blob: d1f2801ce836bea662664f7ca67bf6f9942fdba3 [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) {
Jens Axboe27489a32016-08-24 15:54:25 -0600291 cancel_work_sync(&hctx->run_work);
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600292 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{
Bart Van Assche1b856082016-08-16 16:48:36 -0700518 spin_lock_irq(q->queue_lock);
519 queue_flag_set(QUEUE_FLAG_DYING, q);
520 spin_unlock_irq(q->queue_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700521
522 if (q->mq_ops)
523 blk_mq_wake_waiters(q);
524 else {
525 struct request_list *rl;
526
527 blk_queue_for_each_rl(rl, q) {
528 if (rl->rq_pool) {
529 wake_up(&rl->wait[BLK_RW_SYNC]);
530 wake_up(&rl->wait[BLK_RW_ASYNC]);
531 }
532 }
533 }
534}
535EXPORT_SYMBOL_GPL(blk_set_queue_dying);
536
Tejun Heod7325802012-03-05 13:14:58 -0800537/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200538 * blk_cleanup_queue - shutdown a request queue
539 * @q: request queue to shutdown
540 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100541 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
542 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500543 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100544void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500545{
Tejun Heoc9a929d2011-10-19 14:42:16 +0200546 spinlock_t *lock = q->queue_lock;
Jens Axboee3335de92008-09-18 09:22:54 -0700547
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100548 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500549 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700550 blk_set_queue_dying(q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200551 spin_lock_irq(lock);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800552
Tejun Heo80fd9972012-04-13 14:50:53 -0700553 /*
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100554 * A dying queue is permanently in bypass mode till released. Note
Tejun Heo80fd9972012-04-13 14:50:53 -0700555 * that, unlike blk_queue_bypass_start(), we aren't performing
556 * synchronize_rcu() after entering bypass mode to avoid the delay
557 * as some drivers create and destroy a lot of queues while
558 * probing. This is still safe because blk_release_queue() will be
559 * called only after the queue refcnt drops to zero and nothing,
560 * RCU or not, would be traversing the queue by then.
561 */
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800562 q->bypass_depth++;
563 queue_flag_set(QUEUE_FLAG_BYPASS, q);
564
Tejun Heoc9a929d2011-10-19 14:42:16 +0200565 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
566 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100567 queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200568 spin_unlock_irq(lock);
569 mutex_unlock(&q->sysfs_lock);
570
Bart Van Asschec246e802012-12-06 14:32:01 +0100571 /*
572 * Drain all requests queued before DYING marking. Set DEAD flag to
573 * prevent that q->request_fn() gets invoked after draining finished.
574 */
Dan Williams3ef28e82015-10-21 13:20:12 -0400575 blk_freeze_queue(q);
576 spin_lock_irq(lock);
577 if (!q->mq_ops)
Ming Lei43a5e4e2013-12-26 21:31:35 +0800578 __blk_drain_queue(q, true);
Bart Van Asschec246e802012-12-06 14:32:01 +0100579 queue_flag_set(QUEUE_FLAG_DEAD, q);
Bart Van Assche807592a2012-11-28 13:43:38 +0100580 spin_unlock_irq(lock);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200581
Dan Williams5a48fc12015-10-21 13:20:23 -0400582 /* for synchronous bio-based driver finish in-flight integrity i/o */
583 blk_flush_integrity();
584
Tejun Heoc9a929d2011-10-19 14:42:16 +0200585 /* @q won't process any more request, flush async actions */
586 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
587 blk_sync_queue(q);
588
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100589 if (q->mq_ops)
590 blk_mq_free_queue(q);
Dan Williams3ef28e82015-10-21 13:20:12 -0400591 percpu_ref_exit(&q->q_usage_counter);
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100592
Asias He5e5cfac2012-05-24 23:28:52 +0800593 spin_lock_irq(lock);
594 if (q->queue_lock != &q->__queue_lock)
595 q->queue_lock = &q->__queue_lock;
596 spin_unlock_irq(lock);
597
Tejun Heob02176f2015-09-08 12:20:22 -0400598 bdi_unregister(&q->backing_dev_info);
NeilBrown6cd18e72015-04-27 14:12:22 +1000599
Tejun Heoc9a929d2011-10-19 14:42:16 +0200600 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500601 blk_put_queue(q);
602}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603EXPORT_SYMBOL(blk_cleanup_queue);
604
David Rientjes271508d2015-03-24 16:21:16 -0700605/* Allocate memory local to the request queue */
606static void *alloc_request_struct(gfp_t gfp_mask, void *data)
607{
608 int nid = (int)(long)data;
609 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
610}
611
612static void free_request_struct(void *element, void *unused)
613{
614 kmem_cache_free(request_cachep, element);
615}
616
Tejun Heo5b788ce2012-06-04 20:40:59 -0700617int blk_init_rl(struct request_list *rl, struct request_queue *q,
618 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Mike Snitzer1abec4f2010-05-25 13:15:15 -0400620 if (unlikely(rl->rq_pool))
621 return 0;
622
Tejun Heo5b788ce2012-06-04 20:40:59 -0700623 rl->q = q;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200624 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
625 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200626 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
627 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
David Rientjes271508d2015-03-24 16:21:16 -0700629 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
630 free_request_struct,
631 (void *)(long)q->node, gfp_mask,
632 q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 if (!rl->rq_pool)
634 return -ENOMEM;
635
636 return 0;
637}
638
Tejun Heo5b788ce2012-06-04 20:40:59 -0700639void blk_exit_rl(struct request_list *rl)
640{
641 if (rl->rq_pool)
642 mempool_destroy(rl->rq_pool);
643}
644
Jens Axboe165125e2007-07-24 09:28:11 +0200645struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100647 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700648}
649EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100651int blk_queue_enter(struct request_queue *q, bool nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400652{
653 while (true) {
654 int ret;
655
656 if (percpu_ref_tryget_live(&q->q_usage_counter))
657 return 0;
658
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100659 if (nowait)
Dan Williams3ef28e82015-10-21 13:20:12 -0400660 return -EBUSY;
661
662 ret = wait_event_interruptible(q->mq_freeze_wq,
663 !atomic_read(&q->mq_freeze_depth) ||
664 blk_queue_dying(q));
665 if (blk_queue_dying(q))
666 return -ENODEV;
667 if (ret)
668 return ret;
669 }
670}
671
672void blk_queue_exit(struct request_queue *q)
673{
674 percpu_ref_put(&q->q_usage_counter);
675}
676
677static void blk_queue_usage_counter_release(struct percpu_ref *ref)
678{
679 struct request_queue *q =
680 container_of(ref, struct request_queue, q_usage_counter);
681
682 wake_up_all(&q->mq_freeze_wq);
683}
684
Christoph Hellwig287922e2015-10-30 20:57:30 +0800685static void blk_rq_timed_out_timer(unsigned long data)
686{
687 struct request_queue *q = (struct request_queue *)data;
688
689 kblockd_schedule_work(&q->timeout_work);
690}
691
Jens Axboe165125e2007-07-24 09:28:11 +0200692struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700693{
Jens Axboe165125e2007-07-24 09:28:11 +0200694 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700695 int err;
Christoph Lameter19460892005-06-23 00:08:19 -0700696
Jens Axboe8324aa92008-01-29 14:51:59 +0100697 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700698 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (!q)
700 return NULL;
701
Dan Carpenter00380a42012-03-23 09:58:54 +0100702 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100703 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800704 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100705
Kent Overstreet54efd502015-04-23 22:37:18 -0700706 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
707 if (!q->bio_split)
708 goto fail_id;
709
Jens Axboe0989a022009-06-12 14:42:56 +0200710 q->backing_dev_info.ra_pages =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300711 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
Tejun Heo89e9b9e2015-05-22 17:13:36 -0400712 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
Jens Axboed9938312009-06-12 14:45:52 +0200713 q->backing_dev_info.name = "block";
Mike Snitzer51514122011-11-23 10:59:13 +0100714 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200715
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700716 err = bdi_init(&q->backing_dev_info);
Tejun Heoa73f7302011-12-14 00:33:37 +0100717 if (err)
Kent Overstreet54efd502015-04-23 22:37:18 -0700718 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700719
Matthew Garrett31373d02010-04-06 14:25:14 +0200720 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
721 laptop_mode_timer_fn, (unsigned long) q);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700722 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
Tejun Heob855b042012-03-06 21:24:55 +0100723 INIT_LIST_HEAD(&q->queue_head);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700724 INIT_LIST_HEAD(&q->timeout_list);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100725 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800726#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800727 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800728#endif
Jens Axboe3cca6dc2011-03-02 11:08:00 -0500729 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
Al Viro483f4af2006-03-18 18:34:37 -0500730
Jens Axboe8324aa92008-01-29 14:51:59 +0100731 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Al Viro483f4af2006-03-18 18:34:37 -0500733 mutex_init(&q->sysfs_lock);
Neil Browne7e72bf2008-05-14 16:05:54 -0700734 spin_lock_init(&q->__queue_lock);
Al Viro483f4af2006-03-18 18:34:37 -0500735
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500736 /*
737 * By default initialize queue_lock to internal lock and driver can
738 * override it later if need be.
739 */
740 q->queue_lock = &q->__queue_lock;
741
Tejun Heob82d4b12012-04-13 13:11:31 -0700742 /*
743 * A queue starts its life with bypass turned on to avoid
744 * unnecessary bypass on/off overhead and nasty surprises during
Tejun Heo749fefe2012-09-20 14:08:52 -0700745 * init. The initial bypass will be finished when the queue is
746 * registered by blk_register_queue().
Tejun Heob82d4b12012-04-13 13:11:31 -0700747 */
748 q->bypass_depth = 1;
749 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
750
Jens Axboe320ae512013-10-24 09:20:05 +0100751 init_waitqueue_head(&q->mq_freeze_wq);
752
Dan Williams3ef28e82015-10-21 13:20:12 -0400753 /*
754 * Init percpu_ref in atomic mode so that it's faster to shutdown.
755 * See blk_register_queue() for details.
756 */
757 if (percpu_ref_init(&q->q_usage_counter,
758 blk_queue_usage_counter_release,
759 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
Mikulas Patockafff49962013-10-14 12:11:36 -0400760 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800761
Dan Williams3ef28e82015-10-21 13:20:12 -0400762 if (blkcg_init_queue(q))
763 goto fail_ref;
764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100766
Dan Williams3ef28e82015-10-21 13:20:12 -0400767fail_ref:
768 percpu_ref_exit(&q->q_usage_counter);
Mikulas Patockafff49962013-10-14 12:11:36 -0400769fail_bdi:
770 bdi_destroy(&q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700771fail_split:
772 bioset_free(q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100773fail_id:
774 ida_simple_remove(&blk_queue_ida, q->id);
775fail_q:
776 kmem_cache_free(blk_requestq_cachep, q);
777 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
Christoph Lameter19460892005-06-23 00:08:19 -0700779EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781/**
782 * blk_init_queue - prepare a request queue for use with a block device
783 * @rfn: The function to be called to process requests that have been
784 * placed on the queue.
785 * @lock: Request queue spin lock
786 *
787 * Description:
788 * If a block device wishes to use the standard request handling procedures,
789 * which sorts requests and coalesces adjacent requests, then it must
790 * call blk_init_queue(). The function @rfn will be called when there
791 * are requests on the queue that need to be processed. If the device
792 * supports plugging, then @rfn may not be called immediately when requests
793 * are available on the queue, but may be called at some time later instead.
794 * Plugged queues are generally unplugged when a buffer belonging to one
795 * of the requests on the queue is needed, or due to memory pressure.
796 *
797 * @rfn is not required, or even expected, to remove all requests off the
798 * queue, but only as many as it can handle at a time. If it does leave
799 * requests on the queue, it is responsible for arranging that the requests
800 * get dealt with eventually.
801 *
802 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +0200803 * request queue; this lock will be taken also from interrupt context, so irq
804 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 *
Randy Dunlap710027a2008-08-19 20:13:11 +0200806 * Function returns a pointer to the initialized request queue, or %NULL if
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * it didn't succeed.
808 *
809 * Note:
810 * blk_init_queue() must be paired with a blk_cleanup_queue() call
811 * when the block device is deactivated (such as at module unload).
812 **/
Christoph Lameter19460892005-06-23 00:08:19 -0700813
Jens Axboe165125e2007-07-24 09:28:11 +0200814struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Ezequiel Garciac304a512012-11-10 10:39:44 +0100816 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700817}
818EXPORT_SYMBOL(blk_init_queue);
819
Jens Axboe165125e2007-07-24 09:28:11 +0200820struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -0700821blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
822{
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600823 struct request_queue *uninit_q, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600825 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
826 if (!uninit_q)
827 return NULL;
828
Mike Snitzer51514122011-11-23 10:59:13 +0100829 q = blk_init_allocated_queue(uninit_q, rfn, lock);
Mike Snitzerc86d1b82010-06-03 11:34:52 -0600830 if (!q)
Mike Snitzer7982e902014-03-08 17:20:01 -0700831 blk_cleanup_queue(uninit_q);
Christoph Hellwig18741982014-02-10 09:29:00 -0700832
Mike Snitzer7982e902014-03-08 17:20:01 -0700833 return q;
Mike Snitzer01effb02010-05-11 08:57:42 +0200834}
835EXPORT_SYMBOL(blk_init_queue_node);
836
Jens Axboedece1632015-11-05 10:41:16 -0700837static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
Mike Snitzer336b7e12015-05-11 14:06:32 -0400838
Mike Snitzer01effb02010-05-11 08:57:42 +0200839struct request_queue *
840blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
841 spinlock_t *lock)
842{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (!q)
844 return NULL;
845
Ming Leif70ced02014-09-25 23:23:47 +0800846 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
Ming Leiba483382014-09-25 23:23:44 +0800847 if (!q->fq)
Mike Snitzer7982e902014-03-08 17:20:01 -0700848 return NULL;
849
Tejun Heoa0516612012-06-26 15:05:44 -0700850 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
Dave Jones708f04d2014-03-20 15:03:58 -0600851 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Christoph Hellwig287922e2015-10-30 20:57:30 +0800853 INIT_WORK(&q->timeout_work, blk_timeout_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 q->prep_rq_fn = NULL;
James Bottomley28018c22010-07-01 19:49:17 +0900856 q->unprep_rq_fn = NULL;
Tejun Heo60ea8222012-09-20 14:09:30 -0700857 q->queue_flags |= QUEUE_FLAG_DEFAULT;
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500858
859 /* Override internal queue lock with supplied lock pointer */
860 if (lock)
861 q->queue_lock = lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
Jens Axboef3b144a2009-03-06 08:48:33 +0100863 /*
864 * This also sets hw/phys segments, boundary and size
865 */
Jens Axboec20e8de2011-09-12 12:03:37 +0200866 blk_queue_make_request(q, blk_queue_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Alan Stern44ec9542007-02-20 11:01:57 -0500868 q->sg_reserved_size = INT_MAX;
869
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600870 /* Protect q->elevator from elevator_change */
871 mutex_lock(&q->sysfs_lock);
872
Tejun Heob82d4b12012-04-13 13:11:31 -0700873 /* init elevator */
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600874 if (elevator_init(q, NULL)) {
875 mutex_unlock(&q->sysfs_lock);
Dave Jones708f04d2014-03-20 15:03:58 -0600876 goto fail;
Tomoki Sekiyamaeb1c1602013-10-15 16:42:16 -0600877 }
878
879 mutex_unlock(&q->sysfs_lock);
880
Tejun Heob82d4b12012-04-13 13:11:31 -0700881 return q;
Dave Jones708f04d2014-03-20 15:03:58 -0600882
883fail:
Ming Leiba483382014-09-25 23:23:44 +0800884 blk_free_flush_queue(q->fq);
Dave Jones708f04d2014-03-20 15:03:58 -0600885 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
Mike Snitzer51514122011-11-23 10:59:13 +0100887EXPORT_SYMBOL(blk_init_allocated_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Tejun Heo09ac46c2011-12-14 00:33:38 +0100889bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100891 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100892 __blk_get_queue(q);
893 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 }
895
Tejun Heo09ac46c2011-12-14 00:33:38 +0100896 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
Jens Axboed86e0e82011-05-27 07:44:43 +0200898EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Tejun Heo5b788ce2012-06-04 20:40:59 -0700900static inline void blk_free_request(struct request_list *rl, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Tejun Heof1f8cc92011-12-14 00:33:42 +0100902 if (rq->cmd_flags & REQ_ELVPRIV) {
Tejun Heo5b788ce2012-06-04 20:40:59 -0700903 elv_put_request(rl->q, rq);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100904 if (rq->elv.icq)
Tejun Heo11a31222012-02-07 07:51:30 +0100905 put_io_context(rq->elv.icq->ioc);
Tejun Heof1f8cc92011-12-14 00:33:42 +0100906 }
907
Tejun Heo5b788ce2012-06-04 20:40:59 -0700908 mempool_free(rq, rl->rq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}
910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911/*
912 * ioc_batching returns true if the ioc is a valid batching request and
913 * should be given priority access to a request.
914 */
Jens Axboe165125e2007-07-24 09:28:11 +0200915static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
917 if (!ioc)
918 return 0;
919
920 /*
921 * Make sure the process is able to allocate at least 1 request
922 * even if the batch times out, otherwise we could theoretically
923 * lose wakeups.
924 */
925 return ioc->nr_batch_requests == q->nr_batching ||
926 (ioc->nr_batch_requests > 0
927 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
928}
929
930/*
931 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
932 * will cause the process to be a "batcher" on all queues in the system. This
933 * is the behaviour we want though - once it gets a wakeup it should be given
934 * a nice run.
935 */
Jens Axboe165125e2007-07-24 09:28:11 +0200936static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
938 if (!ioc || ioc_batching(q, ioc))
939 return;
940
941 ioc->nr_batch_requests = q->nr_batching;
942 ioc->last_waited = jiffies;
943}
944
Tejun Heo5b788ce2012-06-04 20:40:59 -0700945static void __freed_request(struct request_list *rl, int sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700947 struct request_queue *q = rl->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Tejun Heod40f75a2015-05-22 17:13:42 -0400949 if (rl->count[sync] < queue_congestion_off_threshold(q))
950 blk_clear_congested(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Jens Axboe1faa16d2009-04-06 14:48:01 +0200952 if (rl->count[sync] + 1 <= q->nr_requests) {
953 if (waitqueue_active(&rl->wait[sync]))
954 wake_up(&rl->wait[sync]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Tejun Heo5b788ce2012-06-04 20:40:59 -0700956 blk_clear_rl_full(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
958}
959
960/*
961 * A request has just been released. Account for it, update the full and
962 * congestion status, wake up any waiters. Called under q->queue_lock.
963 */
Mike Christiee6a40b02016-06-05 14:32:11 -0500964static void freed_request(struct request_list *rl, int op, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
Tejun Heo5b788ce2012-06-04 20:40:59 -0700966 struct request_queue *q = rl->q;
Mike Christied9d8c5c2016-06-05 14:32:16 -0500967 int sync = rw_is_sync(op, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700969 q->nr_rqs[sync]--;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200970 rl->count[sync]--;
Tejun Heo75eb6c32011-10-19 14:31:22 +0200971 if (flags & REQ_ELVPRIV)
Tejun Heo8a5ecdd2012-06-04 20:40:58 -0700972 q->nr_rqs_elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Tejun Heo5b788ce2012-06-04 20:40:59 -0700974 __freed_request(rl, sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Jens Axboe1faa16d2009-04-06 14:48:01 +0200976 if (unlikely(rl->starved[sync ^ 1]))
Tejun Heo5b788ce2012-06-04 20:40:59 -0700977 __freed_request(rl, sync ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600980int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
981{
982 struct request_list *rl;
Tejun Heod40f75a2015-05-22 17:13:42 -0400983 int on_thresh, off_thresh;
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600984
985 spin_lock_irq(q->queue_lock);
986 q->nr_requests = nr;
987 blk_queue_congestion_threshold(q);
Tejun Heod40f75a2015-05-22 17:13:42 -0400988 on_thresh = queue_congestion_on_threshold(q);
989 off_thresh = queue_congestion_off_threshold(q);
Jens Axboee3a2b3f2014-05-20 11:49:02 -0600990
991 blk_queue_for_each_rl(rl, q) {
Tejun Heod40f75a2015-05-22 17:13:42 -0400992 if (rl->count[BLK_RW_SYNC] >= on_thresh)
993 blk_set_congested(rl, BLK_RW_SYNC);
994 else if (rl->count[BLK_RW_SYNC] < off_thresh)
995 blk_clear_congested(rl, BLK_RW_SYNC);
996
997 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
998 blk_set_congested(rl, BLK_RW_ASYNC);
999 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1000 blk_clear_congested(rl, BLK_RW_ASYNC);
1001
Jens Axboee3a2b3f2014-05-20 11:49:02 -06001002 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1003 blk_set_rl_full(rl, BLK_RW_SYNC);
1004 } else {
1005 blk_clear_rl_full(rl, BLK_RW_SYNC);
1006 wake_up(&rl->wait[BLK_RW_SYNC]);
1007 }
1008
1009 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1010 blk_set_rl_full(rl, BLK_RW_ASYNC);
1011 } else {
1012 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1013 wake_up(&rl->wait[BLK_RW_ASYNC]);
1014 }
1015 }
1016
1017 spin_unlock_irq(q->queue_lock);
1018 return 0;
1019}
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021/*
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001022 * Determine if elevator data should be initialized when allocating the
1023 * request associated with @bio.
1024 */
1025static bool blk_rq_should_init_elevator(struct bio *bio)
1026{
1027 if (!bio)
1028 return true;
1029
1030 /*
1031 * Flush requests do not use the elevator so skip initialization.
1032 * This allows a request to share the flush and elevator data.
1033 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001034 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001035 return false;
1036
1037 return true;
1038}
1039
Tejun Heoda8303c2011-10-19 14:33:05 +02001040/**
Tejun Heo852c7882012-03-05 13:15:27 -08001041 * rq_ioc - determine io_context for request allocation
1042 * @bio: request being allocated is for this bio (can be %NULL)
1043 *
1044 * Determine io_context to use for request allocation for @bio. May return
1045 * %NULL if %current->io_context doesn't exist.
1046 */
1047static struct io_context *rq_ioc(struct bio *bio)
1048{
1049#ifdef CONFIG_BLK_CGROUP
1050 if (bio && bio->bi_ioc)
1051 return bio->bi_ioc;
1052#endif
1053 return current->io_context;
1054}
1055
1056/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001057 * __get_request - get a free request
Tejun Heo5b788ce2012-06-04 20:40:59 -07001058 * @rl: request list to allocate from
Mike Christiee6a40b02016-06-05 14:32:11 -05001059 * @op: REQ_OP_READ/REQ_OP_WRITE
1060 * @op_flags: rq_flag_bits
Tejun Heoda8303c2011-10-19 14:33:05 +02001061 * @bio: bio to allocate request for (can be %NULL)
1062 * @gfp_mask: allocation mask
1063 *
1064 * Get a free request from @q. This function may fail under memory
1065 * pressure or if @q is dead.
1066 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001067 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001068 * Returns ERR_PTR on failure, with @q->queue_lock held.
1069 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001071static struct request *__get_request(struct request_list *rl, int op,
1072 int op_flags, struct bio *bio,
1073 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Tejun Heo5b788ce2012-06-04 20:40:59 -07001075 struct request_queue *q = rl->q;
Tejun Heob6792812012-03-05 13:15:23 -08001076 struct request *rq;
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001077 struct elevator_type *et = q->elevator->type;
1078 struct io_context *ioc = rq_ioc(bio);
Tejun Heof1f8cc92011-12-14 00:33:42 +01001079 struct io_cq *icq = NULL;
Mike Christied9d8c5c2016-06-05 14:32:16 -05001080 const bool is_sync = rw_is_sync(op, op_flags) != 0;
Tejun Heo75eb6c32011-10-19 14:31:22 +02001081 int may_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Bart Van Assche3f3299d2012-11-28 13:42:38 +01001083 if (unlikely(blk_queue_dying(q)))
Joe Lawrencea492f072014-08-28 08:15:21 -06001084 return ERR_PTR(-ENODEV);
Tejun Heoda8303c2011-10-19 14:33:05 +02001085
Mike Christieba568ea2016-06-05 14:32:13 -05001086 may_queue = elv_may_queue(q, op, op_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001087 if (may_queue == ELV_MQUEUE_NO)
1088 goto rq_starved;
1089
Jens Axboe1faa16d2009-04-06 14:48:01 +02001090 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1091 if (rl->count[is_sync]+1 >= q->nr_requests) {
Tejun Heof2dbd762011-12-14 00:33:40 +01001092 /*
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001093 * The queue will fill after this allocation, so set
1094 * it as full, and mark this process as "batching".
1095 * This process will be allowed to complete a batch of
1096 * requests, others will be blocked.
1097 */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001098 if (!blk_rl_full(rl, is_sync)) {
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001099 ioc_set_batching(q, ioc);
Tejun Heo5b788ce2012-06-04 20:40:59 -07001100 blk_set_rl_full(rl, is_sync);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001101 } else {
1102 if (may_queue != ELV_MQUEUE_MUST
1103 && !ioc_batching(q, ioc)) {
1104 /*
1105 * The queue is full and the allocating
1106 * process is not a "batcher", and not
1107 * exempted by the IO scheduler
1108 */
Joe Lawrencea492f072014-08-28 08:15:21 -06001109 return ERR_PTR(-ENOMEM);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001110 }
1111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
Tejun Heod40f75a2015-05-22 17:13:42 -04001113 blk_set_congested(rl, is_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
1115
Jens Axboe082cf692005-06-28 16:35:11 +02001116 /*
1117 * Only allow batching queuers to allocate up to 50% over the defined
1118 * limit of requests, otherwise we could have thousands of requests
1119 * allocated with any setting of ->nr_requests
1120 */
Jens Axboe1faa16d2009-04-06 14:48:01 +02001121 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
Joe Lawrencea492f072014-08-28 08:15:21 -06001122 return ERR_PTR(-ENOMEM);
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001123
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001124 q->nr_rqs[is_sync]++;
Jens Axboe1faa16d2009-04-06 14:48:01 +02001125 rl->count[is_sync]++;
1126 rl->starved[is_sync] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001127
Tejun Heof1f8cc92011-12-14 00:33:42 +01001128 /*
1129 * Decide whether the new request will be managed by elevator. If
Mike Christiee6a40b02016-06-05 14:32:11 -05001130 * so, mark @op_flags and increment elvpriv. Non-zero elvpriv will
Tejun Heof1f8cc92011-12-14 00:33:42 +01001131 * prevent the current elevator from being destroyed until the new
1132 * request is freed. This guarantees icq's won't be destroyed and
1133 * makes creating new ones safe.
1134 *
1135 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1136 * it will be created after releasing queue_lock.
1137 */
Tejun Heod7325802012-03-05 13:14:58 -08001138 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
Mike Christiee6a40b02016-06-05 14:32:11 -05001139 op_flags |= REQ_ELVPRIV;
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001140 q->nr_rqs_elvpriv++;
Tejun Heof1f8cc92011-12-14 00:33:42 +01001141 if (et->icq_cache && ioc)
1142 icq = ioc_lookup_icq(ioc, q);
Mike Snitzer9d5a4e92011-02-11 11:05:46 +01001143 }
Tejun Heocb98fc82005-10-28 08:29:39 +02001144
Jens Axboef253b862010-10-24 22:06:02 +02001145 if (blk_queue_io_stat(q))
Mike Christiee6a40b02016-06-05 14:32:11 -05001146 op_flags |= REQ_IO_STAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 spin_unlock_irq(q->queue_lock);
1148
Tejun Heo29e2b092012-04-19 16:29:21 -07001149 /* allocate and init request */
Tejun Heo5b788ce2012-06-04 20:40:59 -07001150 rq = mempool_alloc(rl->rq_pool, gfp_mask);
Tejun Heo29e2b092012-04-19 16:29:21 -07001151 if (!rq)
Tejun Heob6792812012-03-05 13:15:23 -08001152 goto fail_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Tejun Heo29e2b092012-04-19 16:29:21 -07001154 blk_rq_init(q, rq);
Tejun Heoa0516612012-06-26 15:05:44 -07001155 blk_rq_set_rl(rq, rl);
Mike Christiee6a40b02016-06-05 14:32:11 -05001156 req_set_op_attrs(rq, op, op_flags | REQ_ALLOCED);
Tejun Heo29e2b092012-04-19 16:29:21 -07001157
Tejun Heoaaf7c682012-04-19 16:29:22 -07001158 /* init elvpriv */
Mike Christiee6a40b02016-06-05 14:32:11 -05001159 if (op_flags & REQ_ELVPRIV) {
Tejun Heoaaf7c682012-04-19 16:29:22 -07001160 if (unlikely(et->icq_cache && !icq)) {
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001161 if (ioc)
1162 icq = ioc_create_icq(ioc, q, gfp_mask);
Tejun Heoaaf7c682012-04-19 16:29:22 -07001163 if (!icq)
1164 goto fail_elvpriv;
Tejun Heo29e2b092012-04-19 16:29:21 -07001165 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001166
1167 rq->elv.icq = icq;
1168 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1169 goto fail_elvpriv;
1170
1171 /* @rq->elv.icq holds io_context until @rq is freed */
Tejun Heo29e2b092012-04-19 16:29:21 -07001172 if (icq)
1173 get_io_context(icq->ioc);
1174 }
Tejun Heoaaf7c682012-04-19 16:29:22 -07001175out:
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001176 /*
1177 * ioc may be NULL here, and ioc_batching will be false. That's
1178 * OK, if the queue is under the request limit then requests need
1179 * not count toward the nr_batch_requests limit. There will always
1180 * be some limit enforced by BLK_BATCH_TIME.
1181 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (ioc_batching(q, ioc))
1183 ioc->nr_batch_requests--;
Jens Axboe6728cb02008-01-31 13:03:55 +01001184
Mike Christiee6a40b02016-06-05 14:32:11 -05001185 trace_block_getrq(q, bio, op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 return rq;
Tejun Heob6792812012-03-05 13:15:23 -08001187
Tejun Heoaaf7c682012-04-19 16:29:22 -07001188fail_elvpriv:
1189 /*
1190 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1191 * and may fail indefinitely under memory pressure and thus
1192 * shouldn't stall IO. Treat this request as !elvpriv. This will
1193 * disturb iosched and blkcg but weird is bettern than dead.
1194 */
Robert Elliott7b2b10e2014-08-27 10:50:36 -05001195 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1196 __func__, dev_name(q->backing_dev_info.dev));
Tejun Heoaaf7c682012-04-19 16:29:22 -07001197
1198 rq->cmd_flags &= ~REQ_ELVPRIV;
1199 rq->elv.icq = NULL;
1200
1201 spin_lock_irq(q->queue_lock);
Tejun Heo8a5ecdd2012-06-04 20:40:58 -07001202 q->nr_rqs_elvpriv--;
Tejun Heoaaf7c682012-04-19 16:29:22 -07001203 spin_unlock_irq(q->queue_lock);
1204 goto out;
1205
Tejun Heob6792812012-03-05 13:15:23 -08001206fail_alloc:
1207 /*
1208 * Allocation failed presumably due to memory. Undo anything we
1209 * might have messed up.
1210 *
1211 * Allocating task should really be put onto the front of the wait
1212 * queue, but this is pretty rare.
1213 */
1214 spin_lock_irq(q->queue_lock);
Mike Christiee6a40b02016-06-05 14:32:11 -05001215 freed_request(rl, op, op_flags);
Tejun Heob6792812012-03-05 13:15:23 -08001216
1217 /*
1218 * in the very unlikely event that allocation failed and no
1219 * requests for this direction was pending, mark us starved so that
1220 * freeing of a request in the other direction will notice
1221 * us. another possible fix would be to split the rq mempool into
1222 * READ and WRITE
1223 */
1224rq_starved:
1225 if (unlikely(rl->count[is_sync] == 0))
1226 rl->starved[is_sync] = 1;
Joe Lawrencea492f072014-08-28 08:15:21 -06001227 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
1229
Tejun Heoda8303c2011-10-19 14:33:05 +02001230/**
Tejun Heoa06e05e2012-06-04 20:40:55 -07001231 * get_request - get a free request
Tejun Heoda8303c2011-10-19 14:33:05 +02001232 * @q: request_queue to allocate request from
Mike Christiee6a40b02016-06-05 14:32:11 -05001233 * @op: REQ_OP_READ/REQ_OP_WRITE
1234 * @op_flags: rq_flag_bits
Tejun Heoda8303c2011-10-19 14:33:05 +02001235 * @bio: bio to allocate request for (can be %NULL)
Tejun Heoa06e05e2012-06-04 20:40:55 -07001236 * @gfp_mask: allocation mask
Nick Piggind6344532005-06-28 20:45:14 -07001237 *
Mel Gormand0164ad2015-11-06 16:28:21 -08001238 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1239 * this function keeps retrying under memory pressure and fails iff @q is dead.
Tejun Heoda8303c2011-10-19 14:33:05 +02001240 *
Masanari Iidada3dae52014-09-09 01:27:23 +09001241 * Must be called with @q->queue_lock held and,
Joe Lawrencea492f072014-08-28 08:15:21 -06001242 * Returns ERR_PTR on failure, with @q->queue_lock held.
1243 * Returns request pointer on success, with @q->queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001245static struct request *get_request(struct request_queue *q, int op,
1246 int op_flags, struct bio *bio,
1247 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248{
Mike Christied9d8c5c2016-06-05 14:32:16 -05001249 const bool is_sync = rw_is_sync(op, op_flags) != 0;
Tejun Heoa06e05e2012-06-04 20:40:55 -07001250 DEFINE_WAIT(wait);
Tejun Heoa0516612012-06-26 15:05:44 -07001251 struct request_list *rl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 struct request *rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001253
1254 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001255retry:
Mike Christiee6a40b02016-06-05 14:32:11 -05001256 rq = __get_request(rl, op, op_flags, bio, gfp_mask);
Joe Lawrencea492f072014-08-28 08:15:21 -06001257 if (!IS_ERR(rq))
Tejun Heoa06e05e2012-06-04 20:40:55 -07001258 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Mel Gormand0164ad2015-11-06 16:28:21 -08001260 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
Tejun Heoa0516612012-06-26 15:05:44 -07001261 blk_put_rl(rl);
Joe Lawrencea492f072014-08-28 08:15:21 -06001262 return rq;
Tejun Heoa0516612012-06-26 15:05:44 -07001263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Tejun Heoa06e05e2012-06-04 20:40:55 -07001265 /* wait on @rl and retry */
1266 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1267 TASK_UNINTERRUPTIBLE);
Tejun Heoda8303c2011-10-19 14:33:05 +02001268
Mike Christiee6a40b02016-06-05 14:32:11 -05001269 trace_block_sleeprq(q, bio, op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Tejun Heoa06e05e2012-06-04 20:40:55 -07001271 spin_unlock_irq(q->queue_lock);
1272 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Tejun Heoa06e05e2012-06-04 20:40:55 -07001274 /*
1275 * After sleeping, we become a "batching" process and will be able
1276 * to allocate at least one request, and up to a big batch of them
1277 * for a small period time. See ioc_batching, ioc_set_batching
1278 */
Tejun Heoa06e05e2012-06-04 20:40:55 -07001279 ioc_set_batching(q, current->io_context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
Tejun Heoa06e05e2012-06-04 20:40:55 -07001281 spin_lock_irq(q->queue_lock);
1282 finish_wait(&rl->wait[is_sync], &wait);
Jens Axboe2056a782006-03-23 20:00:26 +01001283
Tejun Heoa06e05e2012-06-04 20:40:55 -07001284 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
Jens Axboe320ae512013-10-24 09:20:05 +01001287static struct request *blk_old_get_request(struct request_queue *q, int rw,
1288 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 struct request *rq;
1291
1292 BUG_ON(rw != READ && rw != WRITE);
1293
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001294 /* create ioc upfront */
1295 create_io_context(gfp_mask, q->node);
1296
Nick Piggind6344532005-06-28 20:45:14 -07001297 spin_lock_irq(q->queue_lock);
Mike Christiee6a40b02016-06-05 14:32:11 -05001298 rq = get_request(q, rw, 0, NULL, gfp_mask);
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001299 if (IS_ERR(rq)) {
Tejun Heoda8303c2011-10-19 14:33:05 +02001300 spin_unlock_irq(q->queue_lock);
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001301 return rq;
1302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Christoph Hellwig0c4de0f2016-07-19 11:31:50 +02001304 /* q->queue_lock is unlocked at this point */
1305 rq->__data_len = 0;
1306 rq->__sector = (sector_t) -1;
1307 rq->bio = rq->biotail = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 return rq;
1309}
Jens Axboe320ae512013-10-24 09:20:05 +01001310
1311struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1312{
1313 if (q->mq_ops)
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01001314 return blk_mq_alloc_request(q, rw,
1315 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1316 0 : BLK_MQ_REQ_NOWAIT);
Jens Axboe320ae512013-10-24 09:20:05 +01001317 else
1318 return blk_old_get_request(q, rw, gfp_mask);
1319}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320EXPORT_SYMBOL(blk_get_request);
1321
1322/**
Masanari Iidada3dae52014-09-09 01:27:23 +09001323 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
Jens Axboef27b0872014-06-06 07:57:37 -06001324 * @rq: request to be initialized
1325 *
1326 */
1327void blk_rq_set_block_pc(struct request *rq)
1328{
1329 rq->cmd_type = REQ_TYPE_BLOCK_PC;
Jens Axboef27b0872014-06-06 07:57:37 -06001330 memset(rq->__cmd, 0, sizeof(rq->__cmd));
Jens Axboef27b0872014-06-06 07:57:37 -06001331}
1332EXPORT_SYMBOL(blk_rq_set_block_pc);
1333
1334/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * blk_requeue_request - put a request back on queue
1336 * @q: request queue where request should be inserted
1337 * @rq: request to be inserted
1338 *
1339 * Description:
1340 * Drivers often keep queueing requests until the hardware cannot accept
1341 * more, when that condition happens we need to put the request back
1342 * on the queue. Must be called with queue lock held.
1343 */
Jens Axboe165125e2007-07-24 09:28:11 +02001344void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Jens Axboe242f9dc2008-09-14 05:55:09 -07001346 blk_delete_timer(rq);
1347 blk_clear_rq_complete(rq);
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001348 trace_block_rq_requeue(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01001349
Christoph Hellwig125c99b2014-11-03 12:47:47 +01001350 if (rq->cmd_flags & REQ_QUEUED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 blk_queue_end_tag(q, rq);
1352
James Bottomleyba396a62009-05-27 14:17:08 +02001353 BUG_ON(blk_queued_rq(rq));
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 elv_requeue_request(q, rq);
1356}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357EXPORT_SYMBOL(blk_requeue_request);
1358
Jens Axboe73c10102011-03-08 13:19:51 +01001359static void add_acct_request(struct request_queue *q, struct request *rq,
1360 int where)
1361{
Jens Axboe320ae512013-10-24 09:20:05 +01001362 blk_account_io_start(rq, true);
Jens Axboe7eaceac2011-03-10 08:52:07 +01001363 __elv_add_request(q, rq, where);
Jens Axboe73c10102011-03-08 13:19:51 +01001364}
1365
Tejun Heo074a7ac2008-08-25 19:56:14 +09001366static void part_round_stats_single(int cpu, struct hd_struct *part,
1367 unsigned long now)
1368{
Jens Axboe7276d022014-05-09 15:48:23 -06001369 int inflight;
1370
Tejun Heo074a7ac2008-08-25 19:56:14 +09001371 if (now == part->stamp)
1372 return;
1373
Jens Axboe7276d022014-05-09 15:48:23 -06001374 inflight = part_in_flight(part);
1375 if (inflight) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001376 __part_stat_add(cpu, part, time_in_queue,
Jens Axboe7276d022014-05-09 15:48:23 -06001377 inflight * (now - part->stamp));
Tejun Heo074a7ac2008-08-25 19:56:14 +09001378 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1379 }
1380 part->stamp = now;
1381}
1382
1383/**
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001384 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1385 * @cpu: cpu number for stats access
1386 * @part: target partition
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 *
1388 * The average IO queue length and utilisation statistics are maintained
1389 * by observing the current state of the queue length and the amount of
1390 * time it has been in this state for.
1391 *
1392 * Normally, that accounting is done on IO completion, but that can result
1393 * in more than a second's worth of IO being accounted for within any one
1394 * second, leading to >100% utilisation. To deal with that, we call this
1395 * function to do a round-off before returning the results when reading
1396 * /proc/diskstats. This accounts immediately for all queue usage up to
1397 * the current jiffies and restarts the counters again.
1398 */
Tejun Heoc9959052008-08-25 19:47:21 +09001399void part_round_stats(int cpu, struct hd_struct *part)
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001400{
1401 unsigned long now = jiffies;
1402
Tejun Heo074a7ac2008-08-25 19:56:14 +09001403 if (part->partno)
1404 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1405 part_round_stats_single(cpu, part, now);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001406}
Tejun Heo074a7ac2008-08-25 19:56:14 +09001407EXPORT_SYMBOL_GPL(part_round_stats);
Jerome Marchand6f2576a2008-02-08 11:04:35 +01001408
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01001409#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08001410static void blk_pm_put_request(struct request *rq)
1411{
1412 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1413 pm_runtime_mark_last_busy(rq->q->dev);
1414}
1415#else
1416static inline void blk_pm_put_request(struct request *rq) {}
1417#endif
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419/*
1420 * queue lock must be held
1421 */
Jens Axboe165125e2007-07-24 09:28:11 +02001422void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (unlikely(!q))
1425 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Christoph Hellwig6f5ba582014-02-07 10:22:37 -08001427 if (q->mq_ops) {
1428 blk_mq_free_request(req);
1429 return;
1430 }
1431
Lin Mingc8158812013-03-23 11:42:27 +08001432 blk_pm_put_request(req);
1433
Tejun Heo8922e162005-10-20 16:23:44 +02001434 elv_completed_request(q, req);
1435
Boaz Harrosh1cd96c22009-03-24 12:35:07 +01001436 /* this is a bio leak */
1437 WARN_ON(req->bio != NULL);
1438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 /*
1440 * Request may not have originated from ll_rw_blk. if not,
1441 * it didn't come out of our reserved rq pools
1442 */
Jens Axboe49171e52006-08-10 08:59:11 +02001443 if (req->cmd_flags & REQ_ALLOCED) {
Tejun Heo75eb6c32011-10-19 14:31:22 +02001444 unsigned int flags = req->cmd_flags;
Mike Christiee6a40b02016-06-05 14:32:11 -05001445 int op = req_op(req);
Tejun Heoa0516612012-06-26 15:05:44 -07001446 struct request_list *rl = blk_rq_rl(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe360f92c2014-04-09 20:27:01 -06001449 BUG_ON(ELV_ON_HASH(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Tejun Heoa0516612012-06-26 15:05:44 -07001451 blk_free_request(rl, req);
Mike Christiee6a40b02016-06-05 14:32:11 -05001452 freed_request(rl, op, flags);
Tejun Heoa0516612012-06-26 15:05:44 -07001453 blk_put_rl(rl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
1455}
Mike Christie6e39b69e2005-11-11 05:30:24 -06001456EXPORT_SYMBOL_GPL(__blk_put_request);
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458void blk_put_request(struct request *req)
1459{
Jens Axboe165125e2007-07-24 09:28:11 +02001460 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
Jens Axboe320ae512013-10-24 09:20:05 +01001462 if (q->mq_ops)
1463 blk_mq_free_request(req);
1464 else {
1465 unsigned long flags;
1466
1467 spin_lock_irqsave(q->queue_lock, flags);
1468 __blk_put_request(q, req);
1469 spin_unlock_irqrestore(q->queue_lock, flags);
1470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472EXPORT_SYMBOL(blk_put_request);
1473
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001474/**
1475 * blk_add_request_payload - add a payload to a request
1476 * @rq: request to update
1477 * @page: page backing the payload
Ming Lin37e58232016-03-22 00:24:44 -07001478 * @offset: offset in page
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001479 * @len: length of the payload.
1480 *
1481 * This allows to later add a payload to an already submitted request by
1482 * a block driver. The driver needs to take care of freeing the payload
1483 * itself.
1484 *
1485 * Note that this is a quite horrible hack and nothing but handling of
1486 * discard requests should ever use it.
1487 */
1488void blk_add_request_payload(struct request *rq, struct page *page,
Ming Lin37e58232016-03-22 00:24:44 -07001489 int offset, unsigned int len)
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001490{
1491 struct bio *bio = rq->bio;
1492
1493 bio->bi_io_vec->bv_page = page;
Ming Lin37e58232016-03-22 00:24:44 -07001494 bio->bi_io_vec->bv_offset = offset;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001495 bio->bi_io_vec->bv_len = len;
1496
Kent Overstreet4f024f32013-10-11 15:44:27 -07001497 bio->bi_iter.bi_size = len;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001498 bio->bi_vcnt = 1;
1499 bio->bi_phys_segments = 1;
1500
1501 rq->__data_len = rq->resid_len = len;
1502 rq->nr_phys_segments = 1;
Christoph Hellwig66ac0282010-06-18 16:59:42 +02001503}
1504EXPORT_SYMBOL_GPL(blk_add_request_payload);
1505
Jens Axboe320ae512013-10-24 09:20:05 +01001506bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1507 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001508{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001509 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001510
Jens Axboe73c10102011-03-08 13:19:51 +01001511 if (!ll_back_merge_fn(q, req, bio))
1512 return false;
1513
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001514 trace_block_bio_backmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001515
1516 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1517 blk_rq_set_mixed_merge(req);
1518
1519 req->biotail->bi_next = bio;
1520 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001521 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001522 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1523
Jens Axboe320ae512013-10-24 09:20:05 +01001524 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001525 return true;
1526}
1527
Jens Axboe320ae512013-10-24 09:20:05 +01001528bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1529 struct bio *bio)
Jens Axboe73c10102011-03-08 13:19:51 +01001530{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001531 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +01001532
Jens Axboe73c10102011-03-08 13:19:51 +01001533 if (!ll_front_merge_fn(q, req, bio))
1534 return false;
1535
Tejun Heo8c1cf6b2013-01-11 13:06:34 -08001536 trace_block_bio_frontmerge(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001537
1538 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1539 blk_rq_set_mixed_merge(req);
1540
Jens Axboe73c10102011-03-08 13:19:51 +01001541 bio->bi_next = req->bio;
1542 req->bio = bio;
1543
Kent Overstreet4f024f32013-10-11 15:44:27 -07001544 req->__sector = bio->bi_iter.bi_sector;
1545 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +01001546 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1547
Jens Axboe320ae512013-10-24 09:20:05 +01001548 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +01001549 return true;
1550}
1551
Tejun Heobd87b582011-10-19 14:33:08 +02001552/**
Jens Axboe320ae512013-10-24 09:20:05 +01001553 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +02001554 * @q: request_queue new bio is being queued at
1555 * @bio: new bio being queued
1556 * @request_count: out parameter for number of traversed plugged requests
Randy Dunlapccc26002015-10-30 18:36:16 -07001557 * @same_queue_rq: pointer to &struct request that gets filled in when
1558 * another request associated with @q is found on the plug list
1559 * (optional, may be %NULL)
Tejun Heobd87b582011-10-19 14:33:08 +02001560 *
1561 * Determine whether @bio being queued on @q can be merged with a request
1562 * on %current's plugged list. Returns %true if merge was successful,
1563 * otherwise %false.
1564 *
Tejun Heo07c2bd32012-02-08 09:19:42 +01001565 * Plugging coalesces IOs from the same issuer for the same purpose without
1566 * going through @q->queue_lock. As such it's more of an issuing mechanism
1567 * than scheduling, and the request, while may have elvpriv data, is not
1568 * added on the elevator at this point. In addition, we don't have
1569 * reliable access to the elevator outside queue lock. Only check basic
1570 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -05001571 *
1572 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +01001573 */
Jens Axboe320ae512013-10-24 09:20:05 +01001574bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Shaohua Li5b3f3412015-05-08 10:51:33 -07001575 unsigned int *request_count,
1576 struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +01001577{
1578 struct blk_plug *plug;
1579 struct request *rq;
1580 bool ret = false;
Shaohua Li92f399c2013-10-29 12:01:03 -06001581 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +01001582
Tejun Heobd87b582011-10-19 14:33:08 +02001583 plug = current->plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001584 if (!plug)
1585 goto out;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001586 *request_count = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01001587
Shaohua Li92f399c2013-10-29 12:01:03 -06001588 if (q->mq_ops)
1589 plug_list = &plug->mq_list;
1590 else
1591 plug_list = &plug->list;
1592
1593 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Jens Axboe73c10102011-03-08 13:19:51 +01001594 int el_ret;
1595
Shaohua Li5b3f3412015-05-08 10:51:33 -07001596 if (rq->q == q) {
Shaohua Li1b2e19f2012-04-06 11:37:47 -06001597 (*request_count)++;
Shaohua Li5b3f3412015-05-08 10:51:33 -07001598 /*
1599 * Only blk-mq multiple hardware queues case checks the
1600 * rq in the same queue, there should be only one such
1601 * rq in a queue
1602 **/
1603 if (same_queue_rq)
1604 *same_queue_rq = rq;
1605 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001606
Tejun Heo07c2bd32012-02-08 09:19:42 +01001607 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +01001608 continue;
1609
Tejun Heo050c8ea2012-02-08 09:19:38 +01001610 el_ret = blk_try_merge(rq, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001611 if (el_ret == ELEVATOR_BACK_MERGE) {
1612 ret = bio_attempt_back_merge(q, rq, bio);
1613 if (ret)
1614 break;
1615 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1616 ret = bio_attempt_front_merge(q, rq, bio);
1617 if (ret)
1618 break;
1619 }
1620 }
1621out:
1622 return ret;
1623}
1624
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001625unsigned int blk_plug_queued_count(struct request_queue *q)
1626{
1627 struct blk_plug *plug;
1628 struct request *rq;
1629 struct list_head *plug_list;
1630 unsigned int ret = 0;
1631
1632 plug = current->plug;
1633 if (!plug)
1634 goto out;
1635
1636 if (q->mq_ops)
1637 plug_list = &plug->mq_list;
1638 else
1639 plug_list = &plug->list;
1640
1641 list_for_each_entry(rq, plug_list, queuelist) {
1642 if (rq->q == q)
1643 ret++;
1644 }
1645out:
1646 return ret;
1647}
1648
Jens Axboe86db1e22008-01-29 14:53:40 +01001649void init_request_from_bio(struct request *req, struct bio *bio)
Tejun Heo52d9e672006-01-06 09:49:58 +01001650{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001651 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01001652
Jens Axboe1eff9d32016-08-05 15:35:16 -06001653 req->cmd_flags |= bio->bi_opf & REQ_COMMON_MASK;
1654 if (bio->bi_opf & REQ_RAHEAD)
Tejun Heoa82afdf2009-07-03 17:48:16 +09001655 req->cmd_flags |= REQ_FAILFAST_MASK;
Jens Axboeb31dc662006-06-13 08:26:10 +02001656
Tejun Heo52d9e672006-01-06 09:49:58 +01001657 req->errors = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001658 req->__sector = bio->bi_iter.bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01001659 req->ioprio = bio_prio(bio);
NeilBrownbc1c56f2007-08-16 13:31:30 +02001660 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01001661}
1662
Jens Axboedece1632015-11-05 10:41:16 -07001663static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664{
Jens Axboe1eff9d32016-08-05 15:35:16 -06001665 const bool sync = !!(bio->bi_opf & REQ_SYNC);
Jens Axboe73c10102011-03-08 13:19:51 +01001666 struct blk_plug *plug;
Mike Christiee6a40b02016-06-05 14:32:11 -05001667 int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
Jens Axboe73c10102011-03-08 13:19:51 +01001668 struct request *req;
Shaohua Li56ebdaf2011-08-24 16:04:34 +02001669 unsigned int request_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 /*
1672 * low level driver can indicate that it wants pages above a
1673 * certain limit bounced to low memory (ie for highmem, or even
1674 * ISA dma in theory)
1675 */
1676 blk_queue_bounce(q, &bio);
1677
Junichi Nomura23688bf2015-12-22 10:23:44 -07001678 blk_queue_split(q, &bio, q->bio_split);
1679
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001680 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001681 bio->bi_error = -EIO;
1682 bio_endio(bio);
Jens Axboedece1632015-11-05 10:41:16 -07001683 return BLK_QC_T_NONE;
Darrick J. Wongffecfd12013-02-21 16:42:55 -08001684 }
1685
Jens Axboe1eff9d32016-08-05 15:35:16 -06001686 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
Jens Axboe73c10102011-03-08 13:19:51 +01001687 spin_lock_irq(q->queue_lock);
Tejun Heoae1b1532011-01-25 12:43:54 +01001688 where = ELEVATOR_INSERT_FLUSH;
Tejun Heo28e7d182010-09-03 11:56:16 +02001689 goto get_rq;
1690 }
1691
Jens Axboe73c10102011-03-08 13:19:51 +01001692 /*
1693 * Check if we can merge with the plugged list before grabbing
1694 * any locks.
1695 */
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001696 if (!blk_queue_nomerges(q)) {
1697 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
Jens Axboedece1632015-11-05 10:41:16 -07001698 return BLK_QC_T_NONE;
Jeff Moyer0809e3a2015-10-20 23:13:51 +08001699 } else
1700 request_count = blk_plug_queued_count(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001701
1702 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 el_ret = elv_merge(q, &req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001705 if (el_ret == ELEVATOR_BACK_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001706 if (bio_attempt_back_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001707 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001708 if (!attempt_back_merge(q, req))
1709 elv_merged_request(q, req, el_ret);
1710 goto out_unlock;
Tejun Heo80a761f2009-07-03 17:48:17 +09001711 }
Jens Axboe73c10102011-03-08 13:19:51 +01001712 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
Jens Axboe73c10102011-03-08 13:19:51 +01001713 if (bio_attempt_front_merge(q, req, bio)) {
Tejun Heo07c2bd32012-02-08 09:19:42 +01001714 elv_bio_merged(q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +01001715 if (!attempt_front_merge(q, req))
1716 elv_merged_request(q, req, el_ret);
1717 goto out_unlock;
1718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 }
1720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07001722 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01001723 * This sync check and mask will be re-done in init_request_from_bio(),
1724 * but we need to set it earlier to expose the sync flag to the
1725 * rq allocator and io schedulers.
1726 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01001727 if (sync)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001728 rw_flags |= REQ_SYNC;
Jens Axboe7749a8d2006-12-13 13:02:26 +01001729
1730 /*
Jens Axboeb8269db2016-06-09 15:47:29 -06001731 * Add in META/PRIO flags, if set, before we get to the IO scheduler
1732 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001733 rw_flags |= (bio->bi_opf & (REQ_META | REQ_PRIO));
Jens Axboeb8269db2016-06-09 15:47:29 -06001734
1735 /*
Nick Piggin450991b2005-06-28 20:45:13 -07001736 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07001737 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07001738 */
Mike Christiee6a40b02016-06-05 14:32:11 -05001739 req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
Joe Lawrencea492f072014-08-28 08:15:21 -06001740 if (IS_ERR(req)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001741 bio->bi_error = PTR_ERR(req);
1742 bio_endio(bio);
Tejun Heoda8303c2011-10-19 14:33:05 +02001743 goto out_unlock;
1744 }
Nick Piggind6344532005-06-28 20:45:14 -07001745
Nick Piggin450991b2005-06-28 20:45:13 -07001746 /*
1747 * After dropping the lock and possibly sleeping here, our request
1748 * may now be mergeable after it had proven unmergeable (above).
1749 * We don't worry about that case for efficiency. It won't happen
1750 * often, and the elevators are able to handle it.
1751 */
Tejun Heo52d9e672006-01-06 09:49:58 +01001752 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
Tao Ma9562ad92011-10-24 16:11:30 +02001754 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
Jens Axboe11ccf112011-07-26 15:01:15 +02001755 req->cpu = raw_smp_processor_id();
Tejun Heodd831002010-09-03 11:56:16 +02001756
Jens Axboe73c10102011-03-08 13:19:51 +01001757 plug = current->plug;
Jens Axboe721a9602011-03-09 11:56:30 +01001758 if (plug) {
Jens Axboedc6d36c2011-04-12 10:28:28 +02001759 /*
1760 * If this is the first request added after a plug, fire
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001761 * of a plug trace.
Jens Axboedc6d36c2011-04-12 10:28:28 +02001762 */
Jianpeng Ma7aef2e72013-09-11 13:21:07 -06001763 if (!request_count)
Jens Axboedc6d36c2011-04-12 10:28:28 +02001764 trace_block_plug(q);
Shaohua Li3540d5e2011-11-16 09:21:50 +01001765 else {
Shaohua Li019ceb72011-11-16 09:21:50 +01001766 if (request_count >= BLK_MAX_REQUEST_COUNT) {
Shaohua Li3540d5e2011-11-16 09:21:50 +01001767 blk_flush_plug_list(plug, false);
Shaohua Li019ceb72011-11-16 09:21:50 +01001768 trace_block_plug(q);
1769 }
Jens Axboe73c10102011-03-08 13:19:51 +01001770 }
Shaohua Lia6327162011-08-24 16:04:32 +02001771 list_add_tail(&req->queuelist, &plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01001772 blk_account_io_start(req, true);
Jens Axboe73c10102011-03-08 13:19:51 +01001773 } else {
1774 spin_lock_irq(q->queue_lock);
1775 add_acct_request(q, req, where);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02001776 __blk_run_queue(q);
Jens Axboe73c10102011-03-08 13:19:51 +01001777out_unlock:
1778 spin_unlock_irq(q->queue_lock);
1779 }
Jens Axboedece1632015-11-05 10:41:16 -07001780
1781 return BLK_QC_T_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
1783
1784/*
1785 * If bio->bi_dev is a partition, remap the location
1786 */
1787static inline void blk_partition_remap(struct bio *bio)
1788{
1789 struct block_device *bdev = bio->bi_bdev;
1790
Jens Axboebf2de6f2007-09-27 13:01:25 +02001791 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01001793
Kent Overstreet4f024f32013-10-11 15:44:27 -07001794 bio->bi_iter.bi_sector += p->start_sect;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02001796
Mike Snitzerd07335e2010-11-16 12:52:38 +01001797 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1798 bdev->bd_dev,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001799 bio->bi_iter.bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801}
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803static void handle_bad_sector(struct bio *bio)
1804{
1805 char b[BDEVNAME_SIZE];
1806
1807 printk(KERN_INFO "attempt to access beyond end of device\n");
Mike Christie6296b962016-06-05 14:32:21 -05001808 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 bdevname(bio->bi_bdev, b),
Jens Axboe1eff9d32016-08-05 15:35:16 -06001810 bio->bi_opf,
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001811 (unsigned long long)bio_end_sector(bio),
Mike Snitzer77304d22010-11-08 14:39:12 +01001812 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
1814
Akinobu Mitac17bb492006-12-08 02:39:46 -08001815#ifdef CONFIG_FAIL_MAKE_REQUEST
1816
1817static DECLARE_FAULT_ATTR(fail_make_request);
1818
1819static int __init setup_fail_make_request(char *str)
1820{
1821 return setup_fault_attr(&fail_make_request, str);
1822}
1823__setup("fail_make_request=", setup_fail_make_request);
1824
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001825static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001826{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001827 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001828}
1829
1830static int __init fail_make_request_debugfs(void)
1831{
Akinobu Mitadd48c082011-08-03 16:21:01 -07001832 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1833 NULL, &fail_make_request);
1834
Duan Jiong21f9fcd2014-04-11 15:58:56 +08001835 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -08001836}
1837
1838late_initcall(fail_make_request_debugfs);
1839
1840#else /* CONFIG_FAIL_MAKE_REQUEST */
1841
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001842static inline bool should_fail_request(struct hd_struct *part,
1843 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -08001844{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001845 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -08001846}
1847
1848#endif /* CONFIG_FAIL_MAKE_REQUEST */
1849
Jens Axboec07e2b42007-07-18 13:27:58 +02001850/*
1851 * Check whether this bio extends beyond the end of the device.
1852 */
1853static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1854{
1855 sector_t maxsector;
1856
1857 if (!nr_sectors)
1858 return 0;
1859
1860 /* Test device or partition size, when known. */
Mike Snitzer77304d22010-11-08 14:39:12 +01001861 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
Jens Axboec07e2b42007-07-18 13:27:58 +02001862 if (maxsector) {
Kent Overstreet4f024f32013-10-11 15:44:27 -07001863 sector_t sector = bio->bi_iter.bi_sector;
Jens Axboec07e2b42007-07-18 13:27:58 +02001864
1865 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1866 /*
1867 * This may well happen - the kernel calls bread()
1868 * without checking the size of the device, e.g., when
1869 * mounting a device.
1870 */
1871 handle_bad_sector(bio);
1872 return 1;
1873 }
1874 }
1875
1876 return 0;
1877}
1878
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001879static noinline_for_stack bool
1880generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881{
Jens Axboe165125e2007-07-24 09:28:11 +02001882 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001883 int nr_sectors = bio_sectors(bio);
Jens Axboe51fd77b2007-11-02 08:49:08 +01001884 int err = -EIO;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001885 char b[BDEVNAME_SIZE];
1886 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
1888 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Jens Axboec07e2b42007-07-18 13:27:58 +02001890 if (bio_check_eod(bio, nr_sectors))
1891 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001893 q = bdev_get_queue(bio->bi_bdev);
1894 if (unlikely(!q)) {
1895 printk(KERN_ERR
1896 "generic_make_request: Trying to access "
1897 "nonexistent block-device %s (%Lu)\n",
1898 bdevname(bio->bi_bdev, b),
Kent Overstreet4f024f32013-10-11 15:44:27 -07001899 (long long) bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001900 goto end_io;
1901 }
1902
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001903 part = bio->bi_bdev->bd_part;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001904 if (should_fail_request(part, bio->bi_iter.bi_size) ||
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001905 should_fail_request(&part_to_disk(part)->part0,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001906 bio->bi_iter.bi_size))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001907 goto end_io;
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 /*
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001910 * If this device has partitions, remap block n
1911 * of partition p to block n+start(p) of the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001913 blk_partition_remap(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001915 if (bio_check_eod(bio, nr_sectors))
1916 goto end_io;
1917
1918 /*
1919 * Filter flush bio's early so that make_request based
1920 * drivers without flush support don't have to worry
1921 * about them.
1922 */
Jens Axboe1eff9d32016-08-05 15:35:16 -06001923 if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
Jens Axboec888a8f2016-04-13 13:33:19 -06001924 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
Jens Axboe1eff9d32016-08-05 15:35:16 -06001925 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001926 if (!nr_sectors) {
1927 err = 0;
Tejun Heoa7384672008-11-28 13:32:03 +09001928 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
Christoph Hellwig288dab82016-06-09 16:00:36 +02001932 switch (bio_op(bio)) {
1933 case REQ_OP_DISCARD:
1934 if (!blk_queue_discard(q))
1935 goto not_supported;
1936 break;
1937 case REQ_OP_SECURE_ERASE:
1938 if (!blk_queue_secure_erase(q))
1939 goto not_supported;
1940 break;
1941 case REQ_OP_WRITE_SAME:
1942 if (!bdev_write_same(bio->bi_bdev))
1943 goto not_supported;
1944 break;
1945 default:
1946 break;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
Tejun Heo7f4b35d2012-06-04 20:40:56 -07001949 /*
1950 * Various block parts want %current->io_context and lazy ioc
1951 * allocation ends up trading a lot of pain for a small amount of
1952 * memory. Just allocate it upfront. This may fail and block
1953 * layer knows how to live with it.
1954 */
1955 create_io_context(GFP_ATOMIC, q->node);
1956
Tejun Heoae118892015-08-18 14:55:20 -07001957 if (!blkcg_bio_issue_check(q, bio))
1958 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001959
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001960 trace_block_bio_queue(q, bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001961 return true;
Tejun Heoa7384672008-11-28 13:32:03 +09001962
Christoph Hellwig288dab82016-06-09 16:00:36 +02001963not_supported:
1964 err = -EOPNOTSUPP;
Tejun Heoa7384672008-11-28 13:32:03 +09001965end_io:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001966 bio->bi_error = err;
1967 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001968 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001971/**
1972 * generic_make_request - hand a buffer to its device driver for I/O
1973 * @bio: The bio describing the location in memory and on the device.
1974 *
1975 * generic_make_request() is used to make I/O requests of block
1976 * devices. It is passed a &struct bio, which describes the I/O that needs
1977 * to be done.
1978 *
1979 * generic_make_request() does not return any status. The
1980 * success/failure status of the request, along with notification of
1981 * completion, is delivered asynchronously through the bio->bi_end_io
1982 * function described (one day) else where.
1983 *
1984 * The caller of generic_make_request must make sure that bi_io_vec
1985 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1986 * set to describe the device address, and the
1987 * bi_end_io and optionally bi_private are set to describe how
1988 * completion notification should be signaled.
1989 *
1990 * generic_make_request and the drivers it calls may use bi_next if this
1991 * bio happens to be merged with someone else, and may resubmit the bio to
1992 * a lower device by calling into generic_make_request recursively, which
1993 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001994 */
Jens Axboedece1632015-11-05 10:41:16 -07001995blk_qc_t generic_make_request(struct bio *bio)
Neil Brownd89d8792007-05-01 09:53:42 +02001996{
NeilBrown5959cde2017-03-10 17:00:47 +11001997 /*
1998 * bio_list_on_stack[0] contains bios submitted by the current
1999 * make_request_fn.
2000 * bio_list_on_stack[1] contains bios that were submitted before
2001 * the current make_request_fn, but that haven't been processed
2002 * yet.
2003 */
2004 struct bio_list bio_list_on_stack[2];
Jens Axboedece1632015-11-05 10:41:16 -07002005 blk_qc_t ret = BLK_QC_T_NONE;
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002006
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002007 if (!generic_make_request_checks(bio))
Jens Axboedece1632015-11-05 10:41:16 -07002008 goto out;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002009
2010 /*
2011 * We only want one ->make_request_fn to be active at a time, else
2012 * stack usage with stacked devices could be a problem. So use
2013 * current->bio_list to keep a list of requests submited by a
2014 * make_request_fn function. current->bio_list is also used as a
2015 * flag to say if generic_make_request is currently active in this
2016 * task or not. If it is NULL, then no make_request is active. If
2017 * it is non-NULL, then a make_request is active, and new requests
2018 * should be added at the tail
2019 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002020 if (current->bio_list) {
NeilBrown5959cde2017-03-10 17:00:47 +11002021 bio_list_add(&current->bio_list[0], bio);
Jens Axboedece1632015-11-05 10:41:16 -07002022 goto out;
Neil Brownd89d8792007-05-01 09:53:42 +02002023 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002024
Neil Brownd89d8792007-05-01 09:53:42 +02002025 /* following loop may be a bit non-obvious, and so deserves some
2026 * explanation.
2027 * Before entering the loop, bio->bi_next is NULL (as all callers
2028 * ensure that) so we have a list with a single bio.
2029 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002030 * we assign bio_list to a pointer to the bio_list_on_stack,
2031 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002032 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02002033 * through a recursive call to generic_make_request. If it
2034 * did, we find a non-NULL value in bio_list and re-enter the loop
2035 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002036 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002037 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02002038 */
2039 BUG_ON(bio->bi_next);
NeilBrown5959cde2017-03-10 17:00:47 +11002040 bio_list_init(&bio_list_on_stack[0]);
2041 current->bio_list = bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02002042 do {
Christoph Hellwig27a84d52011-09-15 14:01:40 +02002043 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2044
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +01002045 if (likely(blk_queue_enter(q, false) == 0)) {
NeilBrownd5986e02017-03-08 07:38:05 +11002046 struct bio_list lower, same;
2047
2048 /* Create a fresh bio_list for all subordinate requests */
NeilBrown5959cde2017-03-10 17:00:47 +11002049 bio_list_on_stack[1] = bio_list_on_stack[0];
2050 bio_list_init(&bio_list_on_stack[0]);
Jens Axboedece1632015-11-05 10:41:16 -07002051 ret = q->make_request_fn(q, bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002052
2053 blk_queue_exit(q);
2054
NeilBrownd5986e02017-03-08 07:38:05 +11002055 /* sort new bios into those for a lower level
2056 * and those for the same level
2057 */
2058 bio_list_init(&lower);
2059 bio_list_init(&same);
NeilBrown5959cde2017-03-10 17:00:47 +11002060 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
NeilBrownd5986e02017-03-08 07:38:05 +11002061 if (q == bdev_get_queue(bio->bi_bdev))
2062 bio_list_add(&same, bio);
2063 else
2064 bio_list_add(&lower, bio);
2065 /* now assemble so we handle the lowest level first */
NeilBrown5959cde2017-03-10 17:00:47 +11002066 bio_list_merge(&bio_list_on_stack[0], &lower);
2067 bio_list_merge(&bio_list_on_stack[0], &same);
2068 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
Dan Williams3ef28e82015-10-21 13:20:12 -04002069 } else {
Dan Williams3ef28e82015-10-21 13:20:12 -04002070 bio_io_error(bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04002071 }
NeilBrown5959cde2017-03-10 17:00:47 +11002072 bio = bio_list_pop(&bio_list_on_stack[0]);
Neil Brownd89d8792007-05-01 09:53:42 +02002073 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01002074 current->bio_list = NULL; /* deactivate */
Jens Axboedece1632015-11-05 10:41:16 -07002075
2076out:
2077 return ret;
Neil Brownd89d8792007-05-01 09:53:42 +02002078}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079EXPORT_SYMBOL(generic_make_request);
2080
2081/**
Randy Dunlap710027a2008-08-19 20:13:11 +02002082 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 * @bio: The &struct bio which describes the I/O
2084 *
2085 * submit_bio() is very similar in purpose to generic_make_request(), and
2086 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02002087 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 *
2089 */
Mike Christie4e49ea42016-06-05 14:31:41 -05002090blk_qc_t submit_bio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091{
Jens Axboebf2de6f2007-09-27 13:01:25 +02002092 /*
2093 * If it's a regular read/write or a barrier with data attached,
2094 * go through the normal accounting stuff before submission.
2095 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002096 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002097 unsigned int count;
2098
Mike Christie95fe6c12016-06-05 14:31:48 -05002099 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
Martin K. Petersen4363ac72012-09-18 12:19:27 -04002100 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2101 else
2102 count = bio_sectors(bio);
2103
Mike Christiea8ebb052016-06-05 14:31:45 -05002104 if (op_is_write(bio_op(bio))) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02002105 count_vm_events(PGPGOUT, count);
2106 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -07002107 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002108 count_vm_events(PGPGIN, count);
2109 }
2110
2111 if (unlikely(block_dump)) {
2112 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02002113 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07002114 current->comm, task_pid_nr(current),
Mike Christiea8ebb052016-06-05 14:31:45 -05002115 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07002116 (unsigned long long)bio->bi_iter.bi_sector,
San Mehat8dcbdc72010-09-14 08:48:01 +02002117 bdevname(bio->bi_bdev, b),
2118 count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02002119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 }
2121
Jens Axboedece1632015-11-05 10:41:16 -07002122 return generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124EXPORT_SYMBOL(submit_bio);
2125
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002126/**
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002127 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2128 * for new the queue limits
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002129 * @q: the queue
2130 * @rq: the request being checked
2131 *
2132 * Description:
2133 * @rq may have been made based on weaker limitations of upper-level queues
2134 * in request stacking drivers, and it may violate the limitation of @q.
2135 * Since the block layer and the underlying device driver trust @rq
2136 * after it is inserted to @q, it should be checked against @q before
2137 * the insertion using this generic function.
2138 *
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002139 * Request stacking drivers like request-based dm may change the queue
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002140 * limits when retrying requests on other queues. Those requests need
2141 * to be checked against the new queue limits again during dispatch.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002142 */
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002143static int blk_cloned_rq_check_limits(struct request_queue *q,
2144 struct request *rq)
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002145{
Mike Christie8fe0d472016-06-05 14:32:15 -05002146 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002147 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2148 return -EIO;
2149 }
2150
2151 /*
2152 * queue's settings related to segment counting like q->bounce_pfn
2153 * may differ from that of other stacking queues.
2154 * Recalculate it to check the request correctly on this queue's
2155 * limitation.
2156 */
2157 blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05002158 if (rq->nr_phys_segments > queue_max_segments(q)) {
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002159 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2160 return -EIO;
2161 }
2162
2163 return 0;
2164}
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002165
2166/**
2167 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2168 * @q: the queue to submit the request
2169 * @rq: the request being queued
2170 */
2171int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2172{
2173 unsigned long flags;
Jeff Moyer4853aba2011-08-15 21:37:25 +02002174 int where = ELEVATOR_INSERT_BACK;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002175
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01002176 if (blk_cloned_rq_check_limits(q, rq))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002177 return -EIO;
2178
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07002179 if (rq->rq_disk &&
2180 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002181 return -EIO;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002182
Keith Busch7fb48982014-10-17 17:46:38 -06002183 if (q->mq_ops) {
2184 if (blk_queue_io_stat(q))
2185 blk_account_io_start(rq, true);
Mike Snitzer6acfe682016-02-05 08:49:01 -05002186 blk_mq_insert_request(rq, false, true, false);
Keith Busch7fb48982014-10-17 17:46:38 -06002187 return 0;
2188 }
2189
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002190 spin_lock_irqsave(q->queue_lock, flags);
Bart Van Assche3f3299d2012-11-28 13:42:38 +01002191 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01002192 spin_unlock_irqrestore(q->queue_lock, flags);
2193 return -ENODEV;
2194 }
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002195
2196 /*
2197 * Submitting request must be dequeued before calling this function
2198 * because it will be linked to another request_queue
2199 */
2200 BUG_ON(blk_queued_rq(rq));
2201
Mike Christie28a8f0d2016-06-05 14:32:25 -05002202 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
Jeff Moyer4853aba2011-08-15 21:37:25 +02002203 where = ELEVATOR_INSERT_FLUSH;
2204
2205 add_acct_request(q, rq, where);
Jeff Moyere67b77c2011-10-17 12:57:23 +02002206 if (where == ELEVATOR_INSERT_FLUSH)
2207 __blk_run_queue(q);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04002208 spin_unlock_irqrestore(q->queue_lock, flags);
2209
2210 return 0;
2211}
2212EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2213
Tejun Heo80a761f2009-07-03 17:48:17 +09002214/**
2215 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2216 * @rq: request to examine
2217 *
2218 * Description:
2219 * A request could be merge of IOs which require different failure
2220 * handling. This function determines the number of bytes which
2221 * can be failed from the beginning of the request without
2222 * crossing into area which need to be retried further.
2223 *
2224 * Return:
2225 * The number of bytes to fail.
2226 *
2227 * Context:
2228 * queue_lock must be held.
2229 */
2230unsigned int blk_rq_err_bytes(const struct request *rq)
2231{
2232 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2233 unsigned int bytes = 0;
2234 struct bio *bio;
2235
2236 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2237 return blk_rq_bytes(rq);
2238
2239 /*
2240 * Currently the only 'mixing' which can happen is between
2241 * different fastfail types. We can safely fail portions
2242 * which have all the failfast bits that the first one has -
2243 * the ones which are at least as eager to fail as the first
2244 * one.
2245 */
2246 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -06002247 if ((bio->bi_opf & ff) != ff)
Tejun Heo80a761f2009-07-03 17:48:17 +09002248 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002249 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09002250 }
2251
2252 /* this could lead to infinite loop */
2253 BUG_ON(blk_rq_bytes(rq) && !bytes);
2254 return bytes;
2255}
2256EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2257
Jens Axboe320ae512013-10-24 09:20:05 +01002258void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01002259{
Jens Axboec2553b52009-04-24 08:10:11 +02002260 if (blk_do_io_stat(req)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002261 const int rw = rq_data_dir(req);
2262 struct hd_struct *part;
2263 int cpu;
2264
2265 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002266 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002267 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2268 part_stat_unlock();
2269 }
2270}
2271
Jens Axboe320ae512013-10-24 09:20:05 +01002272void blk_account_io_done(struct request *req)
Jens Axboebc58ba92009-01-23 10:54:44 +01002273{
Jens Axboebc58ba92009-01-23 10:54:44 +01002274 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02002275 * Account IO completion. flush_rq isn't accounted as a
2276 * normal IO on queueing nor completion. Accounting the
2277 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01002278 */
Tejun Heo414b4ff2011-01-25 12:43:49 +01002279 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
Jens Axboebc58ba92009-01-23 10:54:44 +01002280 unsigned long duration = jiffies - req->start_time;
2281 const int rw = rq_data_dir(req);
2282 struct hd_struct *part;
2283 int cpu;
2284
2285 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01002286 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01002287
2288 part_stat_inc(cpu, part, ios[rw]);
2289 part_stat_add(cpu, part, ticks[rw], duration);
2290 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +02002291 part_dec_in_flight(part, rw);
Jens Axboebc58ba92009-01-23 10:54:44 +01002292
Jens Axboe6c23a962011-01-07 08:43:37 +01002293 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01002294 part_stat_unlock();
2295 }
2296}
2297
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01002298#ifdef CONFIG_PM
Lin Mingc8158812013-03-23 11:42:27 +08002299/*
2300 * Don't process normal requests when queue is suspended
2301 * or in the process of suspending/resuming
2302 */
2303static struct request *blk_pm_peek_request(struct request_queue *q,
2304 struct request *rq)
2305{
2306 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2307 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2308 return NULL;
2309 else
2310 return rq;
2311}
2312#else
2313static inline struct request *blk_pm_peek_request(struct request_queue *q,
2314 struct request *rq)
2315{
2316 return rq;
2317}
2318#endif
2319
Jens Axboe320ae512013-10-24 09:20:05 +01002320void blk_account_io_start(struct request *rq, bool new_io)
2321{
2322 struct hd_struct *part;
2323 int rw = rq_data_dir(rq);
2324 int cpu;
2325
2326 if (!blk_do_io_stat(rq))
2327 return;
2328
2329 cpu = part_stat_lock();
2330
2331 if (!new_io) {
2332 part = rq->part;
2333 part_stat_inc(cpu, part, merges[rw]);
2334 } else {
2335 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2336 if (!hd_struct_try_get(part)) {
2337 /*
2338 * The partition is already being removed,
2339 * the request will be accounted on the disk only
2340 *
2341 * We take a reference on disk->part0 although that
2342 * partition will never be deleted, so we can treat
2343 * it as any other partition.
2344 */
2345 part = &rq->rq_disk->part0;
2346 hd_struct_get(part);
2347 }
2348 part_round_stats(cpu, part);
2349 part_inc_in_flight(part, rw);
2350 rq->part = part;
2351 }
2352
2353 part_stat_unlock();
2354}
2355
Tejun Heo53a08802008-12-03 12:41:26 +01002356/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002357 * blk_peek_request - peek at the top of a request queue
2358 * @q: request queue to peek at
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002359 *
2360 * Description:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002361 * Return the request at the top of @q. The returned request
2362 * should be started using blk_start_request() before LLD starts
2363 * processing it.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002364 *
2365 * Return:
Tejun Heo9934c8c2009-05-08 11:54:16 +09002366 * Pointer to the request at the top of @q if available. Null
2367 * otherwise.
2368 *
2369 * Context:
2370 * queue_lock must be held.
2371 */
2372struct request *blk_peek_request(struct request_queue *q)
Tejun Heo158dbda2009-04-23 11:05:18 +09002373{
2374 struct request *rq;
2375 int ret;
2376
2377 while ((rq = __elv_next_request(q)) != NULL) {
Lin Mingc8158812013-03-23 11:42:27 +08002378
2379 rq = blk_pm_peek_request(q, rq);
2380 if (!rq)
2381 break;
2382
Tejun Heo158dbda2009-04-23 11:05:18 +09002383 if (!(rq->cmd_flags & REQ_STARTED)) {
2384 /*
2385 * This is the first time the device driver
2386 * sees this request (possibly after
2387 * requeueing). Notify IO scheduler.
2388 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002389 if (rq->cmd_flags & REQ_SORTED)
Tejun Heo158dbda2009-04-23 11:05:18 +09002390 elv_activate_rq(q, rq);
2391
2392 /*
2393 * just mark as started even if we don't start
2394 * it, a request that has been delayed should
2395 * not be passed by new incoming requests
2396 */
2397 rq->cmd_flags |= REQ_STARTED;
2398 trace_block_rq_issue(q, rq);
2399 }
2400
2401 if (!q->boundary_rq || q->boundary_rq == rq) {
2402 q->end_sector = rq_end_sector(rq);
2403 q->boundary_rq = NULL;
2404 }
2405
2406 if (rq->cmd_flags & REQ_DONTPREP)
2407 break;
2408
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002409 if (q->dma_drain_size && blk_rq_bytes(rq)) {
Tejun Heo158dbda2009-04-23 11:05:18 +09002410 /*
2411 * make sure space for the drain appears we
2412 * know we can do this because max_hw_segments
2413 * has been adjusted to be one fewer than the
2414 * device can handle
2415 */
2416 rq->nr_phys_segments++;
2417 }
2418
2419 if (!q->prep_rq_fn)
2420 break;
2421
2422 ret = q->prep_rq_fn(q, rq);
2423 if (ret == BLKPREP_OK) {
2424 break;
2425 } else if (ret == BLKPREP_DEFER) {
2426 /*
2427 * the request may have been (partially) prepped.
2428 * we need to keep this request in the front to
2429 * avoid resource deadlock. REQ_STARTED will
2430 * prevent other fs requests from passing this one.
2431 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002432 if (q->dma_drain_size && blk_rq_bytes(rq) &&
Tejun Heo158dbda2009-04-23 11:05:18 +09002433 !(rq->cmd_flags & REQ_DONTPREP)) {
2434 /*
2435 * remove the space for the drain we added
2436 * so that we don't add it again
2437 */
2438 --rq->nr_phys_segments;
2439 }
2440
2441 rq = NULL;
2442 break;
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002443 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2444 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2445
Tejun Heo158dbda2009-04-23 11:05:18 +09002446 rq->cmd_flags |= REQ_QUIET;
James Bottomleyc143dc92009-05-30 06:43:49 +02002447 /*
2448 * Mark this request as started so we don't trigger
2449 * any debug logic in the end I/O path.
2450 */
2451 blk_start_request(rq);
Martin K. Petersen0fb5b1f2016-02-04 00:52:12 -05002452 __blk_end_request_all(rq, err);
Tejun Heo158dbda2009-04-23 11:05:18 +09002453 } else {
2454 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2455 break;
2456 }
2457 }
2458
2459 return rq;
2460}
Tejun Heo9934c8c2009-05-08 11:54:16 +09002461EXPORT_SYMBOL(blk_peek_request);
Tejun Heo158dbda2009-04-23 11:05:18 +09002462
Tejun Heo9934c8c2009-05-08 11:54:16 +09002463void blk_dequeue_request(struct request *rq)
Tejun Heo158dbda2009-04-23 11:05:18 +09002464{
Tejun Heo9934c8c2009-05-08 11:54:16 +09002465 struct request_queue *q = rq->q;
2466
Tejun Heo158dbda2009-04-23 11:05:18 +09002467 BUG_ON(list_empty(&rq->queuelist));
2468 BUG_ON(ELV_ON_HASH(rq));
2469
2470 list_del_init(&rq->queuelist);
2471
2472 /*
2473 * the time frame between a request being removed from the lists
2474 * and to it is freed is accounted as io that is in progress at
2475 * the driver side.
2476 */
Divyesh Shah91952912010-04-01 15:01:41 -07002477 if (blk_account_rq(rq)) {
Jens Axboe0a7ae2f2009-05-20 08:54:31 +02002478 q->in_flight[rq_is_sync(rq)]++;
Divyesh Shah91952912010-04-01 15:01:41 -07002479 set_io_start_time_ns(rq);
2480 }
Tejun Heo158dbda2009-04-23 11:05:18 +09002481}
2482
Tejun Heo5efccd12009-04-23 11:05:18 +09002483/**
Tejun Heo9934c8c2009-05-08 11:54:16 +09002484 * blk_start_request - start request processing on the driver
2485 * @req: request to dequeue
2486 *
2487 * Description:
2488 * Dequeue @req and start timeout timer on it. This hands off the
2489 * request to the driver.
2490 *
2491 * Block internal functions which don't want to start timer should
2492 * call blk_dequeue_request().
2493 *
2494 * Context:
2495 * queue_lock must be held.
2496 */
2497void blk_start_request(struct request *req)
2498{
2499 blk_dequeue_request(req);
2500
2501 /*
Tejun Heo5f49f632009-05-19 18:33:05 +09002502 * We are now handing the request to the hardware, initialize
2503 * resid_len to full count and add the timeout handler.
Tejun Heo9934c8c2009-05-08 11:54:16 +09002504 */
Tejun Heo5f49f632009-05-19 18:33:05 +09002505 req->resid_len = blk_rq_bytes(req);
FUJITA Tomonoridbb66c42009-06-09 05:47:10 +02002506 if (unlikely(blk_bidi_rq(req)))
2507 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2508
Jeff Moyer4912aa62013-10-08 14:36:41 -04002509 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
Tejun Heo9934c8c2009-05-08 11:54:16 +09002510 blk_add_timer(req);
2511}
2512EXPORT_SYMBOL(blk_start_request);
2513
2514/**
2515 * blk_fetch_request - fetch a request from a request queue
2516 * @q: request queue to fetch a request from
2517 *
2518 * Description:
2519 * Return the request at the top of @q. The request is started on
2520 * return and LLD can start processing it immediately.
2521 *
2522 * Return:
2523 * Pointer to the request at the top of @q if available. Null
2524 * otherwise.
2525 *
2526 * Context:
2527 * queue_lock must be held.
2528 */
2529struct request *blk_fetch_request(struct request_queue *q)
2530{
2531 struct request *rq;
2532
2533 rq = blk_peek_request(q);
2534 if (rq)
2535 blk_start_request(rq);
2536 return rq;
2537}
2538EXPORT_SYMBOL(blk_fetch_request);
2539
2540/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002541 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002542 * @req: the request being processed
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002543 * @error: %0 for success, < %0 for error
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002544 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002545 *
2546 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002547 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2548 * the request structure even if @req doesn't have leftover.
2549 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09002550 *
2551 * This special helper function is only for request stacking drivers
2552 * (e.g. request-based dm) so that they can handle partial completion.
2553 * Actual device drivers should use blk_end_request instead.
2554 *
2555 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2556 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002557 *
2558 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002559 * %false - this request doesn't have any more data
2560 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05002561 **/
Tejun Heo2e60e022009-04-23 11:05:18 +09002562bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563{
Kent Overstreetf79ea412012-09-20 16:38:30 -07002564 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02002566 trace_block_rq_complete(req->q, req, nr_bytes);
2567
Tejun Heo2e60e022009-04-23 11:05:18 +09002568 if (!req->bio)
2569 return false;
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 /*
Tejun Heo6f414692009-04-19 07:00:41 +09002572 * For fs requests, rq is just carrier of independent bio's
2573 * and each partial completion should be handled separately.
2574 * Reset per-request error on each partial completion.
2575 *
2576 * TODO: tj: This is too subtle. It would be better to let
2577 * low level drivers do what they see fit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 */
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002579 if (req->cmd_type == REQ_TYPE_FS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 req->errors = 0;
2581
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002582 if (error && req->cmd_type == REQ_TYPE_FS &&
2583 !(req->cmd_flags & REQ_QUIET)) {
Hannes Reinecke79775562011-01-18 10:13:13 +01002584 char *error_type;
2585
2586 switch (error) {
2587 case -ENOLINK:
2588 error_type = "recoverable transport";
2589 break;
2590 case -EREMOTEIO:
2591 error_type = "critical target";
2592 break;
2593 case -EBADE:
2594 error_type = "critical nexus";
2595 break;
Hannes Reinecked1ffc1f2013-01-30 09:26:16 +00002596 case -ETIMEDOUT:
2597 error_type = "timeout";
2598 break;
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +02002599 case -ENOSPC:
2600 error_type = "critical space allocation";
2601 break;
Hannes Reinecke7e782af2013-07-01 15:16:26 +02002602 case -ENODATA:
2603 error_type = "critical medium";
2604 break;
Hannes Reinecke79775562011-01-18 10:13:13 +01002605 case -EIO:
2606 default:
2607 error_type = "I/O";
2608 break;
2609 }
Robert Elliottef3ecb62014-08-27 10:50:31 -05002610 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2611 __func__, error_type, req->rq_disk ?
Yi Zou37d7b342012-08-30 16:26:25 -07002612 req->rq_disk->disk_name : "?",
2613 (unsigned long long)blk_rq_pos(req));
2614
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 }
2616
Jens Axboebc58ba92009-01-23 10:54:44 +01002617 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01002618
Kent Overstreetf79ea412012-09-20 16:38:30 -07002619 total_bytes = 0;
2620 while (req->bio) {
2621 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002622 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Kent Overstreet4f024f32013-10-11 15:44:27 -07002624 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Kent Overstreetf79ea412012-09-20 16:38:30 -07002627 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Kent Overstreetf79ea412012-09-20 16:38:30 -07002629 total_bytes += bio_bytes;
2630 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
Kent Overstreetf79ea412012-09-20 16:38:30 -07002632 if (!nr_bytes)
2633 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 }
2635
2636 /*
2637 * completely done
2638 */
Tejun Heo2e60e022009-04-23 11:05:18 +09002639 if (!req->bio) {
2640 /*
2641 * Reset counters so that the request stacking driver
2642 * can find how many bytes remain in the request
2643 * later.
2644 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002645 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09002646 return false;
2647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002649 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002650
2651 /* update sector only for requests with clear definition of sector */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04002652 if (req->cmd_type == REQ_TYPE_FS)
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002653 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002654
Tejun Heo80a761f2009-07-03 17:48:17 +09002655 /* mixed attributes always follow the first bio */
2656 if (req->cmd_flags & REQ_MIXED_MERGE) {
2657 req->cmd_flags &= ~REQ_FAILFAST_MASK;
Jens Axboe1eff9d32016-08-05 15:35:16 -06002658 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
Tejun Heo80a761f2009-07-03 17:48:17 +09002659 }
2660
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002661 /*
2662 * If total number of sectors is less than the first segment
2663 * size, something has gone terribly wrong.
2664 */
2665 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
Jens Axboe81829242011-03-30 09:51:33 +02002666 blk_dump_rq_flags(req, "request botched");
Tejun Heoa2dec7b2009-05-07 22:24:44 +09002667 req->__data_len = blk_rq_cur_bytes(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002668 }
2669
2670 /* recalculate the number of segments */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 blk_recalc_rq_segments(req);
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002672
Tejun Heo2e60e022009-04-23 11:05:18 +09002673 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674}
Tejun Heo2e60e022009-04-23 11:05:18 +09002675EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
Tejun Heo2e60e022009-04-23 11:05:18 +09002677static bool blk_update_bidi_request(struct request *rq, int error,
2678 unsigned int nr_bytes,
2679 unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002680{
Tejun Heo2e60e022009-04-23 11:05:18 +09002681 if (blk_update_request(rq, error, nr_bytes))
2682 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002683
Tejun Heo2e60e022009-04-23 11:05:18 +09002684 /* Bidi request must be completed as a whole */
2685 if (unlikely(blk_bidi_rq(rq)) &&
2686 blk_update_request(rq->next_rq, error, bidi_bytes))
2687 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002688
Jens Axboee2e1a142010-06-09 10:42:09 +02002689 if (blk_queue_add_random(rq->q))
2690 add_disk_randomness(rq->rq_disk);
Tejun Heo2e60e022009-04-23 11:05:18 +09002691
2692 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693}
2694
James Bottomley28018c22010-07-01 19:49:17 +09002695/**
2696 * blk_unprep_request - unprepare a request
2697 * @req: the request
2698 *
2699 * This function makes a request ready for complete resubmission (or
2700 * completion). It happens only after all error handling is complete,
2701 * so represents the appropriate moment to deallocate any resources
2702 * that were allocated to the request in the prep_rq_fn. The queue
2703 * lock is held when calling this.
2704 */
2705void blk_unprep_request(struct request *req)
2706{
2707 struct request_queue *q = req->q;
2708
2709 req->cmd_flags &= ~REQ_DONTPREP;
2710 if (q->unprep_rq_fn)
2711 q->unprep_rq_fn(q, req);
2712}
2713EXPORT_SYMBOL_GPL(blk_unprep_request);
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715/*
2716 * queue lock must be held
2717 */
Christoph Hellwig12120072014-04-16 09:44:59 +02002718void blk_finish_request(struct request *req, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719{
Christoph Hellwig125c99b2014-11-03 12:47:47 +01002720 if (req->cmd_flags & REQ_QUEUED)
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002721 blk_queue_end_tag(req->q, req);
2722
James Bottomleyba396a62009-05-27 14:17:08 +02002723 BUG_ON(blk_queued_rq(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
Christoph Hellwig33659eb2010-08-07 18:17:56 +02002725 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
Matthew Garrett31373d02010-04-06 14:25:14 +02002726 laptop_io_completion(&req->q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
Mike Andersone78042e2008-10-30 02:16:20 -07002728 blk_delete_timer(req);
2729
James Bottomley28018c22010-07-01 19:49:17 +09002730 if (req->cmd_flags & REQ_DONTPREP)
2731 blk_unprep_request(req);
2732
Jens Axboebc58ba92009-01-23 10:54:44 +01002733 blk_account_io_done(req);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002734
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01002736 req->end_io(req, error);
Kiyoshi Uedab8286232007-12-11 17:53:24 -05002737 else {
2738 if (blk_bidi_rq(req))
2739 __blk_put_request(req->next_rq->q, req->next_rq);
2740
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 __blk_put_request(req->q, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
2743}
Christoph Hellwig12120072014-04-16 09:44:59 +02002744EXPORT_SYMBOL(blk_finish_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745
Kiyoshi Ueda3b113132007-12-11 17:41:17 -05002746/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002747 * blk_end_bidi_request - Complete a bidi request
2748 * @rq: the request to complete
Randy Dunlap710027a2008-08-19 20:13:11 +02002749 * @error: %0 for success, < %0 for error
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002750 * @nr_bytes: number of bytes to complete @rq
2751 * @bidi_bytes: number of bytes to complete @rq->next_rq
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002752 *
2753 * Description:
2754 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
Tejun Heo2e60e022009-04-23 11:05:18 +09002755 * Drivers that supports bidi can safely call this member for any
2756 * type of request, bidi or uni. In the later case @bidi_bytes is
2757 * just ignored.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002758 *
2759 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09002760 * %false - we are done with this request
2761 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002762 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002763static bool blk_end_bidi_request(struct request *rq, int error,
2764 unsigned int nr_bytes, unsigned int bidi_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002765{
2766 struct request_queue *q = rq->q;
Tejun Heo2e60e022009-04-23 11:05:18 +09002767 unsigned long flags;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002768
Tejun Heo2e60e022009-04-23 11:05:18 +09002769 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2770 return true;
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002771
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002772 spin_lock_irqsave(q->queue_lock, flags);
Tejun Heo2e60e022009-04-23 11:05:18 +09002773 blk_finish_request(rq, error);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002774 spin_unlock_irqrestore(q->queue_lock, flags);
2775
Tejun Heo2e60e022009-04-23 11:05:18 +09002776 return false;
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002777}
Kiyoshi Uedae3a04fe2007-12-11 17:51:46 -05002778
2779/**
Tejun Heo2e60e022009-04-23 11:05:18 +09002780 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2781 * @rq: the request to complete
2782 * @error: %0 for success, < %0 for error
2783 * @nr_bytes: number of bytes to complete @rq
2784 * @bidi_bytes: number of bytes to complete @rq->next_rq
Tejun Heo5efccd12009-04-23 11:05:18 +09002785 *
2786 * Description:
Tejun Heo2e60e022009-04-23 11:05:18 +09002787 * Identical to blk_end_bidi_request() except that queue lock is
2788 * assumed to be locked on entry and remains so on return.
Tejun Heo5efccd12009-04-23 11:05:18 +09002789 *
Tejun Heo2e60e022009-04-23 11:05:18 +09002790 * Return:
2791 * %false - we are done with this request
2792 * %true - still buffers pending for this request
Tejun Heo5efccd12009-04-23 11:05:18 +09002793 **/
Jeff Moyer4853aba2011-08-15 21:37:25 +02002794bool __blk_end_bidi_request(struct request *rq, int error,
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002795 unsigned int nr_bytes, unsigned int bidi_bytes)
Tejun Heo5efccd12009-04-23 11:05:18 +09002796{
Tejun Heo2e60e022009-04-23 11:05:18 +09002797 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2798 return true;
Tejun Heo5efccd12009-04-23 11:05:18 +09002799
Tejun Heo2e60e022009-04-23 11:05:18 +09002800 blk_finish_request(rq, error);
Tejun Heo5efccd12009-04-23 11:05:18 +09002801
Tejun Heo2e60e022009-04-23 11:05:18 +09002802 return false;
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002803}
2804
2805/**
2806 * blk_end_request - Helper function for drivers to complete the request.
2807 * @rq: the request being processed
2808 * @error: %0 for success, < %0 for error
2809 * @nr_bytes: number of bytes to complete
2810 *
2811 * Description:
2812 * Ends I/O on a number of bytes attached to @rq.
2813 * If @rq has leftover, sets it up for the next range of segments.
2814 *
2815 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002816 * %false - we are done with this request
2817 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002818 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002819bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002820{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002821 return blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002822}
Jens Axboe56ad1742009-07-28 22:11:24 +02002823EXPORT_SYMBOL(blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002824
2825/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002826 * blk_end_request_all - Helper function for drives to finish the request.
2827 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002828 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002829 *
2830 * Description:
2831 * Completely finish @rq.
2832 */
2833void blk_end_request_all(struct request *rq, int error)
2834{
2835 bool pending;
2836 unsigned int bidi_bytes = 0;
2837
2838 if (unlikely(blk_bidi_rq(rq)))
2839 bidi_bytes = blk_rq_bytes(rq->next_rq);
2840
2841 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2842 BUG_ON(pending);
2843}
Jens Axboe56ad1742009-07-28 22:11:24 +02002844EXPORT_SYMBOL(blk_end_request_all);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002845
2846/**
2847 * blk_end_request_cur - Helper function to finish the current request chunk.
2848 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002849 * @error: %0 for success, < %0 for error
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002850 *
2851 * Description:
2852 * Complete the current consecutively mapped chunk from @rq.
2853 *
2854 * Return:
2855 * %false - we are done with this request
2856 * %true - still buffers pending for this request
2857 */
2858bool blk_end_request_cur(struct request *rq, int error)
2859{
2860 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2861}
Jens Axboe56ad1742009-07-28 22:11:24 +02002862EXPORT_SYMBOL(blk_end_request_cur);
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002863
2864/**
Tejun Heo80a761f2009-07-03 17:48:17 +09002865 * blk_end_request_err - Finish a request till the next failure boundary.
2866 * @rq: the request to finish till the next failure boundary for
2867 * @error: must be negative errno
2868 *
2869 * Description:
2870 * Complete @rq till the next failure boundary.
2871 *
2872 * Return:
2873 * %false - we are done with this request
2874 * %true - still buffers pending for this request
2875 */
2876bool blk_end_request_err(struct request *rq, int error)
2877{
2878 WARN_ON(error >= 0);
2879 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2880}
2881EXPORT_SYMBOL_GPL(blk_end_request_err);
2882
2883/**
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002884 * __blk_end_request - Helper function for drivers to complete the request.
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002885 * @rq: the request being processed
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002886 * @error: %0 for success, < %0 for error
2887 * @nr_bytes: number of bytes to complete
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002888 *
2889 * Description:
2890 * Must be called with queue lock held unlike blk_end_request().
2891 *
2892 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002893 * %false - we are done with this request
2894 * %true - still buffers pending for this request
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002895 **/
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002896bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002897{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002898 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002899}
Jens Axboe56ad1742009-07-28 22:11:24 +02002900EXPORT_SYMBOL(__blk_end_request);
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002901
2902/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002903 * __blk_end_request_all - Helper function for drives to finish the request.
2904 * @rq: the request to finish
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002905 * @error: %0 for success, < %0 for error
Kiyoshi Ueda336cdb42007-12-11 17:40:30 -05002906 *
2907 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002908 * Completely finish @rq. Must be called with queue lock held.
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002909 */
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002910void __blk_end_request_all(struct request *rq, int error)
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002911{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002912 bool pending;
2913 unsigned int bidi_bytes = 0;
2914
2915 if (unlikely(blk_bidi_rq(rq)))
2916 bidi_bytes = blk_rq_bytes(rq->next_rq);
2917
2918 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2919 BUG_ON(pending);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002920}
Jens Axboe56ad1742009-07-28 22:11:24 +02002921EXPORT_SYMBOL(__blk_end_request_all);
Kiyoshi Ueda32fab442008-09-18 10:45:09 -04002922
2923/**
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002924 * __blk_end_request_cur - Helper function to finish the current request chunk.
2925 * @rq: the request to finish the current chunk for
Randy Dunlap8ebf9752009-06-11 20:00:41 -07002926 * @error: %0 for success, < %0 for error
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002927 *
2928 * Description:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002929 * Complete the current consecutively mapped chunk from @rq. Must
2930 * be called with queue lock held.
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002931 *
2932 * Return:
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002933 * %false - we are done with this request
2934 * %true - still buffers pending for this request
2935 */
2936bool __blk_end_request_cur(struct request *rq, int error)
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002937{
FUJITA Tomonorib1f74492009-05-11 17:56:09 +09002938 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002939}
Jens Axboe56ad1742009-07-28 22:11:24 +02002940EXPORT_SYMBOL(__blk_end_request_cur);
Kiyoshi Uedae19a3ab2007-12-11 17:51:02 -05002941
Tejun Heo80a761f2009-07-03 17:48:17 +09002942/**
2943 * __blk_end_request_err - Finish a request till the next failure boundary.
2944 * @rq: the request to finish till the next failure boundary for
2945 * @error: must be negative errno
2946 *
2947 * Description:
2948 * Complete @rq till the next failure boundary. Must be called
2949 * with queue lock held.
2950 *
2951 * Return:
2952 * %false - we are done with this request
2953 * %true - still buffers pending for this request
2954 */
2955bool __blk_end_request_err(struct request *rq, int error)
2956{
2957 WARN_ON(error >= 0);
2958 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2959}
2960EXPORT_SYMBOL_GPL(__blk_end_request_err);
2961
Jens Axboe86db1e22008-01-29 14:53:40 +01002962void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2963 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964{
Mike Christie4993b772016-06-05 14:32:10 -05002965 req_set_op(rq, bio_op(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966
Jens Axboeb4f42e22014-04-10 09:46:28 -06002967 if (bio_has_data(bio))
David Woodhousefb2dce82008-08-05 18:01:53 +01002968 rq->nr_phys_segments = bio_phys_segments(q, bio);
Jens Axboeb4f42e22014-04-10 09:46:28 -06002969
Kent Overstreet4f024f32013-10-11 15:44:27 -07002970 rq->__data_len = bio->bi_iter.bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972
NeilBrown66846572007-08-16 13:31:28 +02002973 if (bio->bi_bdev)
2974 rq->rq_disk = bio->bi_bdev->bd_disk;
2975}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002977#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2978/**
2979 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2980 * @rq: the request to be flushed
2981 *
2982 * Description:
2983 * Flush all pages in @rq.
2984 */
2985void rq_flush_dcache_pages(struct request *rq)
2986{
2987 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08002988 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002989
2990 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08002991 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01002992}
2993EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2994#endif
2995
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02002996/**
2997 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2998 * @q : the queue of the device being checked
2999 *
3000 * Description:
3001 * Check if underlying low-level drivers of a device are busy.
3002 * If the drivers want to export their busy state, they must set own
3003 * exporting function using blk_queue_lld_busy() first.
3004 *
3005 * Basically, this function is used only by request stacking drivers
3006 * to stop dispatching requests to underlying devices when underlying
3007 * devices are busy. This behavior helps more I/O merging on the queue
3008 * of the request stacking driver and prevents I/O throughput regression
3009 * on burst I/O load.
3010 *
3011 * Return:
3012 * 0 - Not busy (The request stacking driver should dispatch request)
3013 * 1 - Busy (The request stacking driver should stop dispatching request)
3014 */
3015int blk_lld_busy(struct request_queue *q)
3016{
3017 if (q->lld_busy_fn)
3018 return q->lld_busy_fn(q);
3019
3020 return 0;
3021}
3022EXPORT_SYMBOL_GPL(blk_lld_busy);
3023
Mike Snitzer78d8e582015-06-26 10:01:13 -04003024/**
3025 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3026 * @rq: the clone request to be cleaned up
3027 *
3028 * Description:
3029 * Free all bios in @rq for a cloned request.
3030 */
3031void blk_rq_unprep_clone(struct request *rq)
3032{
3033 struct bio *bio;
3034
3035 while ((bio = rq->bio) != NULL) {
3036 rq->bio = bio->bi_next;
3037
3038 bio_put(bio);
3039 }
3040}
3041EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3042
3043/*
3044 * Copy attributes of the original request to the clone request.
3045 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3046 */
3047static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003048{
3049 dst->cpu = src->cpu;
Mike Christie4993b772016-06-05 14:32:10 -05003050 req_set_op_attrs(dst, req_op(src),
3051 (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003052 dst->cmd_type = src->cmd_type;
3053 dst->__sector = blk_rq_pos(src);
3054 dst->__data_len = blk_rq_bytes(src);
3055 dst->nr_phys_segments = src->nr_phys_segments;
3056 dst->ioprio = src->ioprio;
3057 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04003058}
3059
3060/**
3061 * blk_rq_prep_clone - Helper function to setup clone request
3062 * @rq: the request to be setup
3063 * @rq_src: original request to be cloned
3064 * @bs: bio_set that bios for clone are allocated from
3065 * @gfp_mask: memory allocation mask for bio
3066 * @bio_ctr: setup function to be called for each clone bio.
3067 * Returns %0 for success, non %0 for failure.
3068 * @data: private data to be passed to @bio_ctr
3069 *
3070 * Description:
3071 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3072 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3073 * are not copied, and copying such parts is the caller's responsibility.
3074 * Also, pages which the original bios are pointing to are not copied
3075 * and the cloned bios just point same pages.
3076 * So cloned bios must be completed before original bios, which means
3077 * the caller must complete @rq before @rq_src.
3078 */
3079int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3080 struct bio_set *bs, gfp_t gfp_mask,
3081 int (*bio_ctr)(struct bio *, struct bio *, void *),
3082 void *data)
3083{
3084 struct bio *bio, *bio_src;
3085
3086 if (!bs)
3087 bs = fs_bio_set;
3088
3089 __rq_for_each_bio(bio_src, rq_src) {
3090 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3091 if (!bio)
3092 goto free_and_out;
3093
3094 if (bio_ctr && bio_ctr(bio, bio_src, data))
3095 goto free_and_out;
3096
3097 if (rq->bio) {
3098 rq->biotail->bi_next = bio;
3099 rq->biotail = bio;
3100 } else
3101 rq->bio = rq->biotail = bio;
3102 }
3103
3104 __blk_rq_prep_clone(rq, rq_src);
3105
3106 return 0;
3107
3108free_and_out:
3109 if (bio)
3110 bio_put(bio);
3111 blk_rq_unprep_clone(rq);
3112
3113 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02003114}
3115EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3116
Jens Axboe59c3d452014-04-08 09:15:35 -06003117int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118{
3119 return queue_work(kblockd_workqueue, work);
3120}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121EXPORT_SYMBOL(kblockd_schedule_work);
3122
Jens Axboeee63cfa2016-08-24 15:52:48 -06003123int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3124{
3125 return queue_work_on(cpu, kblockd_workqueue, work);
3126}
3127EXPORT_SYMBOL(kblockd_schedule_work_on);
3128
Jens Axboe59c3d452014-04-08 09:15:35 -06003129int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3130 unsigned long delay)
Vivek Goyale43473b2010-09-15 17:06:35 -04003131{
3132 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3133}
3134EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3135
Jens Axboe8ab14592014-04-08 09:17:40 -06003136int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3137 unsigned long delay)
3138{
3139 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3140}
3141EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3142
Suresh Jayaraman75df7132011-09-21 10:00:16 +02003143/**
3144 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3145 * @plug: The &struct blk_plug that needs to be initialized
3146 *
3147 * Description:
3148 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3149 * pending I/O should the task end up blocking between blk_start_plug() and
3150 * blk_finish_plug(). This is important from a performance perspective, but
3151 * also ensures that we don't deadlock. For instance, if the task is blocking
3152 * for a memory allocation, memory reclaim could end up wanting to free a
3153 * page belonging to that request that is currently residing in our private
3154 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3155 * this kind of deadlock.
3156 */
Jens Axboe73c10102011-03-08 13:19:51 +01003157void blk_start_plug(struct blk_plug *plug)
3158{
3159 struct task_struct *tsk = current;
3160
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003161 /*
3162 * If this is a nested plug, don't actually assign it.
3163 */
3164 if (tsk->plug)
3165 return;
3166
Jens Axboe73c10102011-03-08 13:19:51 +01003167 INIT_LIST_HEAD(&plug->list);
Jens Axboe320ae512013-10-24 09:20:05 +01003168 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02003169 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe73c10102011-03-08 13:19:51 +01003170 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003171 * Store ordering should not be needed here, since a potential
3172 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01003173 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003174 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01003175}
3176EXPORT_SYMBOL(blk_start_plug);
3177
3178static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3179{
3180 struct request *rqa = container_of(a, struct request, queuelist);
3181 struct request *rqb = container_of(b, struct request, queuelist);
3182
Jianpeng Ma975927b2012-10-25 21:58:17 +02003183 return !(rqa->q < rqb->q ||
3184 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
Jens Axboe73c10102011-03-08 13:19:51 +01003185}
3186
Jens Axboe49cac012011-04-16 13:51:05 +02003187/*
3188 * If 'from_schedule' is true, then postpone the dispatch of requests
3189 * until a safe kblockd context. We due this to avoid accidental big
3190 * additional stack usage in driver dispatch, in places where the originally
3191 * plugger did not intend it.
3192 */
Jens Axboef6603782011-04-15 15:49:07 +02003193static void queue_unplugged(struct request_queue *q, unsigned int depth,
Jens Axboe49cac012011-04-16 13:51:05 +02003194 bool from_schedule)
Jens Axboe99e22592011-04-18 09:59:55 +02003195 __releases(q->queue_lock)
Jens Axboe94b5eb22011-04-12 10:12:19 +02003196{
Jens Axboe49cac012011-04-16 13:51:05 +02003197 trace_block_unplug(q, depth, !from_schedule);
Jens Axboe99e22592011-04-18 09:59:55 +02003198
Bart Van Assche70460572012-11-28 13:45:56 +01003199 if (from_schedule)
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003200 blk_run_queue_async(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003201 else
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02003202 __blk_run_queue(q);
Bart Van Assche70460572012-11-28 13:45:56 +01003203 spin_unlock(q->queue_lock);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003204}
3205
NeilBrown74018dc2012-07-31 09:08:15 +02003206static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02003207{
3208 LIST_HEAD(callbacks);
3209
Shaohua Li2a7d5552012-07-31 09:08:15 +02003210 while (!list_empty(&plug->cb_list)) {
3211 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02003212
Shaohua Li2a7d5552012-07-31 09:08:15 +02003213 while (!list_empty(&callbacks)) {
3214 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02003215 struct blk_plug_cb,
3216 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003217 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02003218 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02003219 }
NeilBrown048c9372011-04-18 09:52:22 +02003220 }
3221}
3222
NeilBrown9cbb1752012-07-31 09:08:14 +02003223struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3224 int size)
3225{
3226 struct blk_plug *plug = current->plug;
3227 struct blk_plug_cb *cb;
3228
3229 if (!plug)
3230 return NULL;
3231
3232 list_for_each_entry(cb, &plug->cb_list, list)
3233 if (cb->callback == unplug && cb->data == data)
3234 return cb;
3235
3236 /* Not currently on the callback list */
3237 BUG_ON(size < sizeof(*cb));
3238 cb = kzalloc(size, GFP_ATOMIC);
3239 if (cb) {
3240 cb->data = data;
3241 cb->callback = unplug;
3242 list_add(&cb->list, &plug->cb_list);
3243 }
3244 return cb;
3245}
3246EXPORT_SYMBOL(blk_check_plugged);
3247
Jens Axboe49cac012011-04-16 13:51:05 +02003248void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01003249{
3250 struct request_queue *q;
3251 unsigned long flags;
3252 struct request *rq;
NeilBrown109b8122011-04-11 14:13:10 +02003253 LIST_HEAD(list);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003254 unsigned int depth;
Jens Axboe73c10102011-03-08 13:19:51 +01003255
NeilBrown74018dc2012-07-31 09:08:15 +02003256 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01003257
3258 if (!list_empty(&plug->mq_list))
3259 blk_mq_flush_plug_list(plug, from_schedule);
3260
Jens Axboe73c10102011-03-08 13:19:51 +01003261 if (list_empty(&plug->list))
3262 return;
3263
NeilBrown109b8122011-04-11 14:13:10 +02003264 list_splice_init(&plug->list, &list);
3265
Jianpeng Ma422765c2013-01-11 14:46:09 +01003266 list_sort(NULL, &list, plug_rq_cmp);
Jens Axboe73c10102011-03-08 13:19:51 +01003267
3268 q = NULL;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003269 depth = 0;
Jens Axboe18811272011-04-12 10:11:24 +02003270
3271 /*
3272 * Save and disable interrupts here, to avoid doing it for every
3273 * queue lock we have to take.
3274 */
Jens Axboe73c10102011-03-08 13:19:51 +01003275 local_irq_save(flags);
NeilBrown109b8122011-04-11 14:13:10 +02003276 while (!list_empty(&list)) {
3277 rq = list_entry_rq(list.next);
Jens Axboe73c10102011-03-08 13:19:51 +01003278 list_del_init(&rq->queuelist);
Jens Axboe73c10102011-03-08 13:19:51 +01003279 BUG_ON(!rq->q);
3280 if (rq->q != q) {
Jens Axboe99e22592011-04-18 09:59:55 +02003281 /*
3282 * This drops the queue lock
3283 */
3284 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003285 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003286 q = rq->q;
Jens Axboe94b5eb22011-04-12 10:12:19 +02003287 depth = 0;
Jens Axboe73c10102011-03-08 13:19:51 +01003288 spin_lock(q->queue_lock);
3289 }
Tejun Heo8ba61432011-12-14 00:33:37 +01003290
3291 /*
3292 * Short-circuit if @q is dead
3293 */
Bart Van Assche3f3299d2012-11-28 13:42:38 +01003294 if (unlikely(blk_queue_dying(q))) {
Tejun Heo8ba61432011-12-14 00:33:37 +01003295 __blk_end_request_all(rq, -ENODEV);
3296 continue;
3297 }
3298
Jens Axboe73c10102011-03-08 13:19:51 +01003299 /*
3300 * rq is already accounted, so use raw insert
3301 */
Mike Christie28a8f0d2016-06-05 14:32:25 -05003302 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
Jens Axboe401a18e2011-03-25 16:57:52 +01003303 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3304 else
3305 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
Jens Axboe94b5eb22011-04-12 10:12:19 +02003306
3307 depth++;
Jens Axboe73c10102011-03-08 13:19:51 +01003308 }
3309
Jens Axboe99e22592011-04-18 09:59:55 +02003310 /*
3311 * This drops the queue lock
3312 */
3313 if (q)
Jens Axboe49cac012011-04-16 13:51:05 +02003314 queue_unplugged(q, depth, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01003315
Jens Axboe73c10102011-03-08 13:19:51 +01003316 local_irq_restore(flags);
3317}
Jens Axboe73c10102011-03-08 13:19:51 +01003318
3319void blk_finish_plug(struct blk_plug *plug)
3320{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003321 if (plug != current->plug)
3322 return;
Jens Axboef6603782011-04-15 15:49:07 +02003323 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02003324
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07003325 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01003326}
3327EXPORT_SYMBOL(blk_finish_plug);
3328
Jens Axboe05229be2015-11-05 10:44:55 -07003329bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3330{
3331 struct blk_plug *plug;
3332 long state;
Stephen Bates6e219352016-09-13 12:23:15 -06003333 unsigned int queue_num;
3334 struct blk_mq_hw_ctx *hctx;
Jens Axboe05229be2015-11-05 10:44:55 -07003335
3336 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3337 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3338 return false;
3339
Stephen Bates6e219352016-09-13 12:23:15 -06003340 queue_num = blk_qc_t_to_queue_num(cookie);
3341 hctx = q->queue_hw_ctx[queue_num];
3342 hctx->poll_considered++;
3343
Jens Axboe05229be2015-11-05 10:44:55 -07003344 plug = current->plug;
3345 if (plug)
3346 blk_flush_plug_list(plug, false);
3347
3348 state = current->state;
3349 while (!need_resched()) {
Jens Axboe05229be2015-11-05 10:44:55 -07003350 int ret;
3351
3352 hctx->poll_invoked++;
3353
3354 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3355 if (ret > 0) {
3356 hctx->poll_success++;
3357 set_current_state(TASK_RUNNING);
3358 return true;
3359 }
3360
3361 if (signal_pending_state(state, current))
3362 set_current_state(TASK_RUNNING);
3363
3364 if (current->state == TASK_RUNNING)
3365 return true;
3366 if (ret < 0)
3367 break;
3368 cpu_relax();
3369 }
3370
3371 return false;
3372}
Sagi Grimberg9645c1a2016-06-21 18:04:19 +02003373EXPORT_SYMBOL_GPL(blk_poll);
Jens Axboe05229be2015-11-05 10:44:55 -07003374
Rafael J. Wysocki47fafbc2014-12-04 01:00:23 +01003375#ifdef CONFIG_PM
Lin Ming6c954662013-03-23 11:42:26 +08003376/**
3377 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3378 * @q: the queue of the device
3379 * @dev: the device the queue belongs to
3380 *
3381 * Description:
3382 * Initialize runtime-PM-related fields for @q and start auto suspend for
3383 * @dev. Drivers that want to take advantage of request-based runtime PM
3384 * should call this function after @dev has been initialized, and its
3385 * request queue @q has been allocated, and runtime PM for it can not happen
3386 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3387 * cases, driver should call this function before any I/O has taken place.
3388 *
3389 * This function takes care of setting up using auto suspend for the device,
3390 * the autosuspend delay is set to -1 to make runtime suspend impossible
3391 * until an updated value is either set by user or by driver. Drivers do
3392 * not need to touch other autosuspend settings.
3393 *
3394 * The block layer runtime PM is request based, so only works for drivers
3395 * that use request as their IO unit instead of those directly use bio's.
3396 */
3397void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3398{
3399 q->dev = dev;
3400 q->rpm_status = RPM_ACTIVE;
3401 pm_runtime_set_autosuspend_delay(q->dev, -1);
3402 pm_runtime_use_autosuspend(q->dev);
3403}
3404EXPORT_SYMBOL(blk_pm_runtime_init);
3405
3406/**
3407 * blk_pre_runtime_suspend - Pre runtime suspend check
3408 * @q: the queue of the device
3409 *
3410 * Description:
3411 * This function will check if runtime suspend is allowed for the device
3412 * by examining if there are any requests pending in the queue. If there
3413 * are requests pending, the device can not be runtime suspended; otherwise,
3414 * the queue's status will be updated to SUSPENDING and the driver can
3415 * proceed to suspend the device.
3416 *
3417 * For the not allowed case, we mark last busy for the device so that
3418 * runtime PM core will try to autosuspend it some time later.
3419 *
3420 * This function should be called near the start of the device's
3421 * runtime_suspend callback.
3422 *
3423 * Return:
3424 * 0 - OK to runtime suspend the device
3425 * -EBUSY - Device should not be runtime suspended
3426 */
3427int blk_pre_runtime_suspend(struct request_queue *q)
3428{
3429 int ret = 0;
3430
Ken Xue4fd41a852015-12-01 14:45:46 +08003431 if (!q->dev)
3432 return ret;
3433
Lin Ming6c954662013-03-23 11:42:26 +08003434 spin_lock_irq(q->queue_lock);
3435 if (q->nr_pending) {
3436 ret = -EBUSY;
3437 pm_runtime_mark_last_busy(q->dev);
3438 } else {
3439 q->rpm_status = RPM_SUSPENDING;
3440 }
3441 spin_unlock_irq(q->queue_lock);
3442 return ret;
3443}
3444EXPORT_SYMBOL(blk_pre_runtime_suspend);
3445
3446/**
3447 * blk_post_runtime_suspend - Post runtime suspend processing
3448 * @q: the queue of the device
3449 * @err: return value of the device's runtime_suspend function
3450 *
3451 * Description:
3452 * Update the queue's runtime status according to the return value of the
3453 * device's runtime suspend function and mark last busy for the device so
3454 * that PM core will try to auto suspend the device at a later time.
3455 *
3456 * This function should be called near the end of the device's
3457 * runtime_suspend callback.
3458 */
3459void blk_post_runtime_suspend(struct request_queue *q, int err)
3460{
Ken Xue4fd41a852015-12-01 14:45:46 +08003461 if (!q->dev)
3462 return;
3463
Lin Ming6c954662013-03-23 11:42:26 +08003464 spin_lock_irq(q->queue_lock);
3465 if (!err) {
3466 q->rpm_status = RPM_SUSPENDED;
3467 } else {
3468 q->rpm_status = RPM_ACTIVE;
3469 pm_runtime_mark_last_busy(q->dev);
3470 }
3471 spin_unlock_irq(q->queue_lock);
3472}
3473EXPORT_SYMBOL(blk_post_runtime_suspend);
3474
3475/**
3476 * blk_pre_runtime_resume - Pre runtime resume processing
3477 * @q: the queue of the device
3478 *
3479 * Description:
3480 * Update the queue's runtime status to RESUMING in preparation for the
3481 * runtime resume of the device.
3482 *
3483 * This function should be called near the start of the device's
3484 * runtime_resume callback.
3485 */
3486void blk_pre_runtime_resume(struct request_queue *q)
3487{
Ken Xue4fd41a852015-12-01 14:45:46 +08003488 if (!q->dev)
3489 return;
3490
Lin Ming6c954662013-03-23 11:42:26 +08003491 spin_lock_irq(q->queue_lock);
3492 q->rpm_status = RPM_RESUMING;
3493 spin_unlock_irq(q->queue_lock);
3494}
3495EXPORT_SYMBOL(blk_pre_runtime_resume);
3496
3497/**
3498 * blk_post_runtime_resume - Post runtime resume processing
3499 * @q: the queue of the device
3500 * @err: return value of the device's runtime_resume function
3501 *
3502 * Description:
3503 * Update the queue's runtime status according to the return value of the
3504 * device's runtime_resume function. If it is successfully resumed, process
3505 * the requests that are queued into the device's queue when it is resuming
3506 * and then mark last busy and initiate autosuspend for it.
3507 *
3508 * This function should be called near the end of the device's
3509 * runtime_resume callback.
3510 */
3511void blk_post_runtime_resume(struct request_queue *q, int err)
3512{
Ken Xue4fd41a852015-12-01 14:45:46 +08003513 if (!q->dev)
3514 return;
3515
Lin Ming6c954662013-03-23 11:42:26 +08003516 spin_lock_irq(q->queue_lock);
3517 if (!err) {
3518 q->rpm_status = RPM_ACTIVE;
3519 __blk_run_queue(q);
3520 pm_runtime_mark_last_busy(q->dev);
Aaron Luc60855c2013-05-17 15:47:20 +08003521 pm_request_autosuspend(q->dev);
Lin Ming6c954662013-03-23 11:42:26 +08003522 } else {
3523 q->rpm_status = RPM_SUSPENDED;
3524 }
3525 spin_unlock_irq(q->queue_lock);
3526}
3527EXPORT_SYMBOL(blk_post_runtime_resume);
Mika Westerbergd07ab6d2016-02-18 10:54:11 +02003528
3529/**
3530 * blk_set_runtime_active - Force runtime status of the queue to be active
3531 * @q: the queue of the device
3532 *
3533 * If the device is left runtime suspended during system suspend the resume
3534 * hook typically resumes the device and corrects runtime status
3535 * accordingly. However, that does not affect the queue runtime PM status
3536 * which is still "suspended". This prevents processing requests from the
3537 * queue.
3538 *
3539 * This function can be used in driver's resume hook to correct queue
3540 * runtime PM status and re-enable peeking requests from the queue. It
3541 * should be called before first request is added to the queue.
3542 */
3543void blk_set_runtime_active(struct request_queue *q)
3544{
3545 spin_lock_irq(q->queue_lock);
3546 q->rpm_status = RPM_ACTIVE;
3547 pm_runtime_mark_last_busy(q->dev);
3548 pm_request_autosuspend(q->dev);
3549 spin_unlock_irq(q->queue_lock);
3550}
3551EXPORT_SYMBOL(blk_set_runtime_active);
Lin Ming6c954662013-03-23 11:42:26 +08003552#endif
3553
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554int __init blk_dev_init(void)
3555{
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003556 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05303557 FIELD_SIZEOF(struct request, cmd_flags));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02003558
Tejun Heo89b90be2011-01-03 15:01:47 +01003559 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3560 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02003561 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 if (!kblockd_workqueue)
3563 panic("Failed to create kblockd\n");
3564
3565 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003566 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
Ilya Dryomovc2789bd2015-11-20 22:16:46 +01003568 blk_requestq_cachep = kmem_cache_create("request_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003569 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 return 0;
3572}