blob: b60537b2c35b4152343c0239374d8ba332865b09 [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 */
Wei Tangd674d412015-11-24 09:58:45 +080054struct kmem_cache *request_cachep;
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/**
Jens Axboe21491412015-12-28 13:01:22 -0700210 * blk_start_queue_async - asynchronously restart a previously stopped queue
211 * @q: The &struct request_queue in question
212 *
213 * Description:
214 * blk_start_queue_async() will clear the stop flag on the queue, and
215 * ensure that the request_fn for the queue is run from an async
216 * context.
217 **/
218void blk_start_queue_async(struct request_queue *q)
219{
220 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
221 blk_run_queue_async(q);
222}
223EXPORT_SYMBOL(blk_start_queue_async);
224
225/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +0200227 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 *
229 * Description:
230 * blk_start_queue() will clear the stop flag on the queue, and call
231 * the request_fn for the queue if it was in a stopped state when
232 * entered. Also see blk_stop_queue(). Queue lock must be held.
233 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200234void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200236 WARN_ON(!irqs_disabled());
237
Nick Piggin75ad23b2008-04-29 14:48:33 +0200238 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200239 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241EXPORT_SYMBOL(blk_start_queue);
242
243/**
244 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +0200245 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 *
247 * Description:
248 * The Linux block layer assumes that a block driver will consume all
249 * entries on the request queue when the request_fn strategy is called.
250 * Often this will not happen, because of hardware limitations (queue
251 * depth settings). If a device driver gets a 'queue full' response,
252 * or if it simply chooses not to queue more I/O at one point, it can
253 * call this function to prevent the request_fn from being called until
254 * the driver has signalled it's ready to go again. This happens by calling
255 * blk_start_queue() to restart queue operations. Queue lock must be held.
256 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200257void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Tejun Heo136b5722012-08-21 13:18:24 -0700259 cancel_delayed_work(&q->delay_work);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200260 queue_flag_set(QUEUE_FLAG_STOPPED, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262EXPORT_SYMBOL(blk_stop_queue);
263
264/**
265 * blk_sync_queue - cancel any pending callbacks on a queue
266 * @q: the queue
267 *
268 * Description:
269 * The block layer may perform asynchronous callback activity
270 * on a queue, such as calling the unplug function after a timeout.
271 * A block device may call blk_sync_queue to ensure that any
272 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200273 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 * that its ->make_request_fn will not re-add plugging prior to calling
275 * this function.
276 *
Vivek Goyalda527772011-03-02 19:05:33 -0500277 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900278 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800279 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500280 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
282void blk_sync_queue(struct request_queue *q)
283{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100284 del_timer_sync(&q->timeout);
Ming Leif04c1fe2013-12-26 21:31:36 +0800285
286 if (q->mq_ops) {
287 struct blk_mq_hw_ctx *hctx;
288 int i;
289
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600290 queue_for_each_hw_ctx(q, hctx, i) {
291 cancel_delayed_work_sync(&hctx->run_work);
292 cancel_delayed_work_sync(&hctx->delay_work);
293 }
Ming Leif04c1fe2013-12-26 21:31:36 +0800294 } else {
295 cancel_delayed_work_sync(&q->delay_work);
296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298EXPORT_SYMBOL(blk_sync_queue);
299
300/**
Bart Van Asschec246e802012-12-06 14:32:01 +0100301 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
302 * @q: The queue to run
303 *
304 * Description:
305 * Invoke request handling on a queue if there are any pending requests.
306 * May be used to restart request handling after a request has completed.
307 * This variant runs the queue whether or not the queue has been
308 * stopped. Must be called with the queue lock held and interrupts
309 * disabled. See also @blk_run_queue.
310 */
311inline void __blk_run_queue_uncond(struct request_queue *q)
312{
313 if (unlikely(blk_queue_dead(q)))
314 return;
315
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100316 /*
317 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
318 * the queue lock internally. As a result multiple threads may be
319 * running such a request function concurrently. Keep track of the
320 * number of active request_fn invocations such that blk_drain_queue()
321 * can wait until all these request_fn calls have finished.
322 */
323 q->request_fn_active++;
Bart Van Asschec246e802012-12-06 14:32:01 +0100324 q->request_fn(q);
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100325 q->request_fn_active--;
Bart Van Asschec246e802012-12-06 14:32:01 +0100326}
Christoph Hellwiga7928c12015-04-17 22:37:20 +0200327EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
Bart Van Asschec246e802012-12-06 14:32:01 +0100328
329/**
Jens Axboe80a4b582008-10-14 09:51:06 +0200330 * __blk_run_queue - run a single device queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200332 *
333 * Description:
334 * See @blk_run_queue. This variant must be called with the queue lock
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200335 * held and interrupts disabled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 */
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200337void __blk_run_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
Tejun Heoa538cd02009-04-23 11:05:17 +0900339 if (unlikely(blk_queue_stopped(q)))
340 return;
341
Bart Van Asschec246e802012-12-06 14:32:01 +0100342 __blk_run_queue_uncond(q);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200343}
344EXPORT_SYMBOL(__blk_run_queue);
Jens Axboedac07ec2006-05-11 08:20:16 +0200345
Nick Piggin75ad23b2008-04-29 14:48:33 +0200346/**
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200347 * blk_run_queue_async - run a single device queue in workqueue context
348 * @q: The queue to run
349 *
350 * Description:
351 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
Bart Van Assche70460572012-11-28 13:45:56 +0100352 * of us. The caller must hold the queue lock.
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200353 */
354void blk_run_queue_async(struct request_queue *q)
355{
Bart Van Assche70460572012-11-28 13:45:56 +0100356 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
Tejun Heoe7c2f962012-08-21 13:18:24 -0700357 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200358}
Jens Axboec21e6be2011-04-19 13:32:46 +0200359EXPORT_SYMBOL(blk_run_queue_async);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200360
361/**
Nick Piggin75ad23b2008-04-29 14:48:33 +0200362 * blk_run_queue - run a single device queue
363 * @q: The queue to run
Jens Axboe80a4b582008-10-14 09:51:06 +0200364 *
365 * Description:
366 * Invoke request handling on this queue, if it has pending work to do.
Tejun Heoa7f55792009-04-23 11:05:17 +0900367 * May be used to restart queueing when a request has completed.
Nick Piggin75ad23b2008-04-29 14:48:33 +0200368 */
369void blk_run_queue(struct request_queue *q)
370{
371 unsigned long flags;
372
373 spin_lock_irqsave(q->queue_lock, flags);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +0200374 __blk_run_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 spin_unlock_irqrestore(q->queue_lock, flags);
376}
377EXPORT_SYMBOL(blk_run_queue);
378
Jens Axboe165125e2007-07-24 09:28:11 +0200379void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500380{
381 kobject_put(&q->kobj);
382}
Jens Axboed86e0e82011-05-27 07:44:43 +0200383EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500384
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200385/**
Bart Van Assche807592a2012-11-28 13:43:38 +0100386 * __blk_drain_queue - drain requests from request_queue
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200387 * @q: queue to drain
Tejun Heoc9a929d2011-10-19 14:42:16 +0200388 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200389 *
Tejun Heoc9a929d2011-10-19 14:42:16 +0200390 * Drain requests from @q. If @drain_all is set, all requests are drained.
391 * If not, only ELVPRIV requests are drained. The caller is responsible
392 * for ensuring that no new requests which need to be drained are queued.
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200393 */
Bart Van Assche807592a2012-11-28 13:43:38 +0100394static void __blk_drain_queue(struct request_queue *q, bool drain_all)
395 __releases(q->queue_lock)
396 __acquires(q->queue_lock)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200397{
Asias He458f27a2012-06-15 08:45:25 +0200398 int i;
399
Bart Van Assche807592a2012-11-28 13:43:38 +0100400 lockdep_assert_held(q->queue_lock);
401
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200402 while (true) {
Tejun Heo481a7d62011-12-14 00:33:37 +0100403 bool drain = false;
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200404
Tejun Heob855b042012-03-06 21:24:55 +0100405 /*
406 * The caller might be trying to drain @q before its
407 * elevator is initialized.
408 */
409 if (q->elevator)
410 elv_drain_elevator(q);
411
Tejun Heo5efd6112012-03-05 13:15:12 -0800412 blkcg_drain_queue(q);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200413
Tejun Heo4eabc942011-12-15 20:03:04 +0100414 /*
415 * This function might be called on a queue which failed
Tejun Heob855b042012-03-06 21:24:55 +0100416 * driver init after queue creation or is not yet fully
417 * active yet. Some drivers (e.g. fd and loop) get unhappy
418 * in such cases. Kick queue iff dispatch queue has
419 * something on it and @q has request_fn set.
Tejun Heo4eabc942011-12-15 20:03:04 +0100420 */
Tejun Heob855b042012-03-06 21:24:55 +0100421 if (!list_empty(&q->queue_head) && q->request_fn)
Tejun Heo4eabc942011-12-15 20:03:04 +0100422 __blk_run_queue(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200423
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700424 drain |= q->nr_rqs_elvpriv;
Bart Van Assche24faf6f2012-11-28 13:46:45 +0100425 drain |= q->request_fn_active;
Tejun Heo481a7d62011-12-14 00:33:37 +0100426
427 /*
428 * Unfortunately, requests are queued at and tracked from
429 * multiple places and there's no single counter which can
430 * be drained. Check all the queues and counters.
431 */
432 if (drain_all) {
Ming Leie97c2932014-09-25 23:23:46 +0800433 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
Tejun Heo481a7d62011-12-14 00:33:37 +0100434 drain |= !list_empty(&q->queue_head);
435 for (i = 0; i < 2; i++) {
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700436 drain |= q->nr_rqs[i];
Tejun Heo481a7d62011-12-14 00:33:37 +0100437 drain |= q->in_flight[i];
Ming Lei7c94e1c2014-09-25 23:23:43 +0800438 if (fq)
439 drain |= !list_empty(&fq->flush_queue[i]);
Tejun Heo481a7d62011-12-14 00:33:37 +0100440 }
441 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200442
Tejun Heo481a7d62011-12-14 00:33:37 +0100443 if (!drain)
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200444 break;
Bart Van Assche807592a2012-11-28 13:43:38 +0100445
446 spin_unlock_irq(q->queue_lock);
447
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200448 msleep(10);
Bart Van Assche807592a2012-11-28 13:43:38 +0100449
450 spin_lock_irq(q->queue_lock);
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200451 }
Asias He458f27a2012-06-15 08:45:25 +0200452
453 /*
454 * With queue marked dead, any woken up waiter will fail the
455 * allocation path, so the wakeup chaining is lost and we're
456 * left with hung waiters. We need to wake up those waiters.
457 */
458 if (q->request_fn) {
Tejun Heoa0516612012-06-26 15:05:44 -0700459 struct request_list *rl;
460
Tejun Heoa0516612012-06-26 15:05:44 -0700461 blk_queue_for_each_rl(rl, q)
462 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
463 wake_up_all(&rl->wait[i]);
Asias He458f27a2012-06-15 08:45:25 +0200464 }
Tejun Heoe3c78ca2011-10-19 14:32:38 +0200465}
466
Tejun Heoc9a929d2011-10-19 14:42:16 +0200467/**
Tejun Heod7325802012-03-05 13:14:58 -0800468 * blk_queue_bypass_start - enter queue bypass mode
469 * @q: queue of interest
470 *
471 * In bypass mode, only the dispatch FIFO queue of @q is used. This
472 * function makes @q enter bypass mode and drains all requests which were
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800473 * throttled or issued before. On return, it's guaranteed that no request
Tejun Heo80fd9972012-04-13 14:50:53 -0700474 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
475 * inside queue or RCU read lock.
Tejun Heod7325802012-03-05 13:14:58 -0800476 */
477void blk_queue_bypass_start(struct request_queue *q)
478{
479 spin_lock_irq(q->queue_lock);
Tejun Heo776687b2014-07-01 10:29:17 -0600480 q->bypass_depth++;
Tejun Heod7325802012-03-05 13:14:58 -0800481 queue_flag_set(QUEUE_FLAG_BYPASS, q);
482 spin_unlock_irq(q->queue_lock);
483
Tejun Heo776687b2014-07-01 10:29:17 -0600484 /*
485 * Queues start drained. Skip actual draining till init is
486 * complete. This avoids lenghty delays during queue init which
487 * can happen many times during boot.
488 */
489 if (blk_queue_init_done(q)) {
Bart Van Assche807592a2012-11-28 13:43:38 +0100490 spin_lock_irq(q->queue_lock);
491 __blk_drain_queue(q, false);
492 spin_unlock_irq(q->queue_lock);
493
Tejun Heob82d4b12012-04-13 13:11:31 -0700494 /* ensure blk_queue_bypass() is %true inside RCU read lock */
495 synchronize_rcu();
496 }
Tejun Heod7325802012-03-05 13:14:58 -0800497}
498EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
499
500/**
501 * blk_queue_bypass_end - leave queue bypass mode
502 * @q: queue of interest
503 *
504 * Leave bypass mode and restore the normal queueing behavior.
505 */
506void blk_queue_bypass_end(struct request_queue *q)
507{
508 spin_lock_irq(q->queue_lock);
509 if (!--q->bypass_depth)
510 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
511 WARN_ON_ONCE(q->bypass_depth < 0);
512 spin_unlock_irq(q->queue_lock);
513}
514EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
515
Jens Axboeaed3ea92014-12-22 14:04:42 -0700516void blk_set_queue_dying(struct request_queue *q)
517{
518 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
519
520 if (q->mq_ops)
521 blk_mq_wake_waiters(q);
522 else {
523 struct request_list *rl;
524
525 blk_queue_for_each_rl(rl, q) {
526 if (rl->rq_pool) {
527 wake_up(&rl->wait[BLK_RW_SYNC]);
528 wake_up(&rl->wait[BLK_RW_ASYNC]);
529 }
530 }
531 }
532}
533EXPORT_SYMBOL_GPL(blk_set_queue_dying);
534
Tejun Heod7325802012-03-05 13:14:58 -0800535/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200536 * blk_cleanup_queue - shutdown a request queue
537 * @q: request queue to shutdown
538 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100539 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
540 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500541 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100542void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500543{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200544 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700545
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100546 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500547 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700548 blk_set_queue_dying(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200549 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800550
Tejun Heo80fd9972012-04-13 14:50:53 -0700551 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100552 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700553 * that, unlike blk_queue_bypass_start(), we aren't performing
554 * synchronize_rcu() after entering bypass mode to avoid the delay
555 * as some drivers create and destroy a lot of queues while
556 * probing. This is still safe because blk_release_queue() will be
557 * called only after the queue refcnt drops to zero and nothing,
558 * RCU or not, would be traversing the queue by then.
559 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800560 q->bypass_depth++;
561 queue_flag_set(QUEUE_FLAG_BYPASS, q);
562
Tejun Heoc9a929d2011-10-19 14:42:16 +0200563 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
564 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100565 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200566 spin_unlock_irq(lock);
567 mutex_unlock(&q->sysfs_lock);
568
Bart Van Asschec246e802012-12-06 14:32:01 +0100569 /*
570 * Drain all requests queued before DYING marking. Set DEAD flag to
571 * prevent that q->request_fn() gets invoked after draining finished.
572 */
Dan Williams3ef28e82015-10-21 13:20:12 -0400573 blk_freeze_queue(q);
574 spin_lock_irq(lock);
575 if (!q->mq_ops)
Ming Lei43a5e4e2013-12-26 21:31:35 +0800576 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100577 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100578 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200579
Dan Williams5a48fc12015-10-21 13:20:23 -0400580 /* for synchronous bio-based driver finish in-flight integrity i/o */
581 blk_flush_integrity();
582
Tejun Heoc9a929d2011-10-19 14:42:16 +0200583 /* @q won't process any more request, flush async actions */
584 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
585 blk_sync_queue(q);
586
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100587 if (q->mq_ops)
588 blk_mq_free_queue(q);
Dan Williams3ef28e82015-10-21 13:20:12 -0400589 percpu_ref_exit(&q->q_usage_counter);
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100590
Asias He5e5cfac2012-05-24 23:28:52 +0800591 spin_lock_irq(lock);
592 if (q->queue_lock != &q->__queue_lock)
593 q->queue_lock = &q->__queue_lock;
594 spin_unlock_irq(lock);
595
Tejun Heob02176f2015-09-08 12:20:22 -0400596 bdi_unregister(&q->backing_dev_info);
NeilBrown6cd18e72015-04-27 14:12:22 +1000597
Tejun Heoc9a929d2011-10-19 14:42:16 +0200598 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500599 blk_put_queue(q);
600}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601EXPORT_SYMBOL(blk_cleanup_queue);
602
David Rientjes271508d2015-03-24 16:21:16 -0700603/* Allocate memory local to the request queue */
604static void *alloc_request_struct(gfp_t gfp_mask, void *data)
605{
606 int nid = (int)(long)data;
607 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
608}
609
610static void free_request_struct(void *element, void *unused)
611{
612 kmem_cache_free(request_cachep, element);
613}
614
Tejun Heo5b788ce2012-06-04 20:40:59 -0700615int blk_init_rl(struct request_list *rl, struct request_queue *q,
616 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400618 if (unlikely(rl->rq_pool))
619 return 0;
620
Tejun Heo5b788ce2012-06-04 20:40:59 -0700621 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200622 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
623 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200624 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
625 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
David Rientjes271508d2015-03-24 16:21:16 -0700627 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
628 free_request_struct,
629 (void *)(long)q->node, gfp_mask,
630 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (!rl->rq_pool)
632 return -ENOMEM;
633
634 return 0;
635}
636
Tejun Heo5b788ce2012-06-04 20:40:59 -0700637void blk_exit_rl(struct request_list *rl)
638{
639 if (rl->rq_pool)
640 mempool_destroy(rl->rq_pool);
641}
642
Jens Axboe165125e2007-07-24 09:28:11 +0200643struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100645 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700646}
647EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100649int blk_queue_enter(struct request_queue *q, bool nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400650{
651 while (true) {
652 int ret;
653
654 if (percpu_ref_tryget_live(&q->q_usage_counter))
655 return 0;
656
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100657 if (nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400658 return -EBUSY;
659
660 ret = wait_event_interruptible(q->mq_freeze_wq,
661 !atomic_read(&q->mq_freeze_depth) ||
662 blk_queue_dying(q));
663 if (blk_queue_dying(q))
664 return -ENODEV;
665 if (ret)
666 return ret;
667 }
668}
669
670void blk_queue_exit(struct request_queue *q)
671{
672 percpu_ref_put(&q->q_usage_counter);
673}
674
675static void blk_queue_usage_counter_release(struct percpu_ref *ref)
676{
677 struct request_queue *q =
678 container_of(ref, struct request_queue, q_usage_counter);
679
680 wake_up_all(&q->mq_freeze_wq);
681}
682
Christoph Hellwig287922e2015-10-30 20:57:30 +0800683static void blk_rq_timed_out_timer(unsigned long data)
684{
685 struct request_queue *q = (struct request_queue *)data;
686
687 kblockd_schedule_work(&q->timeout_work);
688}
689
Jens Axboe165125e2007-07-24 09:28:11 +0200690struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700691{
Jens Axboe165125e2007-07-24 09:28:11 +0200692 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700693 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700694
Jens Axboe8324aa92008-01-29 14:51:59 +0100695 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700696 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (!q)
698 return NULL;
699
Dan Carpenter00380a42012-03-23 09:58:54 +0100700 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100701 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800702 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100703
Kent Overstreet54efd502015-04-23 22:37:18 -0700704 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
705 if (!q->bio_split)
706 goto fail_id;
707
Jens Axboe0989a022009-06-12 14:42:56 +0200708 q->backing_dev_info.ra_pages =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300709 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
Tejun Heo89e9b9e2015-05-22 17:13:36 -0400710 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
Jens Axboed9938312009-06-12 14:45:52 +0200711 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100712 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200713
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700714 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100715 if (err)
Kent Overstreet54efd502015-04-23 22:37:18 -0700716 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700717
Matthew Garrett31373d02010-04-06 14:25:14 +0200718 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
719 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700720 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100721 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700722 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100723 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800724#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800725 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800726#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500727 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500728
Jens Axboe8324aa92008-01-29 14:51:59 +0100729 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Al Viro483f4af2006-03-18 18:34:37 -0500731 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700732 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500733
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500734 /*
735 * By default initialize queue_lock to internal lock and driver can
736 * override it later if need be.
737 */
738 q->queue_lock = &q->__queue_lock;
739
Tejun Heob82d4b12012-04-13 13:11:31 -0700740 /*
741 * A queue starts its life with bypass turned on to avoid
742 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700743 * init. The initial bypass will be finished when the queue is
744 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700745 */
746 q->bypass_depth = 1;
747 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
748
Jens Axboe320ae512013-10-24 09:20:05 +0100749 init_waitqueue_head(&q->mq_freeze_wq);
750
Dan Williams3ef28e82015-10-21 13:20:12 -0400751 /*
752 * Init percpu_ref in atomic mode so that it's faster to shutdown.
753 * See blk_register_queue() for details.
754 */
755 if (percpu_ref_init(&q->q_usage_counter,
756 blk_queue_usage_counter_release,
757 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
Mikulas Patockafff49962013-10-14 12:11:36 -0400758 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800759
Dan Williams3ef28e82015-10-21 13:20:12 -0400760 if (blkcg_init_queue(q))
761 goto fail_ref;
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100764
Dan Williams3ef28e82015-10-21 13:20:12 -0400765fail_ref:
766 percpu_ref_exit(&q->q_usage_counter);
Mikulas Patockafff49962013-10-14 12:11:36 -0400767fail_bdi:
768 bdi_destroy(&q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700769fail_split:
770 bioset_free(q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100771fail_id:
772 ida_simple_remove(&blk_queue_ida, q->id);
773fail_q:
774 kmem_cache_free(blk_requestq_cachep, q);
775 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
Christoph Lameter19460892005-06-23 00:08:19 -0700777EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779/**
780 * blk_init_queue - prepare a request queue for use with a block device
781 * @rfn: The function to be called to process requests that have been
782 * placed on the queue.
783 * @lock: Request queue spin lock
784 *
785 * Description:
786 * If a block device wishes to use the standard request handling procedures,
787 * which sorts requests and coalesces adjacent requests, then it must
788 * call blk_init_queue(). The function @rfn will be called when there
789 * are requests on the queue that need to be processed. If the device
790 * supports plugging, then @rfn may not be called immediately when requests
791 * are available on the queue, but may be called at some time later instead.
792 * Plugged queues are generally unplugged when a buffer belonging to one
793 * of the requests on the queue is needed, or due to memory pressure.
794 *
795 * @rfn is not required, or even expected, to remove all requests off the
796 * queue, but only as many as it can handle at a time. If it does leave
797 * requests on the queue, it is responsible for arranging that the requests
798 * get dealt with eventually.
799 *
800 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200801 * request queue; this lock will be taken also from interrupt context, so irq
802 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200804 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 * it didn't succeed.
806 *
807 * Note:
808 * blk_init_queue() must be paired with a blk_cleanup_queue() call
809 * when the block device is deactivated (such as at module unload).
810 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700811
Jens Axboe165125e2007-07-24 09:28:11 +0200812struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100814 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700815}
816EXPORT_SYMBOL(blk_init_queue);
817
Jens Axboe165125e2007-07-24 09:28:11 +0200818struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700819blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
820{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600821 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600823 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
824 if (!uninit_q)
825 return NULL;
826
Mike Snitzer51514122011-11-23 10:59:13 +0100827 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600828 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700829 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700830
Mike Snitzer7982e902014-03-08 17:20:01 -0700831 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200832}
833EXPORT_SYMBOL(blk_init_queue_node);
834
Jens Axboedece1632015-11-05 10:41:16 -0700835static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
Mike Snitzer336b7e12015-05-11 14:06:32 -0400836
Mike Snitzer01effb02010-05-11 08:57:42 +0200837struct request_queue *
838blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
839 spinlock_t *lock)
840{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 if (!q)
842 return NULL;
843
Ming Leif70ced02014-09-25 23:23:47 +0800844 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800845 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700846 return NULL;
847
Tejun Heoa0516612012-06-26 15:05:44 -0700848 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600849 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Christoph Hellwig287922e2015-10-30 20:57:30 +0800851 INIT_WORK(&q->timeout_work, blk_timeout_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900854 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700855 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500856
857 /* Override internal queue lock with supplied lock pointer */
858 if (lock)
859 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Jens Axboef3b144a2009-03-06 08:48:33 +0100861 /*
862 * This also sets hw/phys segments, boundary and size
863 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200864 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Alan Stern44ec9542007-02-20 11:01:57 -0500866 q->sg_reserved_size = INT_MAX;
867
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600868 /* Protect q->elevator from elevator_change */
869 mutex_lock(&q->sysfs_lock);
870
Tejun Heob82d4b12012-04-13 13:11:31 -0700871 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600872 if (elevator_init(q, NULL)) {
873 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600874 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600875 }
876
877 mutex_unlock(&q->sysfs_lock);
878
Tejun Heob82d4b12012-04-13 13:11:31 -0700879 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600880
881fail:
Ming Leiba483382014-09-25 23:23:44 +0800882 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600883 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
Mike Snitzer51514122011-11-23 10:59:13 +0100885EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Tejun Heo09ac46c2011-12-14 00:33:38 +0100887bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100889 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100890 __blk_get_queue(q);
891 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
893
Tejun Heo09ac46c2011-12-14 00:33:38 +0100894 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
Jens Axboed86e0e82011-05-27 07:44:43 +0200896EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Tejun Heo5b788ce2012-06-04 20:40:59 -0700898static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100900 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700901 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100902 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100903 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100904 }
905
Tejun Heo5b788ce2012-06-04 20:40:59 -0700906 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909/*
910 * ioc_batching returns true if the ioc is a valid batching request and
911 * should be given priority access to a request.
912 */
Jens Axboe165125e2007-07-24 09:28:11 +0200913static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
915 if (!ioc)
916 return 0;
917
918 /*
919 * Make sure the process is able to allocate at least 1 request
920 * even if the batch times out, otherwise we could theoretically
921 * lose wakeups.
922 */
923 return ioc->nr_batch_requests == q->nr_batching ||
924 (ioc->nr_batch_requests > 0
925 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
926}
927
928/*
929 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
930 * will cause the process to be a "batcher" on all queues in the system. This
931 * is the behaviour we want though - once it gets a wakeup it should be given
932 * a nice run.
933 */
Jens Axboe165125e2007-07-24 09:28:11 +0200934static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
936 if (!ioc || ioc_batching(q, ioc))
937 return;
938
939 ioc->nr_batch_requests = q->nr_batching;
940 ioc->last_waited = jiffies;
941}
942
Tejun Heo5b788ce2012-06-04 20:40:59 -0700943static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700945 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Tejun Heod40f75a2015-05-22 17:13:42 -0400947 if (rl->count[sync] < queue_congestion_off_threshold(q))
948 blk_clear_congested(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Jens Axboe1faa16d2009-04-06 14:48:01 +0200950 if (rl->count[sync] + 1 <= q->nr_requests) {
951 if (waitqueue_active(&rl->wait[sync]))
952 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Tejun Heo5b788ce2012-06-04 20:40:59 -0700954 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956}
957
958/*
959 * A request has just been released. Account for it, update the full and
960 * congestion status, wake up any waiters. Called under q->queue_lock.
961 */
Tejun Heo5b788ce2012-06-04 20:40:59 -0700962static void freed_request(struct request_list *rl, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700964 struct request_queue *q = rl->q;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200965 int sync = rw_is_sync(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700967 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200968 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200969 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700970 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Tejun Heo5b788ce2012-06-04 20:40:59 -0700972 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Jens Axboe1faa16d2009-04-06 14:48:01 +0200974 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700975 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976}
977
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600978int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
979{
980 struct request_list *rl;
Tejun Heod40f75a2015-05-22 17:13:42 -0400981 int on_thresh, off_thresh;
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600982
983 spin_lock_irq(q->queue_lock);
984 q->nr_requests = nr;
985 blk_queue_congestion_threshold(q);
Tejun Heod40f75a2015-05-22 17:13:42 -0400986 on_thresh = queue_congestion_on_threshold(q);
987 off_thresh = queue_congestion_off_threshold(q);
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600988
989 blk_queue_for_each_rl(rl, q) {
Tejun Heod40f75a2015-05-22 17:13:42 -0400990 if (rl->count[BLK_RW_SYNC] >= on_thresh)
991 blk_set_congested(rl, BLK_RW_SYNC);
992 else if (rl->count[BLK_RW_SYNC] < off_thresh)
993 blk_clear_congested(rl, BLK_RW_SYNC);
994
995 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
996 blk_set_congested(rl, BLK_RW_ASYNC);
997 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
998 blk_clear_congested(rl, BLK_RW_ASYNC);
999
Jens Axboee3a2b3f2014-05-20 11:49:02 -06001000 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1001 blk_set_rl_full(rl, BLK_RW_SYNC);
1002 } else {
1003 blk_clear_rl_full(rl, BLK_RW_SYNC);
1004 wake_up(&rl->wait[BLK_RW_SYNC]);
1005 }
1006
1007 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1008 blk_set_rl_full(rl, BLK_RW_ASYNC);
1009 } else {
1010 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1011 wake_up(&rl->wait[BLK_RW_ASYNC]);
1012 }
1013 }
1014
1015 spin_unlock_irq(q->queue_lock);
1016 return 0;
1017}
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001020 * Determine if elevator data should be initialized when allocating the
1021 * request associated with @bio.
1022 */
1023static bool blk_rq_should_init_elevator(struct bio *bio)
1024{
1025 if (!bio)
1026 return true;
1027
1028 /*
1029 * Flush requests do not use the elevator so skip initialization.
1030 * This allows a request to share the flush and elevator data.
1031 */
1032 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1033 return false;
1034
1035 return true;
1036}
1037
Tejun Heoda8303c2011-10-19 14:33:05 +02001038/**
Tejun Heo852c7882012-03-05 13:15:27 -08001039 * rq_ioc - determine io_context for request allocation
1040 * @bio: request being allocated is for this bio (can be %NULL)
1041 *
1042 * Determine io_context to use for request allocation for @bio. May return
1043 * %NULL if %current->io_context doesn't exist.
1044 */
1045static struct io_context *rq_ioc(struct bio *bio)
1046{
1047#ifdef CONFIG_BLK_CGROUP
1048 if (bio && bio->bi_ioc)
1049 return bio->bi_ioc;
1050#endif
1051 return current->io_context;
1052}
1053
1054/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001055 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -07001056 * @rl: request list to allocate from
Tejun Heoda8303c2011-10-19 14:33:05 +02001057 * @rw_flags: RW and SYNC flags
1058 * @bio: bio to allocate request for (can be %NULL)
1059 * @gfp_mask: allocation mask
1060 *
1061 * Get a free request from @q. This function may fail under memory
1062 * pressure or if @q is dead.
1063 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001064 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001065 * Returns ERR_PTR on failure, with @q->queue_lock held.
1066 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001068static struct request *__get_request(struct request_list *rl, int rw_flags,
Tejun Heoa06e05e2012-06-04 20:40:55 -07001069 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Tejun Heo5b788ce2012-06-04 20:40:59 -07001071 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -08001072 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001073 struct elevator_type *et = q->elevator->type;
1074 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +01001075 struct io_cq *icq = NULL;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001076 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +02001077 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001079 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -06001080 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +02001081
Jens Axboe7749a8d2006-12-13 13:02:26 +01001082 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001083 if (may_queue == ELV_MQUEUE_NO)
1084 goto rq_starved;
1085
Jens Axboe1faa16d2009-04-06 14:48:01 +02001086 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1087 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +01001088 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001089 * The queue will fill after this allocation, so set
1090 * it as full, and mark this process as "batching".
1091 * This process will be allowed to complete a batch of
1092 * requests, others will be blocked.
1093 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001094 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001095 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001096 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001097 } else {
1098 if (may_queue != ELV_MQUEUE_MUST
1099 && !ioc_batching(q, ioc)) {
1100 /*
1101 * The queue is full and the allocating
1102 * process is not a "batcher", and not
1103 * exempted by the IO scheduler
1104 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001105 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001106 }
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 }
Tejun Heod40f75a2015-05-22 17:13:42 -04001109 blk_set_congested(rl, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
1111
Jens Axboe082cf692005-06-28 16:35:11 +02001112 /*
1113 * Only allow batching queuers to allocate up to 50% over the defined
1114 * limit of requests, otherwise we could have thousands of requests
1115 * allocated with any setting of ->nr_requests
1116 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001117 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001118 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001119
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001120 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001121 rl->count[is_sync]++;
1122 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001123
Tejun Heof1f8cc92011-12-14 00:33:42 +01001124 /*
1125 * Decide whether the new request will be managed by elevator. If
1126 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1127 * prevent the current elevator from being destroyed until the new
1128 * request is freed. This guarantees icq's won't be destroyed and
1129 * makes creating new ones safe.
1130 *
1131 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1132 * it will be created after releasing queue_lock.
1133 */
Tejun Heod7325802012-03-05 13:14:58 -08001134 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001135 rw_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001136 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001137 if (et->icq_cache && ioc)
1138 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001139 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001140
Jens Axboef253b862010-10-24 22:06:02 +02001141 if (blk_queue_io_stat(q))
1142 rw_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 spin_unlock_irq(q->queue_lock);
1144
Tejun Heo29e2b092012-04-19 16:29:21 -07001145 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001146 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001147 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001148 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Tejun Heo29e2b092012-04-19 16:29:21 -07001150 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001151 blk_rq_set_rl(rq, rl);
Tejun Heo29e2b092012-04-19 16:29:21 -07001152 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1153
Tejun Heoaaf7c682012-04-19 16:29:22 -07001154 /* init elvpriv */
Tejun Heo29e2b092012-04-19 16:29:21 -07001155 if (rw_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001156 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001157 if (ioc)
1158 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001159 if (!icq)
1160 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001161 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001162
1163 rq->elv.icq = icq;
1164 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1165 goto fail_elvpriv;
1166
1167 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001168 if (icq)
1169 get_io_context(icq->ioc);
1170 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001171out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001172 /*
1173 * ioc may be NULL here, and ioc_batching will be false. That's
1174 * OK, if the queue is under the request limit then requests need
1175 * not count toward the nr_batch_requests limit. There will always
1176 * be some limit enforced by BLK_BATCH_TIME.
1177 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 if (ioc_batching(q, ioc))
1179 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001180
Jens Axboe1faa16d2009-04-06 14:48:01 +02001181 trace_block_getrq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001183
Tejun Heoaaf7c682012-04-19 16:29:22 -07001184fail_elvpriv:
1185 /*
1186 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1187 * and may fail indefinitely under memory pressure and thus
1188 * shouldn't stall IO. Treat this request as !elvpriv. This will
1189 * disturb iosched and blkcg but weird is bettern than dead.
1190 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001191 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1192 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001193
1194 rq->cmd_flags &= ~REQ_ELVPRIV;
1195 rq->elv.icq = NULL;
1196
1197 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001198 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001199 spin_unlock_irq(q->queue_lock);
1200 goto out;
1201
Tejun Heob6792812012-03-05 13:15:23 -08001202fail_alloc:
1203 /*
1204 * Allocation failed presumably due to memory. Undo anything we
1205 * might have messed up.
1206 *
1207 * Allocating task should really be put onto the front of the wait
1208 * queue, but this is pretty rare.
1209 */
1210 spin_lock_irq(q->queue_lock);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001211 freed_request(rl, rw_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001212
1213 /*
1214 * in the very unlikely event that allocation failed and no
1215 * requests for this direction was pending, mark us starved so that
1216 * freeing of a request in the other direction will notice
1217 * us. another possible fix would be to split the rq mempool into
1218 * READ and WRITE
1219 */
1220rq_starved:
1221 if (unlikely(rl->count[is_sync] == 0))
1222 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001223 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
Tejun Heoda8303c2011-10-19 14:33:05 +02001226/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001227 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001228 * @q: request_queue to allocate request from
1229 * @rw_flags: RW and SYNC flags
1230 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001231 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001232 *
Mel Gormand0164ad2015-11-06 16:28:21 -08001233 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1234 * this function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001235 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001236 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001237 * Returns ERR_PTR on failure, with @q->queue_lock held.
1238 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001240static struct request *get_request(struct request_queue *q, int rw_flags,
1241 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001243 const bool is_sync = rw_is_sync(rw_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001244 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001245 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001247
1248 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001249retry:
Tejun Heoa0516612012-06-26 15:05:44 -07001250 rq = __get_request(rl, rw_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001251 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001252 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Mel Gormand0164ad2015-11-06 16:28:21 -08001254 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001255 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001256 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Tejun Heoa06e05e2012-06-04 20:40:55 -07001259 /* wait on @rl and retry */
1260 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1261 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001262
Tejun Heoa06e05e2012-06-04 20:40:55 -07001263 trace_block_sleeprq(q, bio, rw_flags & 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Tejun Heoa06e05e2012-06-04 20:40:55 -07001265 spin_unlock_irq(q->queue_lock);
1266 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Tejun Heoa06e05e2012-06-04 20:40:55 -07001268 /*
1269 * After sleeping, we become a "batching" process and will be able
1270 * to allocate at least one request, and up to a big batch of them
1271 * for a small period time. See ioc_batching, ioc_set_batching
1272 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001273 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Tejun Heoa06e05e2012-06-04 20:40:55 -07001275 spin_lock_irq(q->queue_lock);
1276 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001277
Tejun Heoa06e05e2012-06-04 20:40:55 -07001278 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
Jens Axboe320ae512013-10-24 09:20:05 +01001281static struct request *blk_old_get_request(struct request_queue *q, int rw,
1282 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283{
1284 struct request *rq;
1285
1286 BUG_ON(rw != READ && rw != WRITE);
1287
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001288 /* create ioc upfront */
1289 create_io_context(gfp_mask, q->node);
1290
Nick Piggind6344532005-06-28 20:45:14 -07001291 spin_lock_irq(q->queue_lock);
Tejun Heoa06e05e2012-06-04 20:40:55 -07001292 rq = get_request(q, rw, NULL, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001293 if (IS_ERR(rq))
Tejun Heoda8303c2011-10-19 14:33:05 +02001294 spin_unlock_irq(q->queue_lock);
Nick Piggind6344532005-06-28 20:45:14 -07001295 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 return rq;
1298}
Jens Axboe320ae512013-10-24 09:20:05 +01001299
1300struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1301{
1302 if (q->mq_ops)
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01001303 return blk_mq_alloc_request(q, rw,
1304 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1305 0 : BLK_MQ_REQ_NOWAIT);
Jens Axboe320ae512013-10-24 09:20:05 +01001306 else
1307 return blk_old_get_request(q, rw, gfp_mask);
1308}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309EXPORT_SYMBOL(blk_get_request);
1310
1311/**
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001312 * blk_make_request - given a bio, allocate a corresponding struct request.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001313 * @q: target request queue
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001314 * @bio: The bio describing the memory mappings that will be submitted for IO.
1315 * It may be a chained-bio properly constructed by block/bio layer.
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001316 * @gfp_mask: gfp flags to be used for memory allocation
Jens Axboedc72ef42006-07-20 14:54:05 +02001317 *
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001318 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1319 * type commands. Where the struct request needs to be farther initialized by
1320 * the caller. It is passed a &struct bio, which describes the memory info of
1321 * the I/O transfer.
1322 *
1323 * The caller of blk_make_request must make sure that bi_io_vec
1324 * are set to describe the memory buffers. That bio_data_dir() will return
1325 * the needed direction of the request. (And all bio's in the passed bio-chain
1326 * are properly set accordingly)
1327 *
1328 * If called under none-sleepable conditions, mapped bio buffers must not
1329 * need bouncing, by calling the appropriate masked or flagged allocator,
1330 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1331 * BUG.
Jens Axboe53674ac2009-05-19 19:52:35 +02001332 *
1333 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
Mel Gormand0164ad2015-11-06 16:28:21 -08001334 * given to how you allocate bios. In particular, you cannot use
1335 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1336 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1337 * thus resulting in a deadlock. Alternatively bios should be allocated using
1338 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
Jens Axboe53674ac2009-05-19 19:52:35 +02001339 * If possible a big IO should be split into smaller parts when allocation
1340 * fails. Partial allocation should not be an error, or you risk a live-lock.
Jens Axboedc72ef42006-07-20 14:54:05 +02001341 */
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001342struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1343 gfp_t gfp_mask)
Jens Axboedc72ef42006-07-20 14:54:05 +02001344{
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001345 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1346
Joe Lawrencea492f072014-08-28 08:15:21 -06001347 if (IS_ERR(rq))
1348 return rq;
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001349
Jens Axboef27b0872014-06-06 07:57:37 -06001350 blk_rq_set_block_pc(rq);
1351
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001352 for_each_bio(bio) {
1353 struct bio *bounce_bio = bio;
1354 int ret;
1355
1356 blk_queue_bounce(q, &bounce_bio);
1357 ret = blk_rq_append_bio(q, rq, bounce_bio);
1358 if (unlikely(ret)) {
1359 blk_put_request(rq);
1360 return ERR_PTR(ret);
1361 }
1362 }
1363
1364 return rq;
Jens Axboedc72ef42006-07-20 14:54:05 +02001365}
Boaz Harrosh79eb63e2009-05-17 18:57:15 +03001366EXPORT_SYMBOL(blk_make_request);
Jens Axboedc72ef42006-07-20 14:54:05 +02001367
1368/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001369 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001370 * @rq: request to be initialized
1371 *
1372 */
1373void blk_rq_set_block_pc(struct request *rq)
1374{
1375 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1376 rq->__data_len = 0;
1377 rq->__sector = (sector_t) -1;
1378 rq->bio = rq->biotail = NULL;
1379 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001380}
1381EXPORT_SYMBOL(blk_rq_set_block_pc);
1382
1383/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * blk_requeue_request - put a request back on queue
1385 * @q: request queue where request should be inserted
1386 * @rq: request to be inserted
1387 *
1388 * Description:
1389 * Drivers often keep queueing requests until the hardware cannot accept
1390 * more, when that condition happens we need to put the request back
1391 * on the queue. Must be called with queue lock held.
1392 */
Jens Axboe165125e2007-07-24 09:28:11 +02001393void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001395 blk_delete_timer(rq);
1396 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001397 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001398
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001399 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 blk_queue_end_tag(q, rq);
1401
James Bottomleyba396a62009-05-27 14:17:08 +02001402 BUG_ON(blk_queued_rq(rq));
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 elv_requeue_request(q, rq);
1405}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406EXPORT_SYMBOL(blk_requeue_request);
1407
Jens Axboe73c10102011-03-08 13:19:51 +01001408static void add_acct_request(struct request_queue *q, struct request *rq,
1409 int where)
1410{
Jens Axboe320ae512013-10-24 09:20:05 +01001411 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001412 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001413}
1414
Tejun Heo074a7ac2008-08-25 19:56:14 +09001415static void part_round_stats_single(int cpu, struct hd_struct *part,
1416 unsigned long now)
1417{
Jens Axboe7276d022014-05-09 15:48:23 -06001418 int inflight;
1419
Tejun Heo074a7ac2008-08-25 19:56:14 +09001420 if (now == part->stamp)
1421 return;
1422
Jens Axboe7276d022014-05-09 15:48:23 -06001423 inflight = part_in_flight(part);
1424 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001425 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001426 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001427 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1428 }
1429 part->stamp = now;
1430}
1431
1432/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001433 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1434 * @cpu: cpu number for stats access
1435 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 *
1437 * The average IO queue length and utilisation statistics are maintained
1438 * by observing the current state of the queue length and the amount of
1439 * time it has been in this state for.
1440 *
1441 * Normally, that accounting is done on IO completion, but that can result
1442 * in more than a second's worth of IO being accounted for within any one
1443 * second, leading to >100% utilisation. To deal with that, we call this
1444 * function to do a round-off before returning the results when reading
1445 * /proc/diskstats. This accounts immediately for all queue usage up to
1446 * the current jiffies and restarts the counters again.
1447 */
Tejun Heoc9959052008-08-25 19:47:21 +09001448void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001449{
1450 unsigned long now = jiffies;
1451
Tejun Heo074a7ac2008-08-25 19:56:14 +09001452 if (part->partno)
1453 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1454 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001455}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001456EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001457
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001458#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001459static void blk_pm_put_request(struct request *rq)
1460{
1461 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1462 pm_runtime_mark_last_busy(rq->q->dev);
1463}
1464#else
1465static inline void blk_pm_put_request(struct request *rq) {}
1466#endif
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468/*
1469 * queue lock must be held
1470 */
Jens Axboe165125e2007-07-24 09:28:11 +02001471void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (unlikely(!q))
1474 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001476 if (q->mq_ops) {
1477 blk_mq_free_request(req);
1478 return;
1479 }
1480
Lin Mingc8158812013-03-23 11:42:27 +08001481 blk_pm_put_request(req);
1482
Tejun Heo8922e162005-10-20 16:23:44 +02001483 elv_completed_request(q, req);
1484
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001485 /* this is a bio leak */
1486 WARN_ON(req->bio != NULL);
1487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 /*
1489 * Request may not have originated from ll_rw_blk. if not,
1490 * it didn't come out of our reserved rq pools
1491 */
Jens Axboe49171e52006-08-10 08:59:11 +02001492 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001493 unsigned int flags = req->cmd_flags;
Tejun Heoa0516612012-06-26 15:05:44 -07001494 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001497 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
Tejun Heoa0516612012-06-26 15:05:44 -07001499 blk_free_request(rl, req);
1500 freed_request(rl, flags);
1501 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 }
1503}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001504EXPORT_SYMBOL_GPL(__blk_put_request);
1505
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506void blk_put_request(struct request *req)
1507{
Jens Axboe165125e2007-07-24 09:28:11 +02001508 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Jens Axboe320ae512013-10-24 09:20:05 +01001510 if (q->mq_ops)
1511 blk_mq_free_request(req);
1512 else {
1513 unsigned long flags;
1514
1515 spin_lock_irqsave(q->queue_lock, flags);
1516 __blk_put_request(q, req);
1517 spin_unlock_irqrestore(q->queue_lock, flags);
1518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520EXPORT_SYMBOL(blk_put_request);
1521
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001522/**
1523 * blk_add_request_payload - add a payload to a request
1524 * @rq: request to update
1525 * @page: page backing the payload
1526 * @len: length of the payload.
1527 *
1528 * This allows to later add a payload to an already submitted request by
1529 * a block driver. The driver needs to take care of freeing the payload
1530 * itself.
1531 *
1532 * Note that this is a quite horrible hack and nothing but handling of
1533 * discard requests should ever use it.
1534 */
1535void blk_add_request_payload(struct request *rq, struct page *page,
1536 unsigned int len)
1537{
1538 struct bio *bio = rq->bio;
1539
1540 bio->bi_io_vec->bv_page = page;
1541 bio->bi_io_vec->bv_offset = 0;
1542 bio->bi_io_vec->bv_len = len;
1543
Kent Overstreet4f024f32013-10-11 15:44:27 -07001544 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001545 bio->bi_vcnt = 1;
1546 bio->bi_phys_segments = 1;
1547
1548 rq->__data_len = rq->resid_len = len;
1549 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001550}
1551EXPORT_SYMBOL_GPL(blk_add_request_payload);
1552
Jens Axboe320ae512013-10-24 09:20:05 +01001553bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1554 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001555{
1556 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1557
Jens Axboe73c10102011-03-08 13:19:51 +01001558 if (!ll_back_merge_fn(q, req, bio))
1559 return false;
1560
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001561 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001562
1563 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1564 blk_rq_set_mixed_merge(req);
1565
1566 req->biotail->bi_next = bio;
1567 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001568 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001569 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1570
Jens Axboe320ae512013-10-24 09:20:05 +01001571 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001572 return true;
1573}
1574
Jens Axboe320ae512013-10-24 09:20:05 +01001575bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1576 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001577{
1578 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001579
Jens Axboe73c10102011-03-08 13:19:51 +01001580 if (!ll_front_merge_fn(q, req, bio))
1581 return false;
1582
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001583 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001584
1585 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1586 blk_rq_set_mixed_merge(req);
1587
Jens Axboe73c10102011-03-08 13:19:51 +01001588 bio->bi_next = req->bio;
1589 req->bio = bio;
1590
Kent Overstreet4f024f32013-10-11 15:44:27 -07001591 req->__sector = bio->bi_iter.bi_sector;
1592 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001593 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1594
Jens Axboe320ae512013-10-24 09:20:05 +01001595 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001596 return true;
1597}
1598
Tejun Heobd87b582011-10-19 14:33:08 +02001599/**
Jens Axboe320ae512013-10-24 09:20:05 +01001600 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001601 * @q: request_queue new bio is being queued at
1602 * @bio: new bio being queued
1603 * @request_count: out parameter for number of traversed plugged requests
Randy Dunlapccc26002015-10-30 18:36:16 -07001604 * @same_queue_rq: pointer to &struct request that gets filled in when
1605 * another request associated with @q is found on the plug list
1606 * (optional, may be %NULL)
Tejun Heobd87b582011-10-19 14:33:08 +02001607 *
1608 * Determine whether @bio being queued on @q can be merged with a request
1609 * on %current's plugged list. Returns %true if merge was successful,
1610 * otherwise %false.
1611 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001612 * Plugging coalesces IOs from the same issuer for the same purpose without
1613 * going through @q->queue_lock. As such it's more of an issuing mechanism
1614 * than scheduling, and the request, while may have elvpriv data, is not
1615 * added on the elevator at this point. In addition, we don't have
1616 * reliable access to the elevator outside queue lock. Only check basic
1617 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001618 *
1619 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001620 */
Jens Axboe320ae512013-10-24 09:20:05 +01001621bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Shaohua Li5b3f3412015-05-08 10:51:33 -07001622 unsigned int *request_count,
1623 struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +01001624{
1625 struct blk_plug *plug;
1626 struct request *rq;
1627 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001628 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001629
Tejun Heobd87b582011-10-19 14:33:08 +02001630 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001631 if (!plug)
1632 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001633 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001634
Shaohua Li92f399c2013-10-29 12:01:03 -06001635 if (q->mq_ops)
1636 plug_list = &plug->mq_list;
1637 else
1638 plug_list = &plug->list;
1639
1640 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001641 int el_ret;
1642
Shaohua Li5b3f3412015-05-08 10:51:33 -07001643 if (rq->q == q) {
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001644 (*request_count)++;
Shaohua Li5b3f3412015-05-08 10:51:33 -07001645 /*
1646 * Only blk-mq multiple hardware queues case checks the
1647 * rq in the same queue, there should be only one such
1648 * rq in a queue
1649 **/
1650 if (same_queue_rq)
1651 *same_queue_rq = rq;
1652 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001653
Tejun Heo07c2bd32012-02-08 09:19:42 +01001654 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001655 continue;
1656
Tejun Heo050c8ea2012-02-08 09:19:38 +01001657 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001658 if (el_ret == ELEVATOR_BACK_MERGE) {
1659 ret = bio_attempt_back_merge(q, rq, bio);
1660 if (ret)
1661 break;
1662 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1663 ret = bio_attempt_front_merge(q, rq, bio);
1664 if (ret)
1665 break;
1666 }
1667 }
1668out:
1669 return ret;
1670}
1671
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001672unsigned int blk_plug_queued_count(struct request_queue *q)
1673{
1674 struct blk_plug *plug;
1675 struct request *rq;
1676 struct list_head *plug_list;
1677 unsigned int ret = 0;
1678
1679 plug = current->plug;
1680 if (!plug)
1681 goto out;
1682
1683 if (q->mq_ops)
1684 plug_list = &plug->mq_list;
1685 else
1686 plug_list = &plug->list;
1687
1688 list_for_each_entry(rq, plug_list, queuelist) {
1689 if (rq->q == q)
1690 ret++;
1691 }
1692out:
1693 return ret;
1694}
1695
Jens Axboe86db1e22008-01-29 14:53:40 +01001696void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001697{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001698 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001699
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001700 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1701 if (bio->bi_rw & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001702 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001703
Tejun Heo52d9e672006-01-06 09:49:58 +01001704 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001705 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001706 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001707 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001708}
1709
Jens Axboedece1632015-11-05 10:41:16 -07001710static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
Jiri Slaby5e00d1b2010-08-12 14:31:06 +02001712 const bool sync = !!(bio->bi_rw & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001713 struct blk_plug *plug;
1714 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1715 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001716 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 /*
1719 * low level driver can indicate that it wants pages above a
1720 * certain limit bounced to low memory (ie for highmem, or even
1721 * ISA dma in theory)
1722 */
1723 blk_queue_bounce(q, &bio);
1724
Junichi Nomura23688bf2015-12-22 10:23:44 -07001725 blk_queue_split(q, &bio, q->bio_split);
1726
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001727 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001728 bio->bi_error = -EIO;
1729 bio_endio(bio);
Jens Axboedece1632015-11-05 10:41:16 -07001730 return BLK_QC_T_NONE;
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001731 }
1732
Tejun Heo4fed9472010-09-03 11:56:17 +02001733 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001734 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001735 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001736 goto get_rq;
1737 }
1738
Jens Axboe73c10102011-03-08 13:19:51 +01001739 /*
1740 * Check if we can merge with the plugged list before grabbing
1741 * any locks.
1742 */
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001743 if (!blk_queue_nomerges(q)) {
1744 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
Jens Axboedece1632015-11-05 10:41:16 -07001745 return BLK_QC_T_NONE;
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001746 } else
1747 request_count = blk_plug_queued_count(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001748
1749 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001752 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001753 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001754 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001755 if (!attempt_back_merge(q, req))
1756 elv_merged_request(q, req, el_ret);
1757 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001758 }
Jens Axboe73c10102011-03-08 13:19:51 +01001759 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001760 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001761 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001762 if (!attempt_front_merge(q, req))
1763 elv_merged_request(q, req, el_ret);
1764 goto out_unlock;
1765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001769 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001770 * This sync check and mask will be re-done in init_request_from_bio(),
1771 * but we need to set it earlier to expose the sync flag to the
1772 * rq allocator and io schedulers.
1773 */
1774 rw_flags = bio_data_dir(bio);
1775 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001776 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001777
1778 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001779 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001780 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001781 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001782 req = get_request(q, rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001783 if (IS_ERR(req)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001784 bio->bi_error = PTR_ERR(req);
1785 bio_endio(bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001786 goto out_unlock;
1787 }
Nick Piggind6344532005-06-28 20:45:14 -07001788
Nick Piggin450991b2005-06-28 20:45:13 -07001789 /*
1790 * After dropping the lock and possibly sleeping here, our request
1791 * may now be mergeable after it had proven unmergeable (above).
1792 * We don't worry about that case for efficiency. It won't happen
1793 * often, and the elevators are able to handle it.
1794 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001795 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Tao Ma9562ad92011-10-24 16:11:30 +02001797 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001798 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001799
Jens Axboe73c10102011-03-08 13:19:51 +01001800 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001801 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001802 /*
1803 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001804 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001805 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001806 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001807 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001808 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001809 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001810 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001811 trace_block_plug(q);
1812 }
Jens Axboe73c10102011-03-08 13:19:51 +01001813 }
Shaohua Lia6327162011-08-24 16:04:32 +02001814 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001815 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001816 } else {
1817 spin_lock_irq(q->queue_lock);
1818 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001819 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001820out_unlock:
1821 spin_unlock_irq(q->queue_lock);
1822 }
Jens Axboedece1632015-11-05 10:41:16 -07001823
1824 return BLK_QC_T_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825}
1826
1827/*
1828 * If bio->bi_dev is a partition, remap the location
1829 */
1830static inline void blk_partition_remap(struct bio *bio)
1831{
1832 struct block_device *bdev = bio->bi_bdev;
1833
Jens Axboebf2de6f2007-09-27 13:01:25 +02001834 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001836
Kent Overstreet4f024f32013-10-11 15:44:27 -07001837 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001839
Mike Snitzerd07335e2010-11-16 12:52:38 +01001840 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1841 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001842 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 }
1844}
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846static void handle_bad_sector(struct bio *bio)
1847{
1848 char b[BDEVNAME_SIZE];
1849
1850 printk(KERN_INFO "attempt to access beyond end of device\n");
1851 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1852 bdevname(bio->bi_bdev, b),
1853 bio->bi_rw,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001854 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001855 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
Akinobu Mitac17bb492006-12-08 02:39:46 -08001858#ifdef CONFIG_FAIL_MAKE_REQUEST
1859
1860static DECLARE_FAULT_ATTR(fail_make_request);
1861
1862static int __init setup_fail_make_request(char *str)
1863{
1864 return setup_fault_attr(&fail_make_request, str);
1865}
1866__setup("fail_make_request=", setup_fail_make_request);
1867
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001868static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001869{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001870 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001871}
1872
1873static int __init fail_make_request_debugfs(void)
1874{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001875 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1876 NULL, &fail_make_request);
1877
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001878 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001879}
1880
1881late_initcall(fail_make_request_debugfs);
1882
1883#else /* CONFIG_FAIL_MAKE_REQUEST */
1884
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001885static inline bool should_fail_request(struct hd_struct *part,
1886 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001887{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001888 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001889}
1890
1891#endif /* CONFIG_FAIL_MAKE_REQUEST */
1892
Jens Axboec07e2b42007-07-18 13:27:58 +02001893/*
1894 * Check whether this bio extends beyond the end of the device.
1895 */
1896static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1897{
1898 sector_t maxsector;
1899
1900 if (!nr_sectors)
1901 return 0;
1902
1903 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001904 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001905 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001906 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001907
1908 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1909 /*
1910 * This may well happen - the kernel calls bread()
1911 * without checking the size of the device, e.g., when
1912 * mounting a device.
1913 */
1914 handle_bad_sector(bio);
1915 return 1;
1916 }
1917 }
1918
1919 return 0;
1920}
1921
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001922static noinline_for_stack bool
1923generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924{
Jens Axboe165125e2007-07-24 09:28:11 +02001925 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001926 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001927 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001928 char b[BDEVNAME_SIZE];
1929 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Jens Axboec07e2b42007-07-18 13:27:58 +02001933 if (bio_check_eod(bio, nr_sectors))
1934 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001936 q = bdev_get_queue(bio->bi_bdev);
1937 if (unlikely(!q)) {
1938 printk(KERN_ERR
1939 "generic_make_request: Trying to access "
1940 "nonexistent block-device %s (%Lu)\n",
1941 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001942 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001943 goto end_io;
1944 }
1945
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001946 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001947 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001948 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001949 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001950 goto end_io;
1951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001953 * If this device has partitions, remap block n
1954 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001956 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001958 if (bio_check_eod(bio, nr_sectors))
1959 goto end_io;
1960
1961 /*
1962 * Filter flush bio's early so that make_request based
1963 * drivers without flush support don't have to worry
1964 * about them.
1965 */
1966 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1967 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1968 if (!nr_sectors) {
1969 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001970 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001974 if ((bio->bi_rw & REQ_DISCARD) &&
1975 (!blk_queue_discard(q) ||
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001976 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001977 err = -EOPNOTSUPP;
1978 goto end_io;
1979 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001981 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 err = -EOPNOTSUPP;
1983 goto end_io;
1984 }
1985
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001986 /*
1987 * Various block parts want %current->io_context and lazy ioc
1988 * allocation ends up trading a lot of pain for a small amount of
1989 * memory. Just allocate it upfront. This may fail and block
1990 * layer knows how to live with it.
1991 */
1992 create_io_context(GFP_ATOMIC, q->node);
1993
Tejun Heoae118892015-08-18 14:55:20 -07001994 if (!blkcg_bio_issue_check(q, bio))
1995 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001996
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001997 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001998 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001999
2000end_io:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002001 bio->bi_error = err;
2002 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002003 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004}
2005
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002006/**
2007 * generic_make_request - hand a buffer to its device driver for I/O
2008 * @bio: The bio describing the location in memory and on the device.
2009 *
2010 * generic_make_request() is used to make I/O requests of block
2011 * devices. It is passed a &struct bio, which describes the I/O that needs
2012 * to be done.
2013 *
2014 * generic_make_request() does not return any status. The
2015 * success/failure status of the request, along with notification of
2016 * completion, is delivered asynchronously through the bio->bi_end_io
2017 * function described (one day) else where.
2018 *
2019 * The caller of generic_make_request must make sure that bi_io_vec
2020 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2021 * set to describe the device address, and the
2022 * bi_end_io and optionally bi_private are set to describe how
2023 * completion notification should be signaled.
2024 *
2025 * generic_make_request and the drivers it calls may use bi_next if this
2026 * bio happens to be merged with someone else, and may resubmit the bio to
2027 * a lower device by calling into generic_make_request recursively, which
2028 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02002029 */
Jens Axboedece1632015-11-05 10:41:16 -07002030blk_qc_t generic_make_request(struct bio *bio)
Neil Brownd89d8792007-05-01 09:53:42 +02002031{
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002032 struct bio_list bio_list_on_stack;
Jens Axboedece1632015-11-05 10:41:16 -07002033 blk_qc_t ret = BLK_QC_T_NONE;
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002034
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002035 if (!generic_make_request_checks(bio))
Jens Axboedece1632015-11-05 10:41:16 -07002036 goto out;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002037
2038 /*
2039 * We only want one ->make_request_fn to be active at a time, else
2040 * stack usage with stacked devices could be a problem. So use
2041 * current->bio_list to keep a list of requests submited by a
2042 * make_request_fn function. current->bio_list is also used as a
2043 * flag to say if generic_make_request is currently active in this
2044 * task or not. If it is NULL, then no make_request is active. If
2045 * it is non-NULL, then a make_request is active, and new requests
2046 * should be added at the tail
2047 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002048 if (current->bio_list) {
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002049 bio_list_add(current->bio_list, bio);
Jens Axboedece1632015-11-05 10:41:16 -07002050 goto out;
Neil Brownd89d8792007-05-01 09:53:42 +02002051 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002052
Neil Brownd89d8792007-05-01 09:53:42 +02002053 /* following loop may be a bit non-obvious, and so deserves some
2054 * explanation.
2055 * Before entering the loop, bio->bi_next is NULL (as all callers
2056 * ensure that) so we have a list with a single bio.
2057 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002058 * we assign bio_list to a pointer to the bio_list_on_stack,
2059 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002060 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02002061 * through a recursive call to generic_make_request. If it
2062 * did, we find a non-NULL value in bio_list and re-enter the loop
2063 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002064 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002065 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02002066 */
2067 BUG_ON(bio->bi_next);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002068 bio_list_init(&bio_list_on_stack);
2069 current->bio_list = &bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02002070 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002071 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2072
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01002073 if (likely(blk_queue_enter(q, false) == 0)) {
Jens Axboedece1632015-11-05 10:41:16 -07002074 ret = q->make_request_fn(q, bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002075
2076 blk_queue_exit(q);
2077
2078 bio = bio_list_pop(current->bio_list);
2079 } else {
2080 struct bio *bio_next = bio_list_pop(current->bio_list);
2081
2082 bio_io_error(bio);
2083 bio = bio_next;
2084 }
Neil Brownd89d8792007-05-01 09:53:42 +02002085 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002086 current->bio_list = NULL; /* deactivate */
Jens Axboedece1632015-11-05 10:41:16 -07002087
2088out:
2089 return ret;
Neil Brownd89d8792007-05-01 09:53:42 +02002090}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091EXPORT_SYMBOL(generic_make_request);
2092
2093/**
Randy Dunlap710027a2008-08-19 20:13:11 +02002094 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2096 * @bio: The &struct bio which describes the I/O
2097 *
2098 * submit_bio() is very similar in purpose to generic_make_request(), and
2099 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02002100 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 *
2102 */
Jens Axboedece1632015-11-05 10:41:16 -07002103blk_qc_t submit_bio(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
Jens Axboe22e2c502005-06-27 10:55:12 +02002105 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
Jens Axboebf2de6f2007-09-27 13:01:25 +02002107 /*
2108 * If it's a regular read/write or a barrier with data attached,
2109 * go through the normal accounting stuff before submission.
2110 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002111 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002112 unsigned int count;
2113
2114 if (unlikely(rw & REQ_WRITE_SAME))
2115 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2116 else
2117 count = bio_sectors(bio);
2118
Jens Axboebf2de6f2007-09-27 13:01:25 +02002119 if (rw & WRITE) {
2120 count_vm_events(PGPGOUT, count);
2121 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07002122 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002123 count_vm_events(PGPGIN, count);
2124 }
2125
2126 if (unlikely(block_dump)) {
2127 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02002128 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07002129 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02002130 (rw & WRITE) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002131 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02002132 bdevname(bio->bi_bdev, b),
2133 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 }
2136
Jens Axboedece1632015-11-05 10:41:16 -07002137 return generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139EXPORT_SYMBOL(submit_bio);
2140
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002141/**
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002142 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2143 * for new the queue limits
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002144 * @q: the queue
2145 * @rq: the request being checked
2146 *
2147 * Description:
2148 * @rq may have been made based on weaker limitations of upper-level queues
2149 * in request stacking drivers, and it may violate the limitation of @q.
2150 * Since the block layer and the underlying device driver trust @rq
2151 * after it is inserted to @q, it should be checked against @q before
2152 * the insertion using this generic function.
2153 *
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002154 * Request stacking drivers like request-based dm may change the queue
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002155 * limits when retrying requests on other queues. Those requests need
2156 * to be checked against the new queue limits again during dispatch.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002157 */
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002158static int blk_cloned_rq_check_limits(struct request_queue *q,
2159 struct request *rq)
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002160{
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -04002161 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002162 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2163 return -EIO;
2164 }
2165
2166 /*
2167 * queue's settings related to segment counting like q->bounce_pfn
2168 * may differ from that of other stacking queues.
2169 * Recalculate it to check the request correctly on this queue's
2170 * limitation.
2171 */
2172 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002173 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002174 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2175 return -EIO;
2176 }
2177
2178 return 0;
2179}
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002180
2181/**
2182 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2183 * @q: the queue to submit the request
2184 * @rq: the request being queued
2185 */
2186int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2187{
2188 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002189 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002190
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002191 if (blk_cloned_rq_check_limits(q, rq))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002192 return -EIO;
2193
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002194 if (rq->rq_disk &&
2195 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002196 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002197
Keith Busch7fb48982014-10-17 17:46:38 -06002198 if (q->mq_ops) {
2199 if (blk_queue_io_stat(q))
2200 blk_account_io_start(rq, true);
Mike Snitzer6acfe682016-02-05 08:49:01 -05002201 blk_mq_insert_request(rq, false, true, false);
Keith Busch7fb48982014-10-17 17:46:38 -06002202 return 0;
2203 }
2204
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002205 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002206 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002207 spin_unlock_irqrestore(q->queue_lock, flags);
2208 return -ENODEV;
2209 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002210
2211 /*
2212 * Submitting request must be dequeued before calling this function
2213 * because it will be linked to another request_queue
2214 */
2215 BUG_ON(blk_queued_rq(rq));
2216
Jeff Moyer4853aba2011-08-15 21:37:25 +02002217 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2218 where = ELEVATOR_INSERT_FLUSH;
2219
2220 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002221 if (where == ELEVATOR_INSERT_FLUSH)
2222 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002223 spin_unlock_irqrestore(q->queue_lock, flags);
2224
2225 return 0;
2226}
2227EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2228
Tejun Heo80a761f2009-07-03 17:48:17 +09002229/**
2230 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2231 * @rq: request to examine
2232 *
2233 * Description:
2234 * A request could be merge of IOs which require different failure
2235 * handling. This function determines the number of bytes which
2236 * can be failed from the beginning of the request without
2237 * crossing into area which need to be retried further.
2238 *
2239 * Return:
2240 * The number of bytes to fail.
2241 *
2242 * Context:
2243 * queue_lock must be held.
2244 */
2245unsigned int blk_rq_err_bytes(const struct request *rq)
2246{
2247 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2248 unsigned int bytes = 0;
2249 struct bio *bio;
2250
2251 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2252 return blk_rq_bytes(rq);
2253
2254 /*
2255 * Currently the only 'mixing' which can happen is between
2256 * different fastfail types. We can safely fail portions
2257 * which have all the failfast bits that the first one has -
2258 * the ones which are at least as eager to fail as the first
2259 * one.
2260 */
2261 for (bio = rq->bio; bio; bio = bio->bi_next) {
2262 if ((bio->bi_rw & ff) != ff)
2263 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002264 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002265 }
2266
2267 /* this could lead to infinite loop */
2268 BUG_ON(blk_rq_bytes(rq) && !bytes);
2269 return bytes;
2270}
2271EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2272
Jens Axboe320ae512013-10-24 09:20:05 +01002273void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002274{
Jens Axboec2553b52009-04-24 08:10:11 +02002275 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002276 const int rw = rq_data_dir(req);
2277 struct hd_struct *part;
2278 int cpu;
2279
2280 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002281 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002282 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2283 part_stat_unlock();
2284 }
2285}
2286
Jens Axboe320ae512013-10-24 09:20:05 +01002287void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002288{
Jens Axboebc58ba92009-01-23 10:54:44 +01002289 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002290 * Account IO completion. flush_rq isn't accounted as a
2291 * normal IO on queueing nor completion. Accounting the
2292 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002293 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002294 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002295 unsigned long duration = jiffies - req->start_time;
2296 const int rw = rq_data_dir(req);
2297 struct hd_struct *part;
2298 int cpu;
2299
2300 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002301 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002302
2303 part_stat_inc(cpu, part, ios[rw]);
2304 part_stat_add(cpu, part, ticks[rw], duration);
2305 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002306 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002307
Jens Axboe6c23a962011-01-07 08:43:37 +01002308 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002309 part_stat_unlock();
2310 }
2311}
2312
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002313#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002314/*
2315 * Don't process normal requests when queue is suspended
2316 * or in the process of suspending/resuming
2317 */
2318static struct request *blk_pm_peek_request(struct request_queue *q,
2319 struct request *rq)
2320{
2321 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2322 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2323 return NULL;
2324 else
2325 return rq;
2326}
2327#else
2328static inline struct request *blk_pm_peek_request(struct request_queue *q,
2329 struct request *rq)
2330{
2331 return rq;
2332}
2333#endif
2334
Jens Axboe320ae512013-10-24 09:20:05 +01002335void blk_account_io_start(struct request *rq, bool new_io)
2336{
2337 struct hd_struct *part;
2338 int rw = rq_data_dir(rq);
2339 int cpu;
2340
2341 if (!blk_do_io_stat(rq))
2342 return;
2343
2344 cpu = part_stat_lock();
2345
2346 if (!new_io) {
2347 part = rq->part;
2348 part_stat_inc(cpu, part, merges[rw]);
2349 } else {
2350 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2351 if (!hd_struct_try_get(part)) {
2352 /*
2353 * The partition is already being removed,
2354 * the request will be accounted on the disk only
2355 *
2356 * We take a reference on disk->part0 although that
2357 * partition will never be deleted, so we can treat
2358 * it as any other partition.
2359 */
2360 part = &rq->rq_disk->part0;
2361 hd_struct_get(part);
2362 }
2363 part_round_stats(cpu, part);
2364 part_inc_in_flight(part, rw);
2365 rq->part = part;
2366 }
2367
2368 part_stat_unlock();
2369}
2370
Tejun Heo53a08802008-12-03 12:41:26 +01002371/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002372 * blk_peek_request - peek at the top of a request queue
2373 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002374 *
2375 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002376 * Return the request at the top of @q. The returned request
2377 * should be started using blk_start_request() before LLD starts
2378 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002379 *
2380 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002381 * Pointer to the request at the top of @q if available. Null
2382 * otherwise.
2383 *
2384 * Context:
2385 * queue_lock must be held.
2386 */
2387struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002388{
2389 struct request *rq;
2390 int ret;
2391
2392 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002393
2394 rq = blk_pm_peek_request(q, rq);
2395 if (!rq)
2396 break;
2397
Tejun Heo158dbda2009-04-23 11:05:18 +09002398 if (!(rq->cmd_flags & REQ_STARTED)) {
2399 /*
2400 * This is the first time the device driver
2401 * sees this request (possibly after
2402 * requeueing). Notify IO scheduler.
2403 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002404 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002405 elv_activate_rq(q, rq);
2406
2407 /*
2408 * just mark as started even if we don't start
2409 * it, a request that has been delayed should
2410 * not be passed by new incoming requests
2411 */
2412 rq->cmd_flags |= REQ_STARTED;
2413 trace_block_rq_issue(q, rq);
2414 }
2415
2416 if (!q->boundary_rq || q->boundary_rq == rq) {
2417 q->end_sector = rq_end_sector(rq);
2418 q->boundary_rq = NULL;
2419 }
2420
2421 if (rq->cmd_flags & REQ_DONTPREP)
2422 break;
2423
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002424 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002425 /*
2426 * make sure space for the drain appears we
2427 * know we can do this because max_hw_segments
2428 * has been adjusted to be one fewer than the
2429 * device can handle
2430 */
2431 rq->nr_phys_segments++;
2432 }
2433
2434 if (!q->prep_rq_fn)
2435 break;
2436
2437 ret = q->prep_rq_fn(q, rq);
2438 if (ret == BLKPREP_OK) {
2439 break;
2440 } else if (ret == BLKPREP_DEFER) {
2441 /*
2442 * the request may have been (partially) prepped.
2443 * we need to keep this request in the front to
2444 * avoid resource deadlock. REQ_STARTED will
2445 * prevent other fs requests from passing this one.
2446 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002447 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002448 !(rq->cmd_flags & REQ_DONTPREP)) {
2449 /*
2450 * remove the space for the drain we added
2451 * so that we don't add it again
2452 */
2453 --rq->nr_phys_segments;
2454 }
2455
2456 rq = NULL;
2457 break;
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002458 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2459 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2460
Tejun Heo158dbda2009-04-23 11:05:18 +09002461 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002462 /*
2463 * Mark this request as started so we don't trigger
2464 * any debug logic in the end I/O path.
2465 */
2466 blk_start_request(rq);
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002467 __blk_end_request_all(rq, err);
Tejun Heo158dbda2009-04-23 11:05:18 +09002468 } else {
2469 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2470 break;
2471 }
2472 }
2473
2474 return rq;
2475}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002476EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002477
Tejun Heo9934c8c2009-05-08 11:54:16 +09002478void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002479{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002480 struct request_queue *q = rq->q;
2481
Tejun Heo158dbda2009-04-23 11:05:18 +09002482 BUG_ON(list_empty(&rq->queuelist));
2483 BUG_ON(ELV_ON_HASH(rq));
2484
2485 list_del_init(&rq->queuelist);
2486
2487 /*
2488 * the time frame between a request being removed from the lists
2489 * and to it is freed is accounted as io that is in progress at
2490 * the driver side.
2491 */
Divyesh Shah91952912010-04-01 15:01:41 -07002492 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002493 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002494 set_io_start_time_ns(rq);
2495 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002496}
2497
Tejun Heo5efccd12009-04-23 11:05:18 +09002498/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002499 * blk_start_request - start request processing on the driver
2500 * @req: request to dequeue
2501 *
2502 * Description:
2503 * Dequeue @req and start timeout timer on it. This hands off the
2504 * request to the driver.
2505 *
2506 * Block internal functions which don't want to start timer should
2507 * call blk_dequeue_request().
2508 *
2509 * Context:
2510 * queue_lock must be held.
2511 */
2512void blk_start_request(struct request *req)
2513{
2514 blk_dequeue_request(req);
2515
2516 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002517 * We are now handing the request to the hardware, initialize
2518 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002519 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002520 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002521 if (unlikely(blk_bidi_rq(req)))
2522 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2523
Jeff Moyer4912aa62013-10-08 14:36:41 -04002524 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002525 blk_add_timer(req);
2526}
2527EXPORT_SYMBOL(blk_start_request);
2528
2529/**
2530 * blk_fetch_request - fetch a request from a request queue
2531 * @q: request queue to fetch a request from
2532 *
2533 * Description:
2534 * Return the request at the top of @q. The request is started on
2535 * return and LLD can start processing it immediately.
2536 *
2537 * Return:
2538 * Pointer to the request at the top of @q if available. Null
2539 * otherwise.
2540 *
2541 * Context:
2542 * queue_lock must be held.
2543 */
2544struct request *blk_fetch_request(struct request_queue *q)
2545{
2546 struct request *rq;
2547
2548 rq = blk_peek_request(q);
2549 if (rq)
2550 blk_start_request(rq);
2551 return rq;
2552}
2553EXPORT_SYMBOL(blk_fetch_request);
2554
2555/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002556 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002557 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002558 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002559 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002560 *
2561 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002562 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2563 * the request structure even if @req doesn't have leftover.
2564 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002565 *
2566 * This special helper function is only for request stacking drivers
2567 * (e.g. request-based dm) so that they can handle partial completion.
2568 * Actual device drivers should use blk_end_request instead.
2569 *
2570 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2571 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002572 *
2573 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002574 * %false - this request doesn't have any more data
2575 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002576 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002577bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002579 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002581 trace_block_rq_complete(req->q, req, nr_bytes);
2582
Tejun Heo2e60e022009-04-23 11:05:18 +09002583 if (!req->bio)
2584 return false;
2585
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002587 * For fs requests, rq is just carrier of independent bio's
2588 * and each partial completion should be handled separately.
2589 * Reset per-request error on each partial completion.
2590 *
2591 * TODO: tj: This is too subtle. It would be better to let
2592 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002594 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 req->errors = 0;
2596
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002597 if (error && req->cmd_type == REQ_TYPE_FS &&
2598 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002599 char *error_type;
2600
2601 switch (error) {
2602 case -ENOLINK:
2603 error_type = "recoverable transport";
2604 break;
2605 case -EREMOTEIO:
2606 error_type = "critical target";
2607 break;
2608 case -EBADE:
2609 error_type = "critical nexus";
2610 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002611 case -ETIMEDOUT:
2612 error_type = "timeout";
2613 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002614 case -ENOSPC:
2615 error_type = "critical space allocation";
2616 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002617 case -ENODATA:
2618 error_type = "critical medium";
2619 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002620 case -EIO:
2621 default:
2622 error_type = "I/O";
2623 break;
2624 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002625 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2626 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002627 req->rq_disk->disk_name : "?",
2628 (unsigned long long)blk_rq_pos(req));
2629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 }
2631
Jens Axboebc58ba92009-01-23 10:54:44 +01002632 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002633
Kent Overstreetf79ea412012-09-20 16:38:30 -07002634 total_bytes = 0;
2635 while (req->bio) {
2636 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002637 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
Kent Overstreet4f024f32013-10-11 15:44:27 -07002639 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
Kent Overstreetf79ea412012-09-20 16:38:30 -07002642 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Kent Overstreetf79ea412012-09-20 16:38:30 -07002644 total_bytes += bio_bytes;
2645 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
Kent Overstreetf79ea412012-09-20 16:38:30 -07002647 if (!nr_bytes)
2648 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 }
2650
2651 /*
2652 * completely done
2653 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002654 if (!req->bio) {
2655 /*
2656 * Reset counters so that the request stacking driver
2657 * can find how many bytes remain in the request
2658 * later.
2659 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002660 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002661 return false;
2662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002664 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002665
2666 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002667 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002668 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002669
Tejun Heo80a761f2009-07-03 17:48:17 +09002670 /* mixed attributes always follow the first bio */
2671 if (req->cmd_flags & REQ_MIXED_MERGE) {
2672 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2673 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2674 }
2675
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002676 /*
2677 * If total number of sectors is less than the first segment
2678 * size, something has gone terribly wrong.
2679 */
2680 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002681 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002682 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002683 }
2684
2685 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002687
Tejun Heo2e60e022009-04-23 11:05:18 +09002688 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689}
Tejun Heo2e60e022009-04-23 11:05:18 +09002690EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691
Tejun Heo2e60e022009-04-23 11:05:18 +09002692static bool blk_update_bidi_request(struct request *rq, int error,
2693 unsigned int nr_bytes,
2694 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002695{
Tejun Heo2e60e022009-04-23 11:05:18 +09002696 if (blk_update_request(rq, error, nr_bytes))
2697 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002698
Tejun Heo2e60e022009-04-23 11:05:18 +09002699 /* Bidi request must be completed as a whole */
2700 if (unlikely(blk_bidi_rq(rq)) &&
2701 blk_update_request(rq->next_rq, error, bidi_bytes))
2702 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002703
Jens Axboee2e1a142010-06-09 10:42:09 +02002704 if (blk_queue_add_random(rq->q))
2705 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002706
2707 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708}
2709
James Bottomley28018c22010-07-01 19:49:17 +09002710/**
2711 * blk_unprep_request - unprepare a request
2712 * @req: the request
2713 *
2714 * This function makes a request ready for complete resubmission (or
2715 * completion). It happens only after all error handling is complete,
2716 * so represents the appropriate moment to deallocate any resources
2717 * that were allocated to the request in the prep_rq_fn. The queue
2718 * lock is held when calling this.
2719 */
2720void blk_unprep_request(struct request *req)
2721{
2722 struct request_queue *q = req->q;
2723
2724 req->cmd_flags &= ~REQ_DONTPREP;
2725 if (q->unprep_rq_fn)
2726 q->unprep_rq_fn(q, req);
2727}
2728EXPORT_SYMBOL_GPL(blk_unprep_request);
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730/*
2731 * queue lock must be held
2732 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002733void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002735 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002736 blk_queue_end_tag(req->q, req);
2737
James Bottomleyba396a62009-05-27 14:17:08 +02002738 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002740 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002741 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742
Mike Andersone78042e2008-10-30 02:16:20 -07002743 blk_delete_timer(req);
2744
James Bottomley28018c22010-07-01 19:49:17 +09002745 if (req->cmd_flags & REQ_DONTPREP)
2746 blk_unprep_request(req);
2747
Jens Axboebc58ba92009-01-23 10:54:44 +01002748 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002749
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002751 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002752 else {
2753 if (blk_bidi_rq(req))
2754 __blk_put_request(req->next_rq->q, req->next_rq);
2755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 }
2758}
Christoph Hellwig12120072014-04-16 09:44:59 +02002759EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002761/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002762 * blk_end_bidi_request - Complete a bidi request
2763 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002764 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002765 * @nr_bytes: number of bytes to complete @rq
2766 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002767 *
2768 * Description:
2769 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002770 * Drivers that supports bidi can safely call this member for any
2771 * type of request, bidi or uni. In the later case @bidi_bytes is
2772 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002773 *
2774 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002777 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002778static bool blk_end_bidi_request(struct request *rq, int error,
2779 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002780{
2781 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002782 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002783
Tejun Heo2e60e022009-04-23 11:05:18 +09002784 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2785 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002786
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002787 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002788 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002789 spin_unlock_irqrestore(q->queue_lock, flags);
2790
Tejun Heo2e60e022009-04-23 11:05:18 +09002791 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002792}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002793
2794/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002795 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2796 * @rq: the request to complete
2797 * @error: %0 for success, < %0 for error
2798 * @nr_bytes: number of bytes to complete @rq
2799 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002800 *
2801 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002802 * Identical to blk_end_bidi_request() except that queue lock is
2803 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002804 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002805 * Return:
2806 * %false - we are done with this request
2807 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002808 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002809bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002810 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002811{
Tejun Heo2e60e022009-04-23 11:05:18 +09002812 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2813 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002814
Tejun Heo2e60e022009-04-23 11:05:18 +09002815 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002816
Tejun Heo2e60e022009-04-23 11:05:18 +09002817 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002818}
2819
2820/**
2821 * blk_end_request - Helper function for drivers to complete the request.
2822 * @rq: the request being processed
2823 * @error: %0 for success, < %0 for error
2824 * @nr_bytes: number of bytes to complete
2825 *
2826 * Description:
2827 * Ends I/O on a number of bytes attached to @rq.
2828 * If @rq has leftover, sets it up for the next range of segments.
2829 *
2830 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002831 * %false - we are done with this request
2832 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002833 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002834bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002835{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002836 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002837}
Jens Axboe56ad1742009-07-28 22:11:24 +02002838EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002839
2840/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002841 * blk_end_request_all - Helper function for drives to finish the request.
2842 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002843 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002844 *
2845 * Description:
2846 * Completely finish @rq.
2847 */
2848void blk_end_request_all(struct request *rq, int error)
2849{
2850 bool pending;
2851 unsigned int bidi_bytes = 0;
2852
2853 if (unlikely(blk_bidi_rq(rq)))
2854 bidi_bytes = blk_rq_bytes(rq->next_rq);
2855
2856 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2857 BUG_ON(pending);
2858}
Jens Axboe56ad1742009-07-28 22:11:24 +02002859EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002860
2861/**
2862 * blk_end_request_cur - Helper function to finish the current request chunk.
2863 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002864 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002865 *
2866 * Description:
2867 * Complete the current consecutively mapped chunk from @rq.
2868 *
2869 * Return:
2870 * %false - we are done with this request
2871 * %true - still buffers pending for this request
2872 */
2873bool blk_end_request_cur(struct request *rq, int error)
2874{
2875 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2876}
Jens Axboe56ad1742009-07-28 22:11:24 +02002877EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002878
2879/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002880 * blk_end_request_err - Finish a request till the next failure boundary.
2881 * @rq: the request to finish till the next failure boundary for
2882 * @error: must be negative errno
2883 *
2884 * Description:
2885 * Complete @rq till the next failure boundary.
2886 *
2887 * Return:
2888 * %false - we are done with this request
2889 * %true - still buffers pending for this request
2890 */
2891bool blk_end_request_err(struct request *rq, int error)
2892{
2893 WARN_ON(error >= 0);
2894 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2895}
2896EXPORT_SYMBOL_GPL(blk_end_request_err);
2897
2898/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002899 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002900 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002901 * @error: %0 for success, < %0 for error
2902 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002903 *
2904 * Description:
2905 * Must be called with queue lock held unlike blk_end_request().
2906 *
2907 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002908 * %false - we are done with this request
2909 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002910 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002911bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002912{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002913 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002914}
Jens Axboe56ad1742009-07-28 22:11:24 +02002915EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002916
2917/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002918 * __blk_end_request_all - Helper function for drives to finish the request.
2919 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002920 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002921 *
2922 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002923 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002924 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002925void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002926{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002927 bool pending;
2928 unsigned int bidi_bytes = 0;
2929
2930 if (unlikely(blk_bidi_rq(rq)))
2931 bidi_bytes = blk_rq_bytes(rq->next_rq);
2932
2933 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2934 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002935}
Jens Axboe56ad1742009-07-28 22:11:24 +02002936EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002937
2938/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002939 * __blk_end_request_cur - Helper function to finish the current request chunk.
2940 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002941 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002942 *
2943 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002944 * Complete the current consecutively mapped chunk from @rq. Must
2945 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002946 *
2947 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002948 * %false - we are done with this request
2949 * %true - still buffers pending for this request
2950 */
2951bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002952{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002953 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002954}
Jens Axboe56ad1742009-07-28 22:11:24 +02002955EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002956
Tejun Heo80a761f2009-07-03 17:48:17 +09002957/**
2958 * __blk_end_request_err - Finish a request till the next failure boundary.
2959 * @rq: the request to finish till the next failure boundary for
2960 * @error: must be negative errno
2961 *
2962 * Description:
2963 * Complete @rq till the next failure boundary. Must be called
2964 * with queue lock held.
2965 *
2966 * Return:
2967 * %false - we are done with this request
2968 * %true - still buffers pending for this request
2969 */
2970bool __blk_end_request_err(struct request *rq, int error)
2971{
2972 WARN_ON(error >= 0);
2973 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2974}
2975EXPORT_SYMBOL_GPL(__blk_end_request_err);
2976
Jens Axboe86db1e22008-01-29 14:53:40 +01002977void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2978 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979{
Tejun Heoa82afdf2009-07-03 17:48:16 +09002980 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02002981 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982
Jens Axboeb4f42e22014-04-10 09:46:28 -06002983 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002984 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002985
Kent Overstreet4f024f32013-10-11 15:44:27 -07002986 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
NeilBrown66846572007-08-16 13:31:28 +02002989 if (bio->bi_bdev)
2990 rq->rq_disk = bio->bi_bdev->bd_disk;
2991}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002993#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2994/**
2995 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2996 * @rq: the request to be flushed
2997 *
2998 * Description:
2999 * Flush all pages in @rq.
3000 */
3001void rq_flush_dcache_pages(struct request *rq)
3002{
3003 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08003004 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01003005
3006 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08003007 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01003008}
3009EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3010#endif
3011
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02003012/**
3013 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3014 * @q : the queue of the device being checked
3015 *
3016 * Description:
3017 * Check if underlying low-level drivers of a device are busy.
3018 * If the drivers want to export their busy state, they must set own
3019 * exporting function using blk_queue_lld_busy() first.
3020 *
3021 * Basically, this function is used only by request stacking drivers
3022 * to stop dispatching requests to underlying devices when underlying
3023 * devices are busy. This behavior helps more I/O merging on the queue
3024 * of the request stacking driver and prevents I/O throughput regression
3025 * on burst I/O load.
3026 *
3027 * Return:
3028 * 0 - Not busy (The request stacking driver should dispatch request)
3029 * 1 - Busy (The request stacking driver should stop dispatching request)
3030 */
3031int blk_lld_busy(struct request_queue *q)
3032{
3033 if (q->lld_busy_fn)
3034 return q->lld_busy_fn(q);
3035
3036 return 0;
3037}
3038EXPORT_SYMBOL_GPL(blk_lld_busy);
3039
Mike Snitzer78d8e582015-06-26 10:01:13 -04003040/**
3041 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3042 * @rq: the clone request to be cleaned up
3043 *
3044 * Description:
3045 * Free all bios in @rq for a cloned request.
3046 */
3047void blk_rq_unprep_clone(struct request *rq)
3048{
3049 struct bio *bio;
3050
3051 while ((bio = rq->bio) != NULL) {
3052 rq->bio = bio->bi_next;
3053
3054 bio_put(bio);
3055 }
3056}
3057EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3058
3059/*
3060 * Copy attributes of the original request to the clone request.
3061 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3062 */
3063static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003064{
3065 dst->cpu = src->cpu;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003066 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003067 dst->cmd_type = src->cmd_type;
3068 dst->__sector = blk_rq_pos(src);
3069 dst->__data_len = blk_rq_bytes(src);
3070 dst->nr_phys_segments = src->nr_phys_segments;
3071 dst->ioprio = src->ioprio;
3072 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003073}
3074
3075/**
3076 * blk_rq_prep_clone - Helper function to setup clone request
3077 * @rq: the request to be setup
3078 * @rq_src: original request to be cloned
3079 * @bs: bio_set that bios for clone are allocated from
3080 * @gfp_mask: memory allocation mask for bio
3081 * @bio_ctr: setup function to be called for each clone bio.
3082 * Returns %0 for success, non %0 for failure.
3083 * @data: private data to be passed to @bio_ctr
3084 *
3085 * Description:
3086 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3087 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3088 * are not copied, and copying such parts is the caller's responsibility.
3089 * Also, pages which the original bios are pointing to are not copied
3090 * and the cloned bios just point same pages.
3091 * So cloned bios must be completed before original bios, which means
3092 * the caller must complete @rq before @rq_src.
3093 */
3094int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3095 struct bio_set *bs, gfp_t gfp_mask,
3096 int (*bio_ctr)(struct bio *, struct bio *, void *),
3097 void *data)
3098{
3099 struct bio *bio, *bio_src;
3100
3101 if (!bs)
3102 bs = fs_bio_set;
3103
3104 __rq_for_each_bio(bio_src, rq_src) {
3105 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3106 if (!bio)
3107 goto free_and_out;
3108
3109 if (bio_ctr && bio_ctr(bio, bio_src, data))
3110 goto free_and_out;
3111
3112 if (rq->bio) {
3113 rq->biotail->bi_next = bio;
3114 rq->biotail = bio;
3115 } else
3116 rq->bio = rq->biotail = bio;
3117 }
3118
3119 __blk_rq_prep_clone(rq, rq_src);
3120
3121 return 0;
3122
3123free_and_out:
3124 if (bio)
3125 bio_put(bio);
3126 blk_rq_unprep_clone(rq);
3127
3128 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003129}
3130EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3131
Jens Axboe59c3d452014-04-08 09:15:35 -06003132int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133{
3134 return queue_work(kblockd_workqueue, work);
3135}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136EXPORT_SYMBOL(kblockd_schedule_work);
3137
Jens Axboe59c3d452014-04-08 09:15:35 -06003138int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3139 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003140{
3141 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3142}
3143EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3144
Jens Axboe8ab14592014-04-08 09:17:40 -06003145int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3146 unsigned long delay)
3147{
3148 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3149}
3150EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3151
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003152/**
3153 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3154 * @plug: The &struct blk_plug that needs to be initialized
3155 *
3156 * Description:
3157 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3158 * pending I/O should the task end up blocking between blk_start_plug() and
3159 * blk_finish_plug(). This is important from a performance perspective, but
3160 * also ensures that we don't deadlock. For instance, if the task is blocking
3161 * for a memory allocation, memory reclaim could end up wanting to free a
3162 * page belonging to that request that is currently residing in our private
3163 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3164 * this kind of deadlock.
3165 */
Jens Axboe73c10102011-03-08 13:19:51 +01003166void blk_start_plug(struct blk_plug *plug)
3167{
3168 struct task_struct *tsk = current;
3169
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003170 /*
3171 * If this is a nested plug, don't actually assign it.
3172 */
3173 if (tsk->plug)
3174 return;
3175
Jens Axboe73c10102011-03-08 13:19:51 +01003176 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003177 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003178 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003179 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003180 * Store ordering should not be needed here, since a potential
3181 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01003182 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003183 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01003184}
3185EXPORT_SYMBOL(blk_start_plug);
3186
3187static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3188{
3189 struct request *rqa = container_of(a, struct request, queuelist);
3190 struct request *rqb = container_of(b, struct request, queuelist);
3191
Jianpeng Ma975927b2012-10-25 21:58:17 +02003192 return !(rqa->q < rqb->q ||
3193 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003194}
3195
Jens Axboe49cac012011-04-16 13:51:05 +02003196/*
3197 * If 'from_schedule' is true, then postpone the dispatch of requests
3198 * until a safe kblockd context. We due this to avoid accidental big
3199 * additional stack usage in driver dispatch, in places where the originally
3200 * plugger did not intend it.
3201 */
Jens Axboef6603782011-04-15 15:49:07 +02003202static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003203 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003204 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003205{
Jens Axboe49cac012011-04-16 13:51:05 +02003206 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003207
Bart Van Assche70460572012-11-28 13:45:56 +01003208 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003209 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003210 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003211 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003212 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003213}
3214
NeilBrown74018dc2012-07-31 09:08:15 +02003215static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003216{
3217 LIST_HEAD(callbacks);
3218
Shaohua Li2a7d5552012-07-31 09:08:15 +02003219 while (!list_empty(&plug->cb_list)) {
3220 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003221
Shaohua Li2a7d5552012-07-31 09:08:15 +02003222 while (!list_empty(&callbacks)) {
3223 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003224 struct blk_plug_cb,
3225 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003226 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003227 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003228 }
NeilBrown048c9372011-04-18 09:52:22 +02003229 }
3230}
3231
NeilBrown9cbb1752012-07-31 09:08:14 +02003232struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3233 int size)
3234{
3235 struct blk_plug *plug = current->plug;
3236 struct blk_plug_cb *cb;
3237
3238 if (!plug)
3239 return NULL;
3240
3241 list_for_each_entry(cb, &plug->cb_list, list)
3242 if (cb->callback == unplug && cb->data == data)
3243 return cb;
3244
3245 /* Not currently on the callback list */
3246 BUG_ON(size < sizeof(*cb));
3247 cb = kzalloc(size, GFP_ATOMIC);
3248 if (cb) {
3249 cb->data = data;
3250 cb->callback = unplug;
3251 list_add(&cb->list, &plug->cb_list);
3252 }
3253 return cb;
3254}
3255EXPORT_SYMBOL(blk_check_plugged);
3256
Jens Axboe49cac012011-04-16 13:51:05 +02003257void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003258{
3259 struct request_queue *q;
3260 unsigned long flags;
3261 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003262 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003263 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003264
NeilBrown74018dc2012-07-31 09:08:15 +02003265 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003266
3267 if (!list_empty(&plug->mq_list))
3268 blk_mq_flush_plug_list(plug, from_schedule);
3269
Jens Axboe73c10102011-03-08 13:19:51 +01003270 if (list_empty(&plug->list))
3271 return;
3272
NeilBrown109b8122011-04-11 14:13:10 +02003273 list_splice_init(&plug->list, &list);
3274
Jianpeng Ma422765c2013-01-11 14:46:09 +01003275 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003276
3277 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003278 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003279
3280 /*
3281 * Save and disable interrupts here, to avoid doing it for every
3282 * queue lock we have to take.
3283 */
Jens Axboe73c10102011-03-08 13:19:51 +01003284 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003285 while (!list_empty(&list)) {
3286 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003287 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003288 BUG_ON(!rq->q);
3289 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003290 /*
3291 * This drops the queue lock
3292 */
3293 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003294 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003295 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003296 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003297 spin_lock(q->queue_lock);
3298 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003299
3300 /*
3301 * Short-circuit if @q is dead
3302 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003303 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003304 __blk_end_request_all(rq, -ENODEV);
3305 continue;
3306 }
3307
Jens Axboe73c10102011-03-08 13:19:51 +01003308 /*
3309 * rq is already accounted, so use raw insert
3310 */
Jens Axboe401a18e2011-03-25 16:57:52 +01003311 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3312 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3313 else
3314 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003315
3316 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003317 }
3318
Jens Axboe99e22592011-04-18 09:59:55 +02003319 /*
3320 * This drops the queue lock
3321 */
3322 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003323 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003324
Jens Axboe73c10102011-03-08 13:19:51 +01003325 local_irq_restore(flags);
3326}
Jens Axboe73c10102011-03-08 13:19:51 +01003327
3328void blk_finish_plug(struct blk_plug *plug)
3329{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003330 if (plug != current->plug)
3331 return;
Jens Axboef6603782011-04-15 15:49:07 +02003332 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003333
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003334 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003335}
3336EXPORT_SYMBOL(blk_finish_plug);
3337
Jens Axboe05229be2015-11-05 10:44:55 -07003338bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3339{
3340 struct blk_plug *plug;
3341 long state;
3342
3343 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3344 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3345 return false;
3346
3347 plug = current->plug;
3348 if (plug)
3349 blk_flush_plug_list(plug, false);
3350
3351 state = current->state;
3352 while (!need_resched()) {
3353 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3354 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3355 int ret;
3356
3357 hctx->poll_invoked++;
3358
3359 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3360 if (ret > 0) {
3361 hctx->poll_success++;
3362 set_current_state(TASK_RUNNING);
3363 return true;
3364 }
3365
3366 if (signal_pending_state(state, current))
3367 set_current_state(TASK_RUNNING);
3368
3369 if (current->state == TASK_RUNNING)
3370 return true;
3371 if (ret < 0)
3372 break;
3373 cpu_relax();
3374 }
3375
3376 return false;
3377}
3378
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003379#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003380/**
3381 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3382 * @q: the queue of the device
3383 * @dev: the device the queue belongs to
3384 *
3385 * Description:
3386 * Initialize runtime-PM-related fields for @q and start auto suspend for
3387 * @dev. Drivers that want to take advantage of request-based runtime PM
3388 * should call this function after @dev has been initialized, and its
3389 * request queue @q has been allocated, and runtime PM for it can not happen
3390 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3391 * cases, driver should call this function before any I/O has taken place.
3392 *
3393 * This function takes care of setting up using auto suspend for the device,
3394 * the autosuspend delay is set to -1 to make runtime suspend impossible
3395 * until an updated value is either set by user or by driver. Drivers do
3396 * not need to touch other autosuspend settings.
3397 *
3398 * The block layer runtime PM is request based, so only works for drivers
3399 * that use request as their IO unit instead of those directly use bio's.
3400 */
3401void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3402{
3403 q->dev = dev;
3404 q->rpm_status = RPM_ACTIVE;
3405 pm_runtime_set_autosuspend_delay(q->dev, -1);
3406 pm_runtime_use_autosuspend(q->dev);
3407}
3408EXPORT_SYMBOL(blk_pm_runtime_init);
3409
3410/**
3411 * blk_pre_runtime_suspend - Pre runtime suspend check
3412 * @q: the queue of the device
3413 *
3414 * Description:
3415 * This function will check if runtime suspend is allowed for the device
3416 * by examining if there are any requests pending in the queue. If there
3417 * are requests pending, the device can not be runtime suspended; otherwise,
3418 * the queue's status will be updated to SUSPENDING and the driver can
3419 * proceed to suspend the device.
3420 *
3421 * For the not allowed case, we mark last busy for the device so that
3422 * runtime PM core will try to autosuspend it some time later.
3423 *
3424 * This function should be called near the start of the device's
3425 * runtime_suspend callback.
3426 *
3427 * Return:
3428 * 0 - OK to runtime suspend the device
3429 * -EBUSY - Device should not be runtime suspended
3430 */
3431int blk_pre_runtime_suspend(struct request_queue *q)
3432{
3433 int ret = 0;
3434
Ken Xue4fd41a852015-12-01 14:45:46 +08003435 if (!q->dev)
3436 return ret;
3437
Lin Ming6c954662013-03-23 11:42:26 +08003438 spin_lock_irq(q->queue_lock);
3439 if (q->nr_pending) {
3440 ret = -EBUSY;
3441 pm_runtime_mark_last_busy(q->dev);
3442 } else {
3443 q->rpm_status = RPM_SUSPENDING;
3444 }
3445 spin_unlock_irq(q->queue_lock);
3446 return ret;
3447}
3448EXPORT_SYMBOL(blk_pre_runtime_suspend);
3449
3450/**
3451 * blk_post_runtime_suspend - Post runtime suspend processing
3452 * @q: the queue of the device
3453 * @err: return value of the device's runtime_suspend function
3454 *
3455 * Description:
3456 * Update the queue's runtime status according to the return value of the
3457 * device's runtime suspend function and mark last busy for the device so
3458 * that PM core will try to auto suspend the device at a later time.
3459 *
3460 * This function should be called near the end of the device's
3461 * runtime_suspend callback.
3462 */
3463void blk_post_runtime_suspend(struct request_queue *q, int err)
3464{
Ken Xue4fd41a852015-12-01 14:45:46 +08003465 if (!q->dev)
3466 return;
3467
Lin Ming6c954662013-03-23 11:42:26 +08003468 spin_lock_irq(q->queue_lock);
3469 if (!err) {
3470 q->rpm_status = RPM_SUSPENDED;
3471 } else {
3472 q->rpm_status = RPM_ACTIVE;
3473 pm_runtime_mark_last_busy(q->dev);
3474 }
3475 spin_unlock_irq(q->queue_lock);
3476}
3477EXPORT_SYMBOL(blk_post_runtime_suspend);
3478
3479/**
3480 * blk_pre_runtime_resume - Pre runtime resume processing
3481 * @q: the queue of the device
3482 *
3483 * Description:
3484 * Update the queue's runtime status to RESUMING in preparation for the
3485 * runtime resume of the device.
3486 *
3487 * This function should be called near the start of the device's
3488 * runtime_resume callback.
3489 */
3490void blk_pre_runtime_resume(struct request_queue *q)
3491{
Ken Xue4fd41a852015-12-01 14:45:46 +08003492 if (!q->dev)
3493 return;
3494
Lin Ming6c954662013-03-23 11:42:26 +08003495 spin_lock_irq(q->queue_lock);
3496 q->rpm_status = RPM_RESUMING;
3497 spin_unlock_irq(q->queue_lock);
3498}
3499EXPORT_SYMBOL(blk_pre_runtime_resume);
3500
3501/**
3502 * blk_post_runtime_resume - Post runtime resume processing
3503 * @q: the queue of the device
3504 * @err: return value of the device's runtime_resume function
3505 *
3506 * Description:
3507 * Update the queue's runtime status according to the return value of the
3508 * device's runtime_resume function. If it is successfully resumed, process
3509 * the requests that are queued into the device's queue when it is resuming
3510 * and then mark last busy and initiate autosuspend for it.
3511 *
3512 * This function should be called near the end of the device's
3513 * runtime_resume callback.
3514 */
3515void blk_post_runtime_resume(struct request_queue *q, int err)
3516{
Ken Xue4fd41a852015-12-01 14:45:46 +08003517 if (!q->dev)
3518 return;
3519
Lin Ming6c954662013-03-23 11:42:26 +08003520 spin_lock_irq(q->queue_lock);
3521 if (!err) {
3522 q->rpm_status = RPM_ACTIVE;
3523 __blk_run_queue(q);
3524 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003525 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003526 } else {
3527 q->rpm_status = RPM_SUSPENDED;
3528 }
3529 spin_unlock_irq(q->queue_lock);
3530}
3531EXPORT_SYMBOL(blk_post_runtime_resume);
Mika Westerbergd07ab6d2016-02-18 10:54:11 +02003532
3533/**
3534 * blk_set_runtime_active - Force runtime status of the queue to be active
3535 * @q: the queue of the device
3536 *
3537 * If the device is left runtime suspended during system suspend the resume
3538 * hook typically resumes the device and corrects runtime status
3539 * accordingly. However, that does not affect the queue runtime PM status
3540 * which is still "suspended". This prevents processing requests from the
3541 * queue.
3542 *
3543 * This function can be used in driver's resume hook to correct queue
3544 * runtime PM status and re-enable peeking requests from the queue. It
3545 * should be called before first request is added to the queue.
3546 */
3547void blk_set_runtime_active(struct request_queue *q)
3548{
3549 spin_lock_irq(q->queue_lock);
3550 q->rpm_status = RPM_ACTIVE;
3551 pm_runtime_mark_last_busy(q->dev);
3552 pm_request_autosuspend(q->dev);
3553 spin_unlock_irq(q->queue_lock);
3554}
3555EXPORT_SYMBOL(blk_set_runtime_active);
Lin Ming6c954662013-03-23 11:42:26 +08003556#endif
3557
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558int __init blk_dev_init(void)
3559{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003560 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05303561 FIELD_SIZEOF(struct request, cmd_flags));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003562
Tejun Heo89b90be2011-01-03 15:01:47 +01003563 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3564 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003565 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 if (!kblockd_workqueue)
3567 panic("Failed to create kblockd\n");
3568
3569 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003570 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571
Ilya Dryomovc2789bd2015-11-20 22:16:46 +01003572 blk_requestq_cachep = kmem_cache_create("request_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003573 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 return 0;
3576}