032 rq->bio = bio->bi_next;
3033
3034 bio_put(bio);
3035 }
3036}
3037EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3038
3039/*
3040 * Copy attributes of the original request to the clone request.
3041 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3042 */
3043static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003044{
3045 dst->cpu = src->cpu;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003046 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003047 dst->cmd_type = src->cmd_type;
3048 dst->__sector = blk_rq_pos(src);
3049 dst->__data_len = blk_rq_bytes(src);
3050 dst->nr_phys_segments = src->nr_phys_segments;
3051 dst->ioprio = src->ioprio;
3052 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003053}
3054
3055/**
3056 * blk_rq_prep_clone - Helper function to setup clone request
3057 * @rq: the request to be setup
3058 * @rq_src: original request to be cloned
3059 * @bs: bio_set that bios for clone are allocated from
3060 * @gfp_mask: memory allocation mask for bio
3061 * @bio_ctr: setup function to be called for each clone bio.
3062 * Returns %0 for success, non %0 for failure.
3063 * @data: private data to be passed to @bio_ctr
3064 *
3065 * Description:
3066 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3067 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3068 * are not copied, and copying such parts is the caller's responsibility.
3069 * Also, pages which the original bios are pointing to are not copied
3070 * and the cloned bios just point same pages.
3071 * So cloned bios must be completed before original bios, which means
3072 * the caller must complete @rq before @rq_src.
3073 */
3074int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3075 struct bio_set *bs, gfp_t gfp_mask,
3076 int (*bio_ctr)(struct bio *, struct bio *, void *),
3077 void *data)
3078{
3079 struct bio *bio, *bio_src;
3080
3081 if (!bs)
3082 bs = fs_bio_set;
3083
3084 __rq_for_each_bio(bio_src, rq_src) {
3085 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3086 if (!bio)
3087 goto free_and_out;
3088
3089 if (bio_ctr && bio_ctr(bio, bio_src, data))
3090 goto free_and_out;
3091
3092 if (rq->bio) {
3093 rq->biotail->bi_next = bio;
3094 rq->biotail = bio;
3095 } else
3096 rq->bio = rq->biotail = bio;
3097 }
3098
3099 __blk_rq_prep_clone(rq, rq_src);
3100
3101 return 0;
3102
3103free_and_out:
3104 if (bio)
3105 bio_put(bio);
3106 blk_rq_unprep_clone(rq);
3107
3108 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003109}
3110EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3111
Jens Axboe59c3d452014-04-08 09:15:35 -06003112int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113{
3114 return queue_work(kblockd_workqueue, work);
3115}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116EXPORT_SYMBOL(kblockd_schedule_work);
3117
Jens Axboe59c3d452014-04-08 09:15:35 -06003118int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3119 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003120{
3121 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3122}
3123EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3124
Jens Axboe8ab14592014-04-08 09:17:40 -06003125int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3126 unsigned long delay)
3127{
3128 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3129}
3130EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3131
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003132/**
3133 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3134 * @plug: The &struct blk_plug that needs to be initialized
3135 *
3136 * Description:
3137 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3138 * pending I/O should the task end up blocking between blk_start_plug() and
3139 * blk_finish_plug(). This is important from a performance perspective, but
3140 * also ensures that we don't deadlock. For instance, if the task is blocking
3141 * for a memory allocation, memory reclaim could end up wanting to free a
3142 * page belonging to that request that is currently residing in our private
3143 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3144 * this kind of deadlock.
3145 */
Jens Axboe73c10102011-03-08 13:19:51 +01003146void blk_start_plug(struct blk_plug *plug)
3147{
3148 struct task_struct *tsk = current;
3149
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003150 /*
3151 * If this is a nested plug, don't actually assign it.
3152 */
3153 if (tsk->plug)
3154 return;
3155
Jens Axboe73c10102011-03-08 13:19:51 +01003156 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003157 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003158 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003159 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003160 * Store ordering should not be needed here, since a potential
3161 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01003162 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003163 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01003164}
3165EXPORT_SYMBOL(blk_start_plug);
3166
3167static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3168{
3169 struct request *rqa = container_of(a, struct request, queuelist);
3170 struct request *rqb = container_of(b, struct request, queuelist);
3171
Jianpeng Ma975927b2012-10-25 21:58:17 +02003172 return !(rqa->q < rqb->q ||
3173 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003174}
3175
Jens Axboe49cac012011-04-16 13:51:05 +02003176/*
3177 * If 'from_schedule' is true, then postpone the dispatch of requests
3178 * until a safe kblockd context. We due this to avoid accidental big
3179 * additional stack usage in driver dispatch, in places where the originally
3180 * plugger did not intend it.
3181 */
Jens Axboef6603782011-04-15 15:49:07 +02003182static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003183 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003184 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003185{
Jens Axboe49cac012011-04-16 13:51:05 +02003186 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003187
Bart Van Assche70460572012-11-28 13:45:56 +01003188 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003189 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003190 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003191 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003192 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003193}
3194
NeilBrown74018dc2012-07-31 09:08:15 +02003195static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003196{
3197 LIST_HEAD(callbacks);
3198
Shaohua Li2a7d5552012-07-31 09:08:15 +02003199 while (!list_empty(&plug->cb_list)) {
3200 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003201
Shaohua Li2a7d5552012-07-31 09:08:15 +02003202 while (!list_empty(&callbacks)) {
3203 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003204 struct blk_plug_cb,
3205 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003206 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003207 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003208 }
NeilBrown048c9372011-04-18 09:52:22 +02003209 }
3210}
3211
NeilBrown9cbb1752012-07-31 09:08:14 +02003212struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3213 int size)
3214{
3215 struct blk_plug *plug = current->plug;
3216 struct blk_plug_cb *cb;
3217
3218 if (!plug)
3219 return NULL;
3220
3221 list_for_each_entry(cb, &plug->cb_list, list)
3222 if (cb->callback == unplug && cb->data == data)
3223 return cb;
3224
3225 /* Not currently on the callback list */
3226 BUG_ON(size < sizeof(*cb));
3227 cb = kzalloc(size, GFP_ATOMIC);
3228 if (cb) {
3229 cb->data = data;
3230 cb->callback = unplug;
3231 list_add(&cb->list, &plug->cb_list);
3232 }
3233 return cb;
3234}
3235EXPORT_SYMBOL(blk_check_plugged);
3236
Jens Axboe49cac012011-04-16 13:51:05 +02003237void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003238{
3239 struct request_queue *q;
3240 unsigned long flags;
3241 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003242 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003243 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003244
NeilBrown74018dc2012-07-31 09:08:15 +02003245 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003246
3247 if (!list_empty(&plug->mq_list))
3248 blk_mq_flush_plug_list(plug, from_schedule);
3249
Jens Axboe73c10102011-03-08 13:19:51 +01003250 if (list_empty(&plug->list))
3251 return;
3252
NeilBrown109b8122011-04-11 14:13:10 +02003253 list_splice_init(&plug->list, &list);
3254
Jianpeng Ma422765c2013-01-11 14:46:09 +01003255 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003256
3257 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003258 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003259
3260 /*
3261 * Save and disable interrupts here, to avoid doing it for every
3262 * queue lock we have to take.
3263 */
Jens Axboe73c10102011-03-08 13:19:51 +01003264 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003265 while (!list_empty(&list)) {
3266 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003267 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003268 BUG_ON(!rq->q);
3269 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003270 /*
3271 * This drops the queue lock
3272 */
3273 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003274 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003275 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003276 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003277 spin_lock(q->queue_lock);
3278 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003279
3280 /*
3281 * Short-circuit if @q is dead
3282 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003283 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003284 __blk_end_request_all(rq, -ENODEV);
3285 continue;
3286 }
3287
Jens Axboe73c10102011-03-08 13:19:51 +01003288 /*
3289 * rq is already accounted, so use raw insert
3290 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003291 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3292 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3293 else
3294 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003295
3296 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003297 }
3298
Jens Axboe99e22592011-04-18 09:59:55 +02003299 /*
3300 * This drops the queue lock
3301 */
3302 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003303 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003304
Jens Axboe73c10102011-03-08 13:19:51 +01003305 local_irq_restore(flags);
3306}
Jens Axboe73c10102011-03-08 13:19:51 +01003307
3308void blk_finish_plug(struct blk_plug *plug)
3309{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003310 if (plug != current->plug)
3311 return;
Jens Axboef6603782011-04-15 15:49:07 +02003312 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003313
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003314 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003315}
3316EXPORT_SYMBOL(blk_finish_plug);
3317
Jens Axboe05229be2015-11-05 10:44:55 -07003318bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3319{
3320 struct blk_plug *plug;
3321 long state;
3322
3323 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3324 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3325 return false;
3326
3327 plug = current->plug;
3328 if (plug)
3329 blk_flush_plug_list(plug, false);
3330
3331 state = current->state;
3332 while (!need_resched()) {
3333 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3334 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3335 int ret;
3336
3337 hctx->poll_invoked++;
3338
3339 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3340 if (ret > 0) {
3341 hctx->poll_success++;
3342 set_current_state(TASK_RUNNING);
3343 return true;
3344 }
3345
3346 if (signal_pending_state(state, current))
3347 set_current_state(TASK_RUNNING);
3348
3349 if (current->state == TASK_RUNNING)
3350 return true;
3351 if (ret < 0)
3352 break;
3353 cpu_relax();
3354 }
3355
3356 return false;
3357}
3358
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003359#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003360/**
3361 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3362 * @q: the queue of the device
3363 * @dev: the device the queue belongs to
3364 *
3365 * Description:
3366 * Initialize runtime-PM-related fields for @q and start auto suspend for
3367 * @dev. Drivers that want to take advantage of request-based runtime PM
3368 * should call this function after @dev has been initialized, and its
3369 * request queue @q has been allocated, and runtime PM for it can not happen
3370 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3371 * cases, driver should call this function before any I/O has taken place.
3372 *
3373 * This function takes care of setting up using auto suspend for the device,
3374 * the autosuspend delay is set to -1 to make runtime suspend impossible
3375 * until an updated value is either set by user or by driver. Drivers do
3376 * not need to touch other autosuspend settings.
3377 *
3378 * The block layer runtime PM is request based, so only works for drivers
3379 * that use request as their IO unit instead of those directly use bio's.
3380 */
3381void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3382{
3383 q->dev = dev;
3384 q->rpm_status = RPM_ACTIVE;
3385 pm_runtime_set_autosuspend_delay(q->dev, -1);
3386 pm_runtime_use_autosuspend(q->dev);
3387}
3388EXPORT_SYMBOL(blk_pm_runtime_init);
3389
3390/**
3391 * blk_pre_runtime_suspend - Pre runtime suspend check
3392 * @q: the queue of the device
3393 *
3394 * Description:
3395 * This function will check if runtime suspend is allowed for the device
3396 * by examining if there are any requests pending in the queue. If there
3397 * are requests pending, the device can not be runtime suspended; otherwise,
3398 * the queue's status will be updated to SUSPENDING and the driver can
3399 * proceed to suspend the device.
3400 *
3401 * For the not allowed case, we mark last busy for the device so that
3402 * runtime PM core will try to autosuspend it some time later.
3403 *
3404 * This function should be called near the start of the device's
3405 * runtime_suspend callback.
3406 *
3407 * Return:
3408 * 0 - OK to runtime suspend the device
3409 * -EBUSY - Device should not be runtime suspended
3410 */
3411int blk_pre_runtime_suspend(struct request_queue *q)
3412{
3413 int ret = 0;
3414
3415 spin_lock_irq(q->queue_lock);
3416 if (q->nr_pending) {
3417 ret = -EBUSY;
3418 pm_runtime_mark_last_busy(q->dev);
3419 } else {
3420 q->rpm_status = RPM_SUSPENDING;
3421 }
3422 spin_unlock_irq(q->queue_lock);
3423 return ret;
3424}
3425EXPORT_SYMBOL(blk_pre_runtime_suspend);
3426
3427/**
3428 * blk_post_runtime_suspend - Post runtime suspend processing
3429 * @q: the queue of the device
3430 * @err: return value of the device's runtime_suspend function
3431 *
3432 * Description:
3433 * Update the queue's runtime status according to the return value of the
3434 * device's runtime suspend function and mark last busy for the device so
3435 * that PM core will try to auto suspend the device at a later time.
3436 *
3437 * This function should be called near the end of the device's
3438 * runtime_suspend callback.
3439 */
3440void blk_post_runtime_suspend(struct request_queue *q, int err)
3441{
3442 spin_lock_irq(q->queue_lock);
3443 if (!err) {
3444 q->rpm_status = RPM_SUSPENDED;
3445 } else {
3446 q->rpm_status = RPM_ACTIVE;
3447 pm_runtime_mark_last_busy(q->dev);
3448 }
3449 spin_unlock_irq(q->queue_lock);
3450}
3451EXPORT_SYMBOL(blk_post_runtime_suspend);
3452
3453/**
3454 * blk_pre_runtime_resume - Pre runtime resume processing
3455 * @q: the queue of the device
3456 *
3457 * Description:
3458 * Update the queue's runtime status to RESUMING in preparation for the
3459 * runtime resume of the device.
3460 *
3461 * This function should be called near the start of the device's
3462 * runtime_resume callback.
3463 */
3464void blk_pre_runtime_resume(struct request_queue *q)
3465{
3466 spin_lock_irq(q->queue_lock);
3467 q->rpm_status = RPM_RESUMING;
3468 spin_unlock_irq(q->queue_lock);
3469}
3470EXPORT_SYMBOL(blk_pre_runtime_resume);
3471
3472/**
3473 * blk_post_runtime_resume - Post runtime resume processing
3474 * @q: the queue of the device
3475 * @err: return value of the device's runtime_resume function
3476 *
3477 * Description:
3478 * Update the queue's runtime status according to the return value of the
3479 * device's runtime_resume function. If it is successfully resumed, process
3480 * the requests that are queued into the device's queue when it is resuming
3481 * and then mark last busy and initiate autosuspend for it.
3482 *
3483 * This function should be called near the end of the device's
3484 * runtime_resume callback.
3485 */
3486void blk_post_runtime_resume(struct request_queue *q, int err)
3487{
3488 spin_lock_irq(q->queue_lock);
3489 if (!err) {
3490 q->rpm_status = RPM_ACTIVE;
3491 __blk_run_queue(q);
3492 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003493 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003494 } else {
3495 q->rpm_status = RPM_SUSPENDED;
3496 }
3497 spin_unlock_irq(q->queue_lock);
3498}
3499EXPORT_SYMBOL(blk_post_runtime_resume);
3500#endif
3501
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502int __init blk_dev_init(void)
3503{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003504 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05303505 FIELD_SIZEOF(struct request, cmd_flags));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003506
Tejun Heo89b90be2011-01-03 15:01:47 +01003507 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3508 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003509 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 if (!kblockd_workqueue)
3511 panic("Failed to create kblockd\n");
3512
3513 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003514 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515
Jens Axboe8324aa92008-01-29 14:51:59 +01003516 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003517 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 return 0;
3520}