]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - block/blk-core.c
drbd: Do not redefine __must_hold()
[mirror_ubuntu-eoan-kernel.git] / block / blk-core.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * 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>
6728cb0e
JA
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
1da177e4
LT
8 * 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 */
1da177e4
LT
14#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>
320ae51f 19#include <linux/blk-mq.h>
1da177e4
LT
20#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>
1da177e4
LT
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
faccbd4b 29#include <linux/task_io_accounting_ops.h>
c17bb495 30#include <linux/fault-inject.h>
73c10101 31#include <linux/list_sort.h>
e3c78ca5 32#include <linux/delay.h>
aaf7c680 33#include <linux/ratelimit.h>
6c954667 34#include <linux/pm_runtime.h>
eea8f41c 35#include <linux/blk-cgroup.h>
18fbda91 36#include <linux/debugfs.h>
30abb3a6 37#include <linux/bpf.h>
55782138
LZ
38
39#define CREATE_TRACE_POINTS
40#include <trace/events/block.h>
1da177e4 41
8324aa91 42#include "blk.h"
43a5e4e2 43#include "blk-mq.h"
bd166ef1 44#include "blk-mq-sched.h"
87760e5e 45#include "blk-wbt.h"
8324aa91 46
18fbda91
OS
47#ifdef CONFIG_DEBUG_FS
48struct dentry *blk_debugfs_root;
49#endif
50
d07335e5 51EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
b0da3f0d 52EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
0a82a8d1 53EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
3291fa57 54EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
cbae8d45 55EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
0bfc2455 56
a73f730d
TH
57DEFINE_IDA(blk_queue_ida);
58
1da177e4
LT
59/*
60 * For the allocated request tables
61 */
d674d414 62struct kmem_cache *request_cachep;
1da177e4
LT
63
64/*
65 * For queue allocation
66 */
6728cb0e 67struct kmem_cache *blk_requestq_cachep;
1da177e4 68
1da177e4
LT
69/*
70 * Controlling structure to kblockd
71 */
ff856bad 72static struct workqueue_struct *kblockd_workqueue;
1da177e4 73
8814ce8a
BVA
74/**
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
77 * @q: request queue
78 */
79void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80{
81 unsigned long flags;
82
83 spin_lock_irqsave(q->queue_lock, flags);
84 queue_flag_set(flag, q);
85 spin_unlock_irqrestore(q->queue_lock, flags);
86}
87EXPORT_SYMBOL(blk_queue_flag_set);
88
89/**
90 * blk_queue_flag_clear - atomically clear a queue flag
91 * @flag: flag to be cleared
92 * @q: request queue
93 */
94void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
95{
96 unsigned long flags;
97
98 spin_lock_irqsave(q->queue_lock, flags);
99 queue_flag_clear(flag, q);
100 spin_unlock_irqrestore(q->queue_lock, flags);
101}
102EXPORT_SYMBOL(blk_queue_flag_clear);
103
104/**
105 * blk_queue_flag_test_and_set - atomically test and set a queue flag
106 * @flag: flag to be set
107 * @q: request queue
108 *
109 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
110 * the flag was already set.
111 */
112bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
113{
114 unsigned long flags;
115 bool res;
116
117 spin_lock_irqsave(q->queue_lock, flags);
118 res = queue_flag_test_and_set(flag, q);
119 spin_unlock_irqrestore(q->queue_lock, flags);
120
121 return res;
122}
123EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
124
125/**
126 * blk_queue_flag_test_and_clear - atomically test and clear a queue flag
127 * @flag: flag to be cleared
128 * @q: request queue
129 *
130 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
131 * the flag was set.
132 */
133bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q)
134{
135 unsigned long flags;
136 bool res;
137
138 spin_lock_irqsave(q->queue_lock, flags);
139 res = queue_flag_test_and_clear(flag, q);
140 spin_unlock_irqrestore(q->queue_lock, flags);
141
142 return res;
143}
144EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_clear);
145
d40f75a0
TH
146static void blk_clear_congested(struct request_list *rl, int sync)
147{
d40f75a0
TH
148#ifdef CONFIG_CGROUP_WRITEBACK
149 clear_wb_congested(rl->blkg->wb_congested, sync);
150#else
482cf79c
TH
151 /*
152 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
153 * flip its congestion state for events on other blkcgs.
154 */
155 if (rl == &rl->q->root_rl)
dc3b17cc 156 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
d40f75a0
TH
157#endif
158}
159
160static void blk_set_congested(struct request_list *rl, int sync)
161{
d40f75a0
TH
162#ifdef CONFIG_CGROUP_WRITEBACK
163 set_wb_congested(rl->blkg->wb_congested, sync);
164#else
482cf79c
TH
165 /* see blk_clear_congested() */
166 if (rl == &rl->q->root_rl)
dc3b17cc 167 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
d40f75a0
TH
168#endif
169}
170
8324aa91 171void blk_queue_congestion_threshold(struct request_queue *q)
1da177e4
LT
172{
173 int nr;
174
175 nr = q->nr_requests - (q->nr_requests / 8) + 1;
176 if (nr > q->nr_requests)
177 nr = q->nr_requests;
178 q->nr_congestion_on = nr;
179
180 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
181 if (nr < 1)
182 nr = 1;
183 q->nr_congestion_off = nr;
184}
185
2a4aa30c 186void blk_rq_init(struct request_queue *q, struct request *rq)
1da177e4 187{
1afb20f3
FT
188 memset(rq, 0, sizeof(*rq));
189
1da177e4 190 INIT_LIST_HEAD(&rq->queuelist);
242f9dcb 191 INIT_LIST_HEAD(&rq->timeout_list);
c7c22e4d 192 rq->cpu = -1;
63a71386 193 rq->q = q;
a2dec7b3 194 rq->__sector = (sector_t) -1;
2e662b65
JA
195 INIT_HLIST_NODE(&rq->hash);
196 RB_CLEAR_NODE(&rq->rb_node);
63a71386 197 rq->tag = -1;
bd166ef1 198 rq->internal_tag = -1;
522a7775 199 rq->start_time_ns = ktime_get_ns();
09e099d4 200 rq->part = NULL;
1da177e4 201}
2a4aa30c 202EXPORT_SYMBOL(blk_rq_init);
1da177e4 203
2a842aca
CH
204static const struct {
205 int errno;
206 const char *name;
207} blk_errors[] = {
208 [BLK_STS_OK] = { 0, "" },
209 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
210 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
211 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
212 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
213 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
214 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
215 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
216 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
217 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
86ff7c2a 218 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
03a07c92 219 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
2a842aca 220
4e4cbee9
CH
221 /* device mapper special case, should not leak out: */
222 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
223
2a842aca
CH
224 /* everything else not covered above: */
225 [BLK_STS_IOERR] = { -EIO, "I/O" },
226};
227
228blk_status_t errno_to_blk_status(int errno)
229{
230 int i;
231
232 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
233 if (blk_errors[i].errno == errno)
234 return (__force blk_status_t)i;
235 }
236
237 return BLK_STS_IOERR;
238}
239EXPORT_SYMBOL_GPL(errno_to_blk_status);
240
241int blk_status_to_errno(blk_status_t status)
242{
243 int idx = (__force int)status;
244
34bd9c1c 245 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
2a842aca
CH
246 return -EIO;
247 return blk_errors[idx].errno;
248}
249EXPORT_SYMBOL_GPL(blk_status_to_errno);
250
251static void print_req_error(struct request *req, blk_status_t status)
252{
253 int idx = (__force int)status;
254
34bd9c1c 255 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
2a842aca
CH
256 return;
257
258 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
259 __func__, blk_errors[idx].name, req->rq_disk ?
260 req->rq_disk->disk_name : "?",
261 (unsigned long long)blk_rq_pos(req));
262}
263
5bb23a68 264static void req_bio_endio(struct request *rq, struct bio *bio,
2a842aca 265 unsigned int nbytes, blk_status_t error)
1da177e4 266{
78d8e58a 267 if (error)
4e4cbee9 268 bio->bi_status = error;
797e7dbb 269
e8064021 270 if (unlikely(rq->rq_flags & RQF_QUIET))
b7c44ed9 271 bio_set_flag(bio, BIO_QUIET);
08bafc03 272
f79ea416 273 bio_advance(bio, nbytes);
7ba1ba12 274
143a87f4 275 /* don't actually finish bio if it's part of flush sequence */
e8064021 276 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
4246a0b6 277 bio_endio(bio);
1da177e4 278}
1da177e4 279
1da177e4
LT
280void blk_dump_rq_flags(struct request *rq, char *msg)
281{
aebf526b
CH
282 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
283 rq->rq_disk ? rq->rq_disk->disk_name : "?",
5953316d 284 (unsigned long long) rq->cmd_flags);
1da177e4 285
83096ebf
TH
286 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
287 (unsigned long long)blk_rq_pos(rq),
288 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
b4f42e28
JA
289 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
290 rq->bio, rq->biotail, blk_rq_bytes(rq));
1da177e4 291}
1da177e4
LT
292EXPORT_SYMBOL(blk_dump_rq_flags);
293
3cca6dc1 294static void blk_delay_work(struct work_struct *work)
1da177e4 295{
3cca6dc1 296 struct request_queue *q;
1da177e4 297
3cca6dc1
JA
298 q = container_of(work, struct request_queue, delay_work.work);
299 spin_lock_irq(q->queue_lock);
24ecfbe2 300 __blk_run_queue(q);
3cca6dc1 301 spin_unlock_irq(q->queue_lock);
1da177e4 302}
1da177e4
LT
303
304/**
3cca6dc1
JA
305 * blk_delay_queue - restart queueing after defined interval
306 * @q: The &struct request_queue in question
307 * @msecs: Delay in msecs
1da177e4
LT
308 *
309 * Description:
3cca6dc1
JA
310 * Sometimes queueing needs to be postponed for a little while, to allow
311 * resources to come back. This function will make sure that queueing is
2fff8a92 312 * restarted around the specified time.
3cca6dc1
JA
313 */
314void blk_delay_queue(struct request_queue *q, unsigned long msecs)
2ad8b1ef 315{
2fff8a92 316 lockdep_assert_held(q->queue_lock);
332ebbf7 317 WARN_ON_ONCE(q->mq_ops);
2fff8a92 318
70460571
BVA
319 if (likely(!blk_queue_dead(q)))
320 queue_delayed_work(kblockd_workqueue, &q->delay_work,
321 msecs_to_jiffies(msecs));
2ad8b1ef 322}
3cca6dc1 323EXPORT_SYMBOL(blk_delay_queue);
2ad8b1ef 324
21491412
JA
325/**
326 * blk_start_queue_async - asynchronously restart a previously stopped queue
327 * @q: The &struct request_queue in question
328 *
329 * Description:
330 * blk_start_queue_async() will clear the stop flag on the queue, and
331 * ensure that the request_fn for the queue is run from an async
332 * context.
333 **/
334void blk_start_queue_async(struct request_queue *q)
335{
2fff8a92 336 lockdep_assert_held(q->queue_lock);
332ebbf7 337 WARN_ON_ONCE(q->mq_ops);
2fff8a92 338
21491412
JA
339 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
340 blk_run_queue_async(q);
341}
342EXPORT_SYMBOL(blk_start_queue_async);
343
1da177e4
LT
344/**
345 * blk_start_queue - restart a previously stopped queue
165125e1 346 * @q: The &struct request_queue in question
1da177e4
LT
347 *
348 * Description:
349 * blk_start_queue() will clear the stop flag on the queue, and call
350 * the request_fn for the queue if it was in a stopped state when
2fff8a92 351 * entered. Also see blk_stop_queue().
1da177e4 352 **/
165125e1 353void blk_start_queue(struct request_queue *q)
1da177e4 354{
2fff8a92 355 lockdep_assert_held(q->queue_lock);
332ebbf7 356 WARN_ON_ONCE(q->mq_ops);
a038e253 357
75ad23bc 358 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
24ecfbe2 359 __blk_run_queue(q);
1da177e4 360}
1da177e4
LT
361EXPORT_SYMBOL(blk_start_queue);
362
363/**
364 * blk_stop_queue - stop a queue
165125e1 365 * @q: The &struct request_queue in question
1da177e4
LT
366 *
367 * Description:
368 * The Linux block layer assumes that a block driver will consume all
369 * entries on the request queue when the request_fn strategy is called.
370 * Often this will not happen, because of hardware limitations (queue
371 * depth settings). If a device driver gets a 'queue full' response,
372 * or if it simply chooses not to queue more I/O at one point, it can
373 * call this function to prevent the request_fn from being called until
374 * the driver has signalled it's ready to go again. This happens by calling
2fff8a92 375 * blk_start_queue() to restart queue operations.
1da177e4 376 **/
165125e1 377void blk_stop_queue(struct request_queue *q)
1da177e4 378{
2fff8a92 379 lockdep_assert_held(q->queue_lock);
332ebbf7 380 WARN_ON_ONCE(q->mq_ops);
2fff8a92 381
136b5721 382 cancel_delayed_work(&q->delay_work);
75ad23bc 383 queue_flag_set(QUEUE_FLAG_STOPPED, q);
1da177e4
LT
384}
385EXPORT_SYMBOL(blk_stop_queue);
386
387/**
388 * blk_sync_queue - cancel any pending callbacks on a queue
389 * @q: the queue
390 *
391 * Description:
392 * The block layer may perform asynchronous callback activity
393 * on a queue, such as calling the unplug function after a timeout.
394 * A block device may call blk_sync_queue to ensure that any
395 * such activity is cancelled, thus allowing it to release resources
59c51591 396 * that the callbacks might use. The caller must already have made sure
1da177e4
LT
397 * that its ->make_request_fn will not re-add plugging prior to calling
398 * this function.
399 *
da527770 400 * This function does not cancel any asynchronous activity arising
da3dae54 401 * out of elevator or throttling code. That would require elevator_exit()
5efd6113 402 * and blkcg_exit_queue() to be called with queue lock initialized.
da527770 403 *
1da177e4
LT
404 */
405void blk_sync_queue(struct request_queue *q)
406{
70ed28b9 407 del_timer_sync(&q->timeout);
4e9b6f20 408 cancel_work_sync(&q->timeout_work);
f04c1fe7
ML
409
410 if (q->mq_ops) {
411 struct blk_mq_hw_ctx *hctx;
412 int i;
413
aba7afc5 414 cancel_delayed_work_sync(&q->requeue_work);
21c6e939 415 queue_for_each_hw_ctx(q, hctx, i)
9f993737 416 cancel_delayed_work_sync(&hctx->run_work);
f04c1fe7
ML
417 } else {
418 cancel_delayed_work_sync(&q->delay_work);
419 }
1da177e4
LT
420}
421EXPORT_SYMBOL(blk_sync_queue);
422
c9254f2d
BVA
423/**
424 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
425 * @q: request queue pointer
426 *
427 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
428 * set and 1 if the flag was already set.
429 */
430int blk_set_preempt_only(struct request_queue *q)
431{
8814ce8a 432 return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
c9254f2d
BVA
433}
434EXPORT_SYMBOL_GPL(blk_set_preempt_only);
435
436void blk_clear_preempt_only(struct request_queue *q)
437{
8814ce8a 438 blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
3a0a5299 439 wake_up_all(&q->mq_freeze_wq);
c9254f2d
BVA
440}
441EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
442
c246e80d
BVA
443/**
444 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
445 * @q: The queue to run
446 *
447 * Description:
448 * Invoke request handling on a queue if there are any pending requests.
449 * May be used to restart request handling after a request has completed.
450 * This variant runs the queue whether or not the queue has been
451 * stopped. Must be called with the queue lock held and interrupts
452 * disabled. See also @blk_run_queue.
453 */
454inline void __blk_run_queue_uncond(struct request_queue *q)
455{
2fff8a92 456 lockdep_assert_held(q->queue_lock);
332ebbf7 457 WARN_ON_ONCE(q->mq_ops);
2fff8a92 458
c246e80d
BVA
459 if (unlikely(blk_queue_dead(q)))
460 return;
461
24faf6f6
BVA
462 /*
463 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
464 * the queue lock internally. As a result multiple threads may be
465 * running such a request function concurrently. Keep track of the
466 * number of active request_fn invocations such that blk_drain_queue()
467 * can wait until all these request_fn calls have finished.
468 */
469 q->request_fn_active++;
c246e80d 470 q->request_fn(q);
24faf6f6 471 q->request_fn_active--;
c246e80d 472}
a7928c15 473EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
c246e80d 474
1da177e4 475/**
80a4b58e 476 * __blk_run_queue - run a single device queue
1da177e4 477 * @q: The queue to run
80a4b58e
JA
478 *
479 * Description:
2fff8a92 480 * See @blk_run_queue.
1da177e4 481 */
24ecfbe2 482void __blk_run_queue(struct request_queue *q)
1da177e4 483{
2fff8a92 484 lockdep_assert_held(q->queue_lock);
332ebbf7 485 WARN_ON_ONCE(q->mq_ops);
2fff8a92 486
a538cd03
TH
487 if (unlikely(blk_queue_stopped(q)))
488 return;
489
c246e80d 490 __blk_run_queue_uncond(q);
75ad23bc
NP
491}
492EXPORT_SYMBOL(__blk_run_queue);
dac07ec1 493
24ecfbe2
CH
494/**
495 * blk_run_queue_async - run a single device queue in workqueue context
496 * @q: The queue to run
497 *
498 * Description:
499 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
2fff8a92
BVA
500 * of us.
501 *
502 * Note:
503 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
504 * has canceled q->delay_work, callers must hold the queue lock to avoid
505 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
24ecfbe2
CH
506 */
507void blk_run_queue_async(struct request_queue *q)
508{
2fff8a92 509 lockdep_assert_held(q->queue_lock);
332ebbf7 510 WARN_ON_ONCE(q->mq_ops);
2fff8a92 511
70460571 512 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
e7c2f967 513 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
24ecfbe2 514}
c21e6beb 515EXPORT_SYMBOL(blk_run_queue_async);
24ecfbe2 516
75ad23bc
NP
517/**
518 * blk_run_queue - run a single device queue
519 * @q: The queue to run
80a4b58e
JA
520 *
521 * Description:
522 * Invoke request handling on this queue, if it has pending work to do.
a7f55792 523 * May be used to restart queueing when a request has completed.
75ad23bc
NP
524 */
525void blk_run_queue(struct request_queue *q)
526{
527 unsigned long flags;
528
332ebbf7
BVA
529 WARN_ON_ONCE(q->mq_ops);
530
75ad23bc 531 spin_lock_irqsave(q->queue_lock, flags);
24ecfbe2 532 __blk_run_queue(q);
1da177e4
LT
533 spin_unlock_irqrestore(q->queue_lock, flags);
534}
535EXPORT_SYMBOL(blk_run_queue);
536
165125e1 537void blk_put_queue(struct request_queue *q)
483f4afc
AV
538{
539 kobject_put(&q->kobj);
540}
d86e0e83 541EXPORT_SYMBOL(blk_put_queue);
483f4afc 542
e3c78ca5 543/**
807592a4 544 * __blk_drain_queue - drain requests from request_queue
e3c78ca5 545 * @q: queue to drain
c9a929dd 546 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
e3c78ca5 547 *
c9a929dd
TH
548 * Drain requests from @q. If @drain_all is set, all requests are drained.
549 * If not, only ELVPRIV requests are drained. The caller is responsible
550 * for ensuring that no new requests which need to be drained are queued.
e3c78ca5 551 */
807592a4
BVA
552static void __blk_drain_queue(struct request_queue *q, bool drain_all)
553 __releases(q->queue_lock)
554 __acquires(q->queue_lock)
e3c78ca5 555{
458f27a9
AH
556 int i;
557
807592a4 558 lockdep_assert_held(q->queue_lock);
332ebbf7 559 WARN_ON_ONCE(q->mq_ops);
807592a4 560
e3c78ca5 561 while (true) {
481a7d64 562 bool drain = false;
e3c78ca5 563
b855b04a
TH
564 /*
565 * The caller might be trying to drain @q before its
566 * elevator is initialized.
567 */
568 if (q->elevator)
569 elv_drain_elevator(q);
570
5efd6113 571 blkcg_drain_queue(q);
e3c78ca5 572
4eabc941
TH
573 /*
574 * This function might be called on a queue which failed
b855b04a
TH
575 * driver init after queue creation or is not yet fully
576 * active yet. Some drivers (e.g. fd and loop) get unhappy
577 * in such cases. Kick queue iff dispatch queue has
578 * something on it and @q has request_fn set.
4eabc941 579 */
b855b04a 580 if (!list_empty(&q->queue_head) && q->request_fn)
4eabc941 581 __blk_run_queue(q);
c9a929dd 582
8a5ecdd4 583 drain |= q->nr_rqs_elvpriv;
24faf6f6 584 drain |= q->request_fn_active;
481a7d64
TH
585
586 /*
587 * Unfortunately, requests are queued at and tracked from
588 * multiple places and there's no single counter which can
589 * be drained. Check all the queues and counters.
590 */
591 if (drain_all) {
e97c293c 592 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
481a7d64
TH
593 drain |= !list_empty(&q->queue_head);
594 for (i = 0; i < 2; i++) {
8a5ecdd4 595 drain |= q->nr_rqs[i];
481a7d64 596 drain |= q->in_flight[i];
7c94e1c1
ML
597 if (fq)
598 drain |= !list_empty(&fq->flush_queue[i]);
481a7d64
TH
599 }
600 }
e3c78ca5 601
481a7d64 602 if (!drain)
e3c78ca5 603 break;
807592a4
BVA
604
605 spin_unlock_irq(q->queue_lock);
606
e3c78ca5 607 msleep(10);
807592a4
BVA
608
609 spin_lock_irq(q->queue_lock);
e3c78ca5 610 }
458f27a9
AH
611
612 /*
613 * With queue marked dead, any woken up waiter will fail the
614 * allocation path, so the wakeup chaining is lost and we're
615 * left with hung waiters. We need to wake up those waiters.
616 */
617 if (q->request_fn) {
a051661c
TH
618 struct request_list *rl;
619
a051661c
TH
620 blk_queue_for_each_rl(rl, q)
621 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
622 wake_up_all(&rl->wait[i]);
458f27a9 623 }
e3c78ca5
TH
624}
625
454be724
ML
626void blk_drain_queue(struct request_queue *q)
627{
628 spin_lock_irq(q->queue_lock);
629 __blk_drain_queue(q, true);
630 spin_unlock_irq(q->queue_lock);
631}
632
d732580b
TH
633/**
634 * blk_queue_bypass_start - enter queue bypass mode
635 * @q: queue of interest
636 *
637 * In bypass mode, only the dispatch FIFO queue of @q is used. This
638 * function makes @q enter bypass mode and drains all requests which were
6ecf23af 639 * throttled or issued before. On return, it's guaranteed that no request
80fd9979
TH
640 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
641 * inside queue or RCU read lock.
d732580b
TH
642 */
643void blk_queue_bypass_start(struct request_queue *q)
644{
332ebbf7
BVA
645 WARN_ON_ONCE(q->mq_ops);
646
d732580b 647 spin_lock_irq(q->queue_lock);
776687bc 648 q->bypass_depth++;
d732580b
TH
649 queue_flag_set(QUEUE_FLAG_BYPASS, q);
650 spin_unlock_irq(q->queue_lock);
651
776687bc
TH
652 /*
653 * Queues start drained. Skip actual draining till init is
654 * complete. This avoids lenghty delays during queue init which
655 * can happen many times during boot.
656 */
657 if (blk_queue_init_done(q)) {
807592a4
BVA
658 spin_lock_irq(q->queue_lock);
659 __blk_drain_queue(q, false);
660 spin_unlock_irq(q->queue_lock);
661
b82d4b19
TH
662 /* ensure blk_queue_bypass() is %true inside RCU read lock */
663 synchronize_rcu();
664 }
d732580b
TH
665}
666EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
667
668/**
669 * blk_queue_bypass_end - leave queue bypass mode
670 * @q: queue of interest
671 *
672 * Leave bypass mode and restore the normal queueing behavior.
332ebbf7
BVA
673 *
674 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
675 * this function is called for both blk-sq and blk-mq queues.
d732580b
TH
676 */
677void blk_queue_bypass_end(struct request_queue *q)
678{
679 spin_lock_irq(q->queue_lock);
680 if (!--q->bypass_depth)
681 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
682 WARN_ON_ONCE(q->bypass_depth < 0);
683 spin_unlock_irq(q->queue_lock);
684}
685EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
686
aed3ea94
JA
687void blk_set_queue_dying(struct request_queue *q)
688{
8814ce8a 689 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
aed3ea94 690
d3cfb2a0
ML
691 /*
692 * When queue DYING flag is set, we need to block new req
693 * entering queue, so we call blk_freeze_queue_start() to
694 * prevent I/O from crossing blk_queue_enter().
695 */
696 blk_freeze_queue_start(q);
697
aed3ea94
JA
698 if (q->mq_ops)
699 blk_mq_wake_waiters(q);
700 else {
701 struct request_list *rl;
702
bbfc3c5d 703 spin_lock_irq(q->queue_lock);
aed3ea94
JA
704 blk_queue_for_each_rl(rl, q) {
705 if (rl->rq_pool) {
34d9715a
ML
706 wake_up_all(&rl->wait[BLK_RW_SYNC]);
707 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
aed3ea94
JA
708 }
709 }
bbfc3c5d 710 spin_unlock_irq(q->queue_lock);
aed3ea94 711 }
055f6e18
ML
712
713 /* Make blk_queue_enter() reexamine the DYING flag. */
714 wake_up_all(&q->mq_freeze_wq);
aed3ea94
JA
715}
716EXPORT_SYMBOL_GPL(blk_set_queue_dying);
717
c9a929dd
TH
718/**
719 * blk_cleanup_queue - shutdown a request queue
720 * @q: request queue to shutdown
721 *
c246e80d
BVA
722 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
723 * put it. All future requests will be failed immediately with -ENODEV.
c94a96ac 724 */
6728cb0e 725void blk_cleanup_queue(struct request_queue *q)
483f4afc 726{
c9a929dd 727 spinlock_t *lock = q->queue_lock;
e3335de9 728
3f3299d5 729 /* mark @q DYING, no new request or merges will be allowed afterwards */
483f4afc 730 mutex_lock(&q->sysfs_lock);
aed3ea94 731 blk_set_queue_dying(q);
c9a929dd 732 spin_lock_irq(lock);
6ecf23af 733
80fd9979 734 /*
3f3299d5 735 * A dying queue is permanently in bypass mode till released. Note
80fd9979
TH
736 * that, unlike blk_queue_bypass_start(), we aren't performing
737 * synchronize_rcu() after entering bypass mode to avoid the delay
738 * as some drivers create and destroy a lot of queues while
739 * probing. This is still safe because blk_release_queue() will be
740 * called only after the queue refcnt drops to zero and nothing,
741 * RCU or not, would be traversing the queue by then.
742 */
6ecf23af
TH
743 q->bypass_depth++;
744 queue_flag_set(QUEUE_FLAG_BYPASS, q);
745
c9a929dd
TH
746 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
747 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
3f3299d5 748 queue_flag_set(QUEUE_FLAG_DYING, q);
c9a929dd
TH
749 spin_unlock_irq(lock);
750 mutex_unlock(&q->sysfs_lock);
751
c246e80d
BVA
752 /*
753 * Drain all requests queued before DYING marking. Set DEAD flag to
754 * prevent that q->request_fn() gets invoked after draining finished.
755 */
3ef28e83 756 blk_freeze_queue(q);
9c1051aa 757 spin_lock_irq(lock);
c246e80d 758 queue_flag_set(QUEUE_FLAG_DEAD, q);
807592a4 759 spin_unlock_irq(lock);
c9a929dd 760
c2856ae2
ML
761 /*
762 * make sure all in-progress dispatch are completed because
763 * blk_freeze_queue() can only complete all requests, and
764 * dispatch may still be in-progress since we dispatch requests
1311326c
ML
765 * from more than one contexts.
766 *
767 * No need to quiesce queue if it isn't initialized yet since
768 * blk_freeze_queue() should be enough for cases of passthrough
769 * request.
c2856ae2 770 */
1311326c 771 if (q->mq_ops && blk_queue_init_done(q))
c2856ae2
ML
772 blk_mq_quiesce_queue(q);
773
5a48fc14
DW
774 /* for synchronous bio-based driver finish in-flight integrity i/o */
775 blk_flush_integrity();
776
c9a929dd 777 /* @q won't process any more request, flush async actions */
dc3b17cc 778 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
c9a929dd
TH
779 blk_sync_queue(q);
780
a063057d
BVA
781 /*
782 * I/O scheduler exit is only safe after the sysfs scheduler attribute
783 * has been removed.
784 */
785 WARN_ON_ONCE(q->kobj.state_in_sysfs);
786
787 /*
788 * Since the I/O scheduler exit code may access cgroup information,
789 * perform I/O scheduler exit before disassociating from the block
790 * cgroup controller.
791 */
792 if (q->elevator) {
793 ioc_clear_queue(q);
794 elevator_exit(q, q->elevator);
795 q->elevator = NULL;
796 }
797
798 /*
799 * Remove all references to @q from the block cgroup controller before
800 * restoring @q->queue_lock to avoid that restoring this pointer causes
801 * e.g. blkcg_print_blkgs() to crash.
802 */
803 blkcg_exit_queue(q);
804
805 /*
806 * Since the cgroup code may dereference the @q->backing_dev_info
807 * pointer, only decrease its reference count after having removed the
808 * association with the block cgroup controller.
809 */
810 bdi_put(q->backing_dev_info);
811
45a9c9d9
BVA
812 if (q->mq_ops)
813 blk_mq_free_queue(q);
3ef28e83 814 percpu_ref_exit(&q->q_usage_counter);
45a9c9d9 815
5e5cfac0
AH
816 spin_lock_irq(lock);
817 if (q->queue_lock != &q->__queue_lock)
818 q->queue_lock = &q->__queue_lock;
819 spin_unlock_irq(lock);
820
c9a929dd 821 /* @q is and will stay empty, shutdown and put */
483f4afc
AV
822 blk_put_queue(q);
823}
1da177e4
LT
824EXPORT_SYMBOL(blk_cleanup_queue);
825
271508db 826/* Allocate memory local to the request queue */
6d247d7f 827static void *alloc_request_simple(gfp_t gfp_mask, void *data)
271508db 828{
6d247d7f
CH
829 struct request_queue *q = data;
830
831 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
271508db
DR
832}
833
6d247d7f 834static void free_request_simple(void *element, void *data)
271508db
DR
835{
836 kmem_cache_free(request_cachep, element);
837}
838
6d247d7f
CH
839static void *alloc_request_size(gfp_t gfp_mask, void *data)
840{
841 struct request_queue *q = data;
842 struct request *rq;
843
844 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
845 q->node);
846 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
847 kfree(rq);
848 rq = NULL;
849 }
850 return rq;
851}
852
853static void free_request_size(void *element, void *data)
854{
855 struct request_queue *q = data;
856
857 if (q->exit_rq_fn)
858 q->exit_rq_fn(q, element);
859 kfree(element);
860}
861
5b788ce3
TH
862int blk_init_rl(struct request_list *rl, struct request_queue *q,
863 gfp_t gfp_mask)
1da177e4 864{
85acb3ba 865 if (unlikely(rl->rq_pool) || q->mq_ops)
1abec4fd
MS
866 return 0;
867
5b788ce3 868 rl->q = q;
1faa16d2
JA
869 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
870 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
1faa16d2
JA
871 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
872 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
1da177e4 873
6d247d7f
CH
874 if (q->cmd_size) {
875 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
876 alloc_request_size, free_request_size,
877 q, gfp_mask, q->node);
878 } else {
879 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
880 alloc_request_simple, free_request_simple,
881 q, gfp_mask, q->node);
882 }
1da177e4
LT
883 if (!rl->rq_pool)
884 return -ENOMEM;
885
b425e504
BVA
886 if (rl != &q->root_rl)
887 WARN_ON_ONCE(!blk_get_queue(q));
888
1da177e4
LT
889 return 0;
890}
891
b425e504 892void blk_exit_rl(struct request_queue *q, struct request_list *rl)
5b788ce3 893{
b425e504 894 if (rl->rq_pool) {
5b788ce3 895 mempool_destroy(rl->rq_pool);
b425e504
BVA
896 if (rl != &q->root_rl)
897 blk_put_queue(q);
898 }
5b788ce3
TH
899}
900
165125e1 901struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
1da177e4 902{
5ee0524b 903 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
1946089a
CL
904}
905EXPORT_SYMBOL(blk_alloc_queue);
1da177e4 906
3a0a5299
BVA
907/**
908 * blk_queue_enter() - try to increase q->q_usage_counter
909 * @q: request queue pointer
910 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
911 */
9a95e4ef 912int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
3ef28e83 913{
3a0a5299
BVA
914 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
915
3ef28e83 916 while (true) {
3a0a5299 917 bool success = false;
3ef28e83 918
818e0fa2 919 rcu_read_lock();
3a0a5299
BVA
920 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
921 /*
922 * The code that sets the PREEMPT_ONLY flag is
923 * responsible for ensuring that that flag is globally
924 * visible before the queue is unfrozen.
925 */
926 if (preempt || !blk_queue_preempt_only(q)) {
927 success = true;
928 } else {
929 percpu_ref_put(&q->q_usage_counter);
930 }
931 }
818e0fa2 932 rcu_read_unlock();
3a0a5299
BVA
933
934 if (success)
3ef28e83
DW
935 return 0;
936
3a0a5299 937 if (flags & BLK_MQ_REQ_NOWAIT)
3ef28e83
DW
938 return -EBUSY;
939
5ed61d3f 940 /*
1671d522 941 * read pair of barrier in blk_freeze_queue_start(),
5ed61d3f 942 * we need to order reading __PERCPU_REF_DEAD flag of
d3cfb2a0
ML
943 * .q_usage_counter and reading .mq_freeze_depth or
944 * queue dying flag, otherwise the following wait may
945 * never return if the two reads are reordered.
5ed61d3f
ML
946 */
947 smp_rmb();
948
1dc3039b
AJ
949 wait_event(q->mq_freeze_wq,
950 (atomic_read(&q->mq_freeze_depth) == 0 &&
951 (preempt || !blk_queue_preempt_only(q))) ||
952 blk_queue_dying(q));
3ef28e83
DW
953 if (blk_queue_dying(q))
954 return -ENODEV;
3ef28e83
DW
955 }
956}
957
958void blk_queue_exit(struct request_queue *q)
959{
960 percpu_ref_put(&q->q_usage_counter);
961}
962
963static void blk_queue_usage_counter_release(struct percpu_ref *ref)
964{
965 struct request_queue *q =
966 container_of(ref, struct request_queue, q_usage_counter);
967
968 wake_up_all(&q->mq_freeze_wq);
969}
970
bca237a5 971static void blk_rq_timed_out_timer(struct timer_list *t)
287922eb 972{
bca237a5 973 struct request_queue *q = from_timer(q, t, timeout);
287922eb
CH
974
975 kblockd_schedule_work(&q->timeout_work);
976}
977
498f6650
BVA
978/**
979 * blk_alloc_queue_node - allocate a request queue
980 * @gfp_mask: memory allocation flags
981 * @node_id: NUMA node to allocate memory from
982 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
983 * serialize calls to the legacy .request_fn() callback. Ignored for
984 * blk-mq request queues.
985 *
986 * Note: pass the queue lock as the third argument to this function instead of
987 * setting the queue lock pointer explicitly to avoid triggering a sporadic
988 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
989 * the queue lock pointer must be set before blkcg_init_queue() is called.
990 */
5ee0524b
BVA
991struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
992 spinlock_t *lock)
1946089a 993{
165125e1 994 struct request_queue *q;
338aa96d 995 int ret;
1946089a 996
8324aa91 997 q = kmem_cache_alloc_node(blk_requestq_cachep,
94f6030c 998 gfp_mask | __GFP_ZERO, node_id);
1da177e4
LT
999 if (!q)
1000 return NULL;
1001
cbf62af3
CH
1002 INIT_LIST_HEAD(&q->queue_head);
1003 q->last_merge = NULL;
1004 q->end_sector = 0;
1005 q->boundary_rq = NULL;
1006
00380a40 1007 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
a73f730d 1008 if (q->id < 0)
3d2936f4 1009 goto fail_q;
a73f730d 1010
338aa96d
KO
1011 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1012 if (ret)
54efd50b
KO
1013 goto fail_id;
1014
d03f6cdc
JK
1015 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1016 if (!q->backing_dev_info)
1017 goto fail_split;
1018
a83b576c
JA
1019 q->stats = blk_alloc_queue_stats();
1020 if (!q->stats)
1021 goto fail_stats;
1022
dc3b17cc 1023 q->backing_dev_info->ra_pages =
09cbfeaf 1024 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
dc3b17cc
JK
1025 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1026 q->backing_dev_info->name = "block";
5151412d 1027 q->node = node_id;
0989a025 1028
bca237a5
KC
1029 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1030 laptop_mode_timer_fn, 0);
1031 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
4e9b6f20 1032 INIT_WORK(&q->timeout_work, NULL);
b855b04a 1033 INIT_LIST_HEAD(&q->queue_head);
242f9dcb 1034 INIT_LIST_HEAD(&q->timeout_list);
a612fddf 1035 INIT_LIST_HEAD(&q->icq_list);
4eef3049 1036#ifdef CONFIG_BLK_CGROUP
e8989fae 1037 INIT_LIST_HEAD(&q->blkg_list);
4eef3049 1038#endif
3cca6dc1 1039 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
483f4afc 1040
8324aa91 1041 kobject_init(&q->kobj, &blk_queue_ktype);
1da177e4 1042
5acb3cc2
WL
1043#ifdef CONFIG_BLK_DEV_IO_TRACE
1044 mutex_init(&q->blk_trace_mutex);
1045#endif
483f4afc 1046 mutex_init(&q->sysfs_lock);
e7e72bf6 1047 spin_lock_init(&q->__queue_lock);
483f4afc 1048
498f6650
BVA
1049 if (!q->mq_ops)
1050 q->queue_lock = lock ? : &q->__queue_lock;
c94a96ac 1051
b82d4b19
TH
1052 /*
1053 * A queue starts its life with bypass turned on to avoid
1054 * unnecessary bypass on/off overhead and nasty surprises during
749fefe6
TH
1055 * init. The initial bypass will be finished when the queue is
1056 * registered by blk_register_queue().
b82d4b19
TH
1057 */
1058 q->bypass_depth = 1;
f78bac2c 1059 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
b82d4b19 1060
320ae51f
JA
1061 init_waitqueue_head(&q->mq_freeze_wq);
1062
3ef28e83
DW
1063 /*
1064 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1065 * See blk_register_queue() for details.
1066 */
1067 if (percpu_ref_init(&q->q_usage_counter,
1068 blk_queue_usage_counter_release,
1069 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
fff4996b 1070 goto fail_bdi;
f51b802c 1071
3ef28e83
DW
1072 if (blkcg_init_queue(q))
1073 goto fail_ref;
1074
1da177e4 1075 return q;
a73f730d 1076
3ef28e83
DW
1077fail_ref:
1078 percpu_ref_exit(&q->q_usage_counter);
fff4996b 1079fail_bdi:
a83b576c
JA
1080 blk_free_queue_stats(q->stats);
1081fail_stats:
d03f6cdc 1082 bdi_put(q->backing_dev_info);
54efd50b 1083fail_split:
338aa96d 1084 bioset_exit(&q->bio_split);
a73f730d
TH
1085fail_id:
1086 ida_simple_remove(&blk_queue_ida, q->id);
1087fail_q:
1088 kmem_cache_free(blk_requestq_cachep, q);
1089 return NULL;
1da177e4 1090}
1946089a 1091EXPORT_SYMBOL(blk_alloc_queue_node);
1da177e4
LT
1092
1093/**
1094 * blk_init_queue - prepare a request queue for use with a block device
1095 * @rfn: The function to be called to process requests that have been
1096 * placed on the queue.
1097 * @lock: Request queue spin lock
1098 *
1099 * Description:
1100 * If a block device wishes to use the standard request handling procedures,
1101 * which sorts requests and coalesces adjacent requests, then it must
1102 * call blk_init_queue(). The function @rfn will be called when there
1103 * are requests on the queue that need to be processed. If the device
1104 * supports plugging, then @rfn may not be called immediately when requests
1105 * are available on the queue, but may be called at some time later instead.
1106 * Plugged queues are generally unplugged when a buffer belonging to one
1107 * of the requests on the queue is needed, or due to memory pressure.
1108 *
1109 * @rfn is not required, or even expected, to remove all requests off the
1110 * queue, but only as many as it can handle at a time. If it does leave
1111 * requests on the queue, it is responsible for arranging that the requests
1112 * get dealt with eventually.
1113 *
1114 * The queue spin lock must be held while manipulating the requests on the
a038e253
PBG
1115 * request queue; this lock will be taken also from interrupt context, so irq
1116 * disabling is needed for it.
1da177e4 1117 *
710027a4 1118 * Function returns a pointer to the initialized request queue, or %NULL if
1da177e4
LT
1119 * it didn't succeed.
1120 *
1121 * Note:
1122 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1123 * when the block device is deactivated (such as at module unload).
1124 **/
1946089a 1125
165125e1 1126struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1da177e4 1127{
c304a51b 1128 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1946089a
CL
1129}
1130EXPORT_SYMBOL(blk_init_queue);
1131
165125e1 1132struct request_queue *
1946089a
CL
1133blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1134{
5ea708d1 1135 struct request_queue *q;
1da177e4 1136
498f6650 1137 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
5ea708d1 1138 if (!q)
c86d1b8a
MS
1139 return NULL;
1140
5ea708d1 1141 q->request_fn = rfn;
5ea708d1
CH
1142 if (blk_init_allocated_queue(q) < 0) {
1143 blk_cleanup_queue(q);
1144 return NULL;
1145 }
18741986 1146
7982e90c 1147 return q;
01effb0d
MS
1148}
1149EXPORT_SYMBOL(blk_init_queue_node);
1150
dece1635 1151static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
336b7e1f 1152
1da177e4 1153
5ea708d1
CH
1154int blk_init_allocated_queue(struct request_queue *q)
1155{
332ebbf7
BVA
1156 WARN_ON_ONCE(q->mq_ops);
1157
6d247d7f 1158 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
ba483388 1159 if (!q->fq)
5ea708d1 1160 return -ENOMEM;
7982e90c 1161
6d247d7f
CH
1162 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1163 goto out_free_flush_queue;
7982e90c 1164
a051661c 1165 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
6d247d7f 1166 goto out_exit_flush_rq;
1da177e4 1167
287922eb 1168 INIT_WORK(&q->timeout_work, blk_timeout_work);
60ea8226 1169 q->queue_flags |= QUEUE_FLAG_DEFAULT;
c94a96ac 1170
f3b144aa
JA
1171 /*
1172 * This also sets hw/phys segments, boundary and size
1173 */
c20e8de2 1174 blk_queue_make_request(q, blk_queue_bio);
1da177e4 1175
44ec9542
AS
1176 q->sg_reserved_size = INT_MAX;
1177
acddf3b3 1178 if (elevator_init(q))
6d247d7f 1179 goto out_exit_flush_rq;
5ea708d1 1180 return 0;
eb1c160b 1181
6d247d7f
CH
1182out_exit_flush_rq:
1183 if (q->exit_rq_fn)
1184 q->exit_rq_fn(q, q->fq->flush_rq);
1185out_free_flush_queue:
ba483388 1186 blk_free_flush_queue(q->fq);
5ea708d1 1187 return -ENOMEM;
1da177e4 1188}
5151412d 1189EXPORT_SYMBOL(blk_init_allocated_queue);
1da177e4 1190
09ac46c4 1191bool blk_get_queue(struct request_queue *q)
1da177e4 1192{
3f3299d5 1193 if (likely(!blk_queue_dying(q))) {
09ac46c4
TH
1194 __blk_get_queue(q);
1195 return true;
1da177e4
LT
1196 }
1197
09ac46c4 1198 return false;
1da177e4 1199}
d86e0e83 1200EXPORT_SYMBOL(blk_get_queue);
1da177e4 1201
5b788ce3 1202static inline void blk_free_request(struct request_list *rl, struct request *rq)
1da177e4 1203{
e8064021 1204 if (rq->rq_flags & RQF_ELVPRIV) {
5b788ce3 1205 elv_put_request(rl->q, rq);
f1f8cc94 1206 if (rq->elv.icq)
11a3122f 1207 put_io_context(rq->elv.icq->ioc);
f1f8cc94
TH
1208 }
1209
5b788ce3 1210 mempool_free(rq, rl->rq_pool);
1da177e4
LT
1211}
1212
1da177e4
LT
1213/*
1214 * ioc_batching returns true if the ioc is a valid batching request and
1215 * should be given priority access to a request.
1216 */
165125e1 1217static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1218{
1219 if (!ioc)
1220 return 0;
1221
1222 /*
1223 * Make sure the process is able to allocate at least 1 request
1224 * even if the batch times out, otherwise we could theoretically
1225 * lose wakeups.
1226 */
1227 return ioc->nr_batch_requests == q->nr_batching ||
1228 (ioc->nr_batch_requests > 0
1229 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1230}
1231
1232/*
1233 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1234 * will cause the process to be a "batcher" on all queues in the system. This
1235 * is the behaviour we want though - once it gets a wakeup it should be given
1236 * a nice run.
1237 */
165125e1 1238static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1239{
1240 if (!ioc || ioc_batching(q, ioc))
1241 return;
1242
1243 ioc->nr_batch_requests = q->nr_batching;
1244 ioc->last_waited = jiffies;
1245}
1246
5b788ce3 1247static void __freed_request(struct request_list *rl, int sync)
1da177e4 1248{
5b788ce3 1249 struct request_queue *q = rl->q;
1da177e4 1250
d40f75a0
TH
1251 if (rl->count[sync] < queue_congestion_off_threshold(q))
1252 blk_clear_congested(rl, sync);
1da177e4 1253
1faa16d2
JA
1254 if (rl->count[sync] + 1 <= q->nr_requests) {
1255 if (waitqueue_active(&rl->wait[sync]))
1256 wake_up(&rl->wait[sync]);
1da177e4 1257
5b788ce3 1258 blk_clear_rl_full(rl, sync);
1da177e4
LT
1259 }
1260}
1261
1262/*
1263 * A request has just been released. Account for it, update the full and
1264 * congestion status, wake up any waiters. Called under q->queue_lock.
1265 */
e8064021
CH
1266static void freed_request(struct request_list *rl, bool sync,
1267 req_flags_t rq_flags)
1da177e4 1268{
5b788ce3 1269 struct request_queue *q = rl->q;
1da177e4 1270
8a5ecdd4 1271 q->nr_rqs[sync]--;
1faa16d2 1272 rl->count[sync]--;
e8064021 1273 if (rq_flags & RQF_ELVPRIV)
8a5ecdd4 1274 q->nr_rqs_elvpriv--;
1da177e4 1275
5b788ce3 1276 __freed_request(rl, sync);
1da177e4 1277
1faa16d2 1278 if (unlikely(rl->starved[sync ^ 1]))
5b788ce3 1279 __freed_request(rl, sync ^ 1);
1da177e4
LT
1280}
1281
e3a2b3f9
JA
1282int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1283{
1284 struct request_list *rl;
d40f75a0 1285 int on_thresh, off_thresh;
e3a2b3f9 1286
332ebbf7
BVA
1287 WARN_ON_ONCE(q->mq_ops);
1288
e3a2b3f9
JA
1289 spin_lock_irq(q->queue_lock);
1290 q->nr_requests = nr;
1291 blk_queue_congestion_threshold(q);
d40f75a0
TH
1292 on_thresh = queue_congestion_on_threshold(q);
1293 off_thresh = queue_congestion_off_threshold(q);
e3a2b3f9 1294
d40f75a0
TH
1295 blk_queue_for_each_rl(rl, q) {
1296 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1297 blk_set_congested(rl, BLK_RW_SYNC);
1298 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1299 blk_clear_congested(rl, BLK_RW_SYNC);
e3a2b3f9 1300
d40f75a0
TH
1301 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1302 blk_set_congested(rl, BLK_RW_ASYNC);
1303 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1304 blk_clear_congested(rl, BLK_RW_ASYNC);
e3a2b3f9 1305
e3a2b3f9
JA
1306 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1307 blk_set_rl_full(rl, BLK_RW_SYNC);
1308 } else {
1309 blk_clear_rl_full(rl, BLK_RW_SYNC);
1310 wake_up(&rl->wait[BLK_RW_SYNC]);
1311 }
1312
1313 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1314 blk_set_rl_full(rl, BLK_RW_ASYNC);
1315 } else {
1316 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1317 wake_up(&rl->wait[BLK_RW_ASYNC]);
1318 }
1319 }
1320
1321 spin_unlock_irq(q->queue_lock);
1322 return 0;
1323}
1324
da8303c6 1325/**
a06e05e6 1326 * __get_request - get a free request
5b788ce3 1327 * @rl: request list to allocate from
ef295ecf 1328 * @op: operation and flags
da8303c6 1329 * @bio: bio to allocate request for (can be %NULL)
6a15674d 1330 * @flags: BLQ_MQ_REQ_* flags
4accf5fc 1331 * @gfp_mask: allocator flags
da8303c6
TH
1332 *
1333 * Get a free request from @q. This function may fail under memory
1334 * pressure or if @q is dead.
1335 *
da3dae54 1336 * Must be called with @q->queue_lock held and,
a492f075
JL
1337 * Returns ERR_PTR on failure, with @q->queue_lock held.
1338 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1339 */
ef295ecf 1340static struct request *__get_request(struct request_list *rl, unsigned int op,
4accf5fc 1341 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1da177e4 1342{
5b788ce3 1343 struct request_queue *q = rl->q;
b679281a 1344 struct request *rq;
7f4b35d1
TH
1345 struct elevator_type *et = q->elevator->type;
1346 struct io_context *ioc = rq_ioc(bio);
f1f8cc94 1347 struct io_cq *icq = NULL;
ef295ecf 1348 const bool is_sync = op_is_sync(op);
75eb6c37 1349 int may_queue;
e8064021 1350 req_flags_t rq_flags = RQF_ALLOCED;
88ee5ef1 1351
2fff8a92
BVA
1352 lockdep_assert_held(q->queue_lock);
1353
3f3299d5 1354 if (unlikely(blk_queue_dying(q)))
a492f075 1355 return ERR_PTR(-ENODEV);
da8303c6 1356
ef295ecf 1357 may_queue = elv_may_queue(q, op);
88ee5ef1
JA
1358 if (may_queue == ELV_MQUEUE_NO)
1359 goto rq_starved;
1360
1faa16d2
JA
1361 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1362 if (rl->count[is_sync]+1 >= q->nr_requests) {
88ee5ef1
JA
1363 /*
1364 * The queue will fill after this allocation, so set
1365 * it as full, and mark this process as "batching".
1366 * This process will be allowed to complete a batch of
1367 * requests, others will be blocked.
1368 */
5b788ce3 1369 if (!blk_rl_full(rl, is_sync)) {
88ee5ef1 1370 ioc_set_batching(q, ioc);
5b788ce3 1371 blk_set_rl_full(rl, is_sync);
88ee5ef1
JA
1372 } else {
1373 if (may_queue != ELV_MQUEUE_MUST
1374 && !ioc_batching(q, ioc)) {
1375 /*
1376 * The queue is full and the allocating
1377 * process is not a "batcher", and not
1378 * exempted by the IO scheduler
1379 */
a492f075 1380 return ERR_PTR(-ENOMEM);
88ee5ef1
JA
1381 }
1382 }
1da177e4 1383 }
d40f75a0 1384 blk_set_congested(rl, is_sync);
1da177e4
LT
1385 }
1386
082cf69e
JA
1387 /*
1388 * Only allow batching queuers to allocate up to 50% over the defined
1389 * limit of requests, otherwise we could have thousands of requests
1390 * allocated with any setting of ->nr_requests
1391 */
1faa16d2 1392 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
a492f075 1393 return ERR_PTR(-ENOMEM);
fd782a4a 1394
8a5ecdd4 1395 q->nr_rqs[is_sync]++;
1faa16d2
JA
1396 rl->count[is_sync]++;
1397 rl->starved[is_sync] = 0;
cb98fc8b 1398
f1f8cc94
TH
1399 /*
1400 * Decide whether the new request will be managed by elevator. If
e8064021 1401 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
f1f8cc94
TH
1402 * prevent the current elevator from being destroyed until the new
1403 * request is freed. This guarantees icq's won't be destroyed and
1404 * makes creating new ones safe.
1405 *
e6f7f93d
CH
1406 * Flush requests do not use the elevator so skip initialization.
1407 * This allows a request to share the flush and elevator data.
1408 *
f1f8cc94
TH
1409 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1410 * it will be created after releasing queue_lock.
1411 */
e6f7f93d 1412 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
e8064021 1413 rq_flags |= RQF_ELVPRIV;
8a5ecdd4 1414 q->nr_rqs_elvpriv++;
f1f8cc94
TH
1415 if (et->icq_cache && ioc)
1416 icq = ioc_lookup_icq(ioc, q);
9d5a4e94 1417 }
cb98fc8b 1418
f253b86b 1419 if (blk_queue_io_stat(q))
e8064021 1420 rq_flags |= RQF_IO_STAT;
1da177e4
LT
1421 spin_unlock_irq(q->queue_lock);
1422
29e2b09a 1423 /* allocate and init request */
5b788ce3 1424 rq = mempool_alloc(rl->rq_pool, gfp_mask);
29e2b09a 1425 if (!rq)
b679281a 1426 goto fail_alloc;
1da177e4 1427
29e2b09a 1428 blk_rq_init(q, rq);
a051661c 1429 blk_rq_set_rl(rq, rl);
ef295ecf 1430 rq->cmd_flags = op;
e8064021 1431 rq->rq_flags = rq_flags;
1b6d65a0
BVA
1432 if (flags & BLK_MQ_REQ_PREEMPT)
1433 rq->rq_flags |= RQF_PREEMPT;
29e2b09a 1434
aaf7c680 1435 /* init elvpriv */
e8064021 1436 if (rq_flags & RQF_ELVPRIV) {
aaf7c680 1437 if (unlikely(et->icq_cache && !icq)) {
7f4b35d1
TH
1438 if (ioc)
1439 icq = ioc_create_icq(ioc, q, gfp_mask);
aaf7c680
TH
1440 if (!icq)
1441 goto fail_elvpriv;
29e2b09a 1442 }
aaf7c680
TH
1443
1444 rq->elv.icq = icq;
1445 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1446 goto fail_elvpriv;
1447
1448 /* @rq->elv.icq holds io_context until @rq is freed */
29e2b09a
TH
1449 if (icq)
1450 get_io_context(icq->ioc);
1451 }
aaf7c680 1452out:
88ee5ef1
JA
1453 /*
1454 * ioc may be NULL here, and ioc_batching will be false. That's
1455 * OK, if the queue is under the request limit then requests need
1456 * not count toward the nr_batch_requests limit. There will always
1457 * be some limit enforced by BLK_BATCH_TIME.
1458 */
1da177e4
LT
1459 if (ioc_batching(q, ioc))
1460 ioc->nr_batch_requests--;
6728cb0e 1461
e6a40b09 1462 trace_block_getrq(q, bio, op);
1da177e4 1463 return rq;
b679281a 1464
aaf7c680
TH
1465fail_elvpriv:
1466 /*
1467 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1468 * and may fail indefinitely under memory pressure and thus
1469 * shouldn't stall IO. Treat this request as !elvpriv. This will
1470 * disturb iosched and blkcg but weird is bettern than dead.
1471 */
7b2b10e0 1472 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
dc3b17cc 1473 __func__, dev_name(q->backing_dev_info->dev));
aaf7c680 1474
e8064021 1475 rq->rq_flags &= ~RQF_ELVPRIV;
aaf7c680
TH
1476 rq->elv.icq = NULL;
1477
1478 spin_lock_irq(q->queue_lock);
8a5ecdd4 1479 q->nr_rqs_elvpriv--;
aaf7c680
TH
1480 spin_unlock_irq(q->queue_lock);
1481 goto out;
1482
b679281a
TH
1483fail_alloc:
1484 /*
1485 * Allocation failed presumably due to memory. Undo anything we
1486 * might have messed up.
1487 *
1488 * Allocating task should really be put onto the front of the wait
1489 * queue, but this is pretty rare.
1490 */
1491 spin_lock_irq(q->queue_lock);
e8064021 1492 freed_request(rl, is_sync, rq_flags);
b679281a
TH
1493
1494 /*
1495 * in the very unlikely event that allocation failed and no
1496 * requests for this direction was pending, mark us starved so that
1497 * freeing of a request in the other direction will notice
1498 * us. another possible fix would be to split the rq mempool into
1499 * READ and WRITE
1500 */
1501rq_starved:
1502 if (unlikely(rl->count[is_sync] == 0))
1503 rl->starved[is_sync] = 1;
a492f075 1504 return ERR_PTR(-ENOMEM);
1da177e4
LT
1505}
1506
da8303c6 1507/**
a06e05e6 1508 * get_request - get a free request
da8303c6 1509 * @q: request_queue to allocate request from
ef295ecf 1510 * @op: operation and flags
da8303c6 1511 * @bio: bio to allocate request for (can be %NULL)
6a15674d 1512 * @flags: BLK_MQ_REQ_* flags.
4accf5fc 1513 * @gfp: allocator flags
da8303c6 1514 *
a9a14d36 1515 * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags,
d0164adc 1516 * this function keeps retrying under memory pressure and fails iff @q is dead.
d6344532 1517 *
da3dae54 1518 * Must be called with @q->queue_lock held and,
a492f075
JL
1519 * Returns ERR_PTR on failure, with @q->queue_lock held.
1520 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1521 */
ef295ecf 1522static struct request *get_request(struct request_queue *q, unsigned int op,
4accf5fc 1523 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1da177e4 1524{
ef295ecf 1525 const bool is_sync = op_is_sync(op);
a06e05e6 1526 DEFINE_WAIT(wait);
a051661c 1527 struct request_list *rl;
1da177e4 1528 struct request *rq;
a051661c 1529
2fff8a92 1530 lockdep_assert_held(q->queue_lock);
332ebbf7 1531 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1532
a051661c 1533 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
a06e05e6 1534retry:
4accf5fc 1535 rq = __get_request(rl, op, bio, flags, gfp);
a492f075 1536 if (!IS_ERR(rq))
a06e05e6 1537 return rq;
1da177e4 1538
03a07c92
GR
1539 if (op & REQ_NOWAIT) {
1540 blk_put_rl(rl);
1541 return ERR_PTR(-EAGAIN);
1542 }
1543
6a15674d 1544 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
a051661c 1545 blk_put_rl(rl);
a492f075 1546 return rq;
a051661c 1547 }
1da177e4 1548
a06e05e6
TH
1549 /* wait on @rl and retry */
1550 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1551 TASK_UNINTERRUPTIBLE);
1da177e4 1552
e6a40b09 1553 trace_block_sleeprq(q, bio, op);
1da177e4 1554
a06e05e6
TH
1555 spin_unlock_irq(q->queue_lock);
1556 io_schedule();
d6344532 1557
a06e05e6
TH
1558 /*
1559 * After sleeping, we become a "batching" process and will be able
1560 * to allocate at least one request, and up to a big batch of them
1561 * for a small period time. See ioc_batching, ioc_set_batching
1562 */
a06e05e6 1563 ioc_set_batching(q, current->io_context);
05caf8db 1564
a06e05e6
TH
1565 spin_lock_irq(q->queue_lock);
1566 finish_wait(&rl->wait[is_sync], &wait);
1da177e4 1567
a06e05e6 1568 goto retry;
1da177e4
LT
1569}
1570
6a15674d 1571/* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
cd6ce148 1572static struct request *blk_old_get_request(struct request_queue *q,
9a95e4ef 1573 unsigned int op, blk_mq_req_flags_t flags)
1da177e4
LT
1574{
1575 struct request *rq;
c3036021 1576 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
055f6e18 1577 int ret = 0;
1da177e4 1578
332ebbf7
BVA
1579 WARN_ON_ONCE(q->mq_ops);
1580
7f4b35d1
TH
1581 /* create ioc upfront */
1582 create_io_context(gfp_mask, q->node);
1583
3a0a5299 1584 ret = blk_queue_enter(q, flags);
055f6e18
ML
1585 if (ret)
1586 return ERR_PTR(ret);
d6344532 1587 spin_lock_irq(q->queue_lock);
4accf5fc 1588 rq = get_request(q, op, NULL, flags, gfp_mask);
0c4de0f3 1589 if (IS_ERR(rq)) {
da8303c6 1590 spin_unlock_irq(q->queue_lock);
055f6e18 1591 blk_queue_exit(q);
0c4de0f3
CH
1592 return rq;
1593 }
1da177e4 1594
0c4de0f3
CH
1595 /* q->queue_lock is unlocked at this point */
1596 rq->__data_len = 0;
1597 rq->__sector = (sector_t) -1;
1598 rq->bio = rq->biotail = NULL;
1da177e4
LT
1599 return rq;
1600}
320ae51f 1601
6a15674d 1602/**
ff005a06 1603 * blk_get_request - allocate a request
6a15674d
BVA
1604 * @q: request queue to allocate a request for
1605 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1606 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1607 */
ff005a06
CH
1608struct request *blk_get_request(struct request_queue *q, unsigned int op,
1609 blk_mq_req_flags_t flags)
320ae51f 1610{
d280bab3
BVA
1611 struct request *req;
1612
6a15674d 1613 WARN_ON_ONCE(op & REQ_NOWAIT);
1b6d65a0 1614 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
6a15674d 1615
d280bab3 1616 if (q->mq_ops) {
6a15674d 1617 req = blk_mq_alloc_request(q, op, flags);
d280bab3
BVA
1618 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1619 q->mq_ops->initialize_rq_fn(req);
1620 } else {
6a15674d 1621 req = blk_old_get_request(q, op, flags);
d280bab3
BVA
1622 if (!IS_ERR(req) && q->initialize_rq_fn)
1623 q->initialize_rq_fn(req);
1624 }
1625
1626 return req;
320ae51f 1627}
1da177e4
LT
1628EXPORT_SYMBOL(blk_get_request);
1629
1630/**
1631 * blk_requeue_request - put a request back on queue
1632 * @q: request queue where request should be inserted
1633 * @rq: request to be inserted
1634 *
1635 * Description:
1636 * Drivers often keep queueing requests until the hardware cannot accept
1637 * more, when that condition happens we need to put the request back
1638 * on the queue. Must be called with queue lock held.
1639 */
165125e1 1640void blk_requeue_request(struct request_queue *q, struct request *rq)
1da177e4 1641{
2fff8a92 1642 lockdep_assert_held(q->queue_lock);
332ebbf7 1643 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1644
242f9dcb
JA
1645 blk_delete_timer(rq);
1646 blk_clear_rq_complete(rq);
5f3ea37c 1647 trace_block_rq_requeue(q, rq);
a8a45941 1648 wbt_requeue(q->rq_wb, rq);
2056a782 1649
e8064021 1650 if (rq->rq_flags & RQF_QUEUED)
1da177e4
LT
1651 blk_queue_end_tag(q, rq);
1652
ba396a6c
JB
1653 BUG_ON(blk_queued_rq(rq));
1654
1da177e4
LT
1655 elv_requeue_request(q, rq);
1656}
1da177e4
LT
1657EXPORT_SYMBOL(blk_requeue_request);
1658
73c10101
JA
1659static void add_acct_request(struct request_queue *q, struct request *rq,
1660 int where)
1661{
320ae51f 1662 blk_account_io_start(rq, true);
7eaceacc 1663 __elv_add_request(q, rq, where);
73c10101
JA
1664}
1665
d62e26b3 1666static void part_round_stats_single(struct request_queue *q, int cpu,
b8d62b3a
JA
1667 struct hd_struct *part, unsigned long now,
1668 unsigned int inflight)
074a7aca 1669{
b8d62b3a 1670 if (inflight) {
074a7aca 1671 __part_stat_add(cpu, part, time_in_queue,
b8d62b3a 1672 inflight * (now - part->stamp));
074a7aca
TH
1673 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1674 }
1675 part->stamp = now;
1676}
1677
1678/**
496aa8a9 1679 * part_round_stats() - Round off the performance stats on a struct disk_stats.
d62e26b3 1680 * @q: target block queue
496aa8a9
RD
1681 * @cpu: cpu number for stats access
1682 * @part: target partition
1da177e4
LT
1683 *
1684 * The average IO queue length and utilisation statistics are maintained
1685 * by observing the current state of the queue length and the amount of
1686 * time it has been in this state for.
1687 *
1688 * Normally, that accounting is done on IO completion, but that can result
1689 * in more than a second's worth of IO being accounted for within any one
1690 * second, leading to >100% utilisation. To deal with that, we call this
1691 * function to do a round-off before returning the results when reading
1692 * /proc/diskstats. This accounts immediately for all queue usage up to
1693 * the current jiffies and restarts the counters again.
1694 */
d62e26b3 1695void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
6f2576af 1696{
b8d62b3a 1697 struct hd_struct *part2 = NULL;
6f2576af 1698 unsigned long now = jiffies;
b8d62b3a
JA
1699 unsigned int inflight[2];
1700 int stats = 0;
1701
1702 if (part->stamp != now)
1703 stats |= 1;
1704
1705 if (part->partno) {
1706 part2 = &part_to_disk(part)->part0;
1707 if (part2->stamp != now)
1708 stats |= 2;
1709 }
1710
1711 if (!stats)
1712 return;
1713
1714 part_in_flight(q, part, inflight);
6f2576af 1715
b8d62b3a
JA
1716 if (stats & 2)
1717 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1718 if (stats & 1)
1719 part_round_stats_single(q, cpu, part, now, inflight[0]);
6f2576af 1720}
074a7aca 1721EXPORT_SYMBOL_GPL(part_round_stats);
6f2576af 1722
47fafbc7 1723#ifdef CONFIG_PM
c8158819
LM
1724static void blk_pm_put_request(struct request *rq)
1725{
e8064021 1726 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
c8158819
LM
1727 pm_runtime_mark_last_busy(rq->q->dev);
1728}
1729#else
1730static inline void blk_pm_put_request(struct request *rq) {}
1731#endif
1732
165125e1 1733void __blk_put_request(struct request_queue *q, struct request *req)
1da177e4 1734{
e8064021
CH
1735 req_flags_t rq_flags = req->rq_flags;
1736
1da177e4
LT
1737 if (unlikely(!q))
1738 return;
1da177e4 1739
6f5ba581
CH
1740 if (q->mq_ops) {
1741 blk_mq_free_request(req);
1742 return;
1743 }
1744
2fff8a92
BVA
1745 lockdep_assert_held(q->queue_lock);
1746
6cc77e9c 1747 blk_req_zone_write_unlock(req);
c8158819
LM
1748 blk_pm_put_request(req);
1749
8922e16c
TH
1750 elv_completed_request(q, req);
1751
1cd96c24
BH
1752 /* this is a bio leak */
1753 WARN_ON(req->bio != NULL);
1754
a8a45941 1755 wbt_done(q->rq_wb, req);
87760e5e 1756
1da177e4
LT
1757 /*
1758 * Request may not have originated from ll_rw_blk. if not,
1759 * it didn't come out of our reserved rq pools
1760 */
e8064021 1761 if (rq_flags & RQF_ALLOCED) {
a051661c 1762 struct request_list *rl = blk_rq_rl(req);
ef295ecf 1763 bool sync = op_is_sync(req->cmd_flags);
1da177e4 1764
1da177e4 1765 BUG_ON(!list_empty(&req->queuelist));
360f92c2 1766 BUG_ON(ELV_ON_HASH(req));
1da177e4 1767
a051661c 1768 blk_free_request(rl, req);
e8064021 1769 freed_request(rl, sync, rq_flags);
a051661c 1770 blk_put_rl(rl);
055f6e18 1771 blk_queue_exit(q);
1da177e4
LT
1772 }
1773}
6e39b69e
MC
1774EXPORT_SYMBOL_GPL(__blk_put_request);
1775
1da177e4
LT
1776void blk_put_request(struct request *req)
1777{
165125e1 1778 struct request_queue *q = req->q;
8922e16c 1779
320ae51f
JA
1780 if (q->mq_ops)
1781 blk_mq_free_request(req);
1782 else {
1783 unsigned long flags;
1784
1785 spin_lock_irqsave(q->queue_lock, flags);
1786 __blk_put_request(q, req);
1787 spin_unlock_irqrestore(q->queue_lock, flags);
1788 }
1da177e4 1789}
1da177e4
LT
1790EXPORT_SYMBOL(blk_put_request);
1791
320ae51f
JA
1792bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1793 struct bio *bio)
73c10101 1794{
1eff9d32 1795 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1796
73c10101
JA
1797 if (!ll_back_merge_fn(q, req, bio))
1798 return false;
1799
8c1cf6bb 1800 trace_block_bio_backmerge(q, req, bio);
73c10101
JA
1801
1802 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1803 blk_rq_set_mixed_merge(req);
1804
1805 req->biotail->bi_next = bio;
1806 req->biotail = bio;
4f024f37 1807 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1808 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1809
320ae51f 1810 blk_account_io_start(req, false);
73c10101
JA
1811 return true;
1812}
1813
320ae51f
JA
1814bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1815 struct bio *bio)
73c10101 1816{
1eff9d32 1817 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1818
73c10101
JA
1819 if (!ll_front_merge_fn(q, req, bio))
1820 return false;
1821
8c1cf6bb 1822 trace_block_bio_frontmerge(q, req, bio);
73c10101
JA
1823
1824 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1825 blk_rq_set_mixed_merge(req);
1826
73c10101
JA
1827 bio->bi_next = req->bio;
1828 req->bio = bio;
1829
4f024f37
KO
1830 req->__sector = bio->bi_iter.bi_sector;
1831 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1832 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1833
320ae51f 1834 blk_account_io_start(req, false);
73c10101
JA
1835 return true;
1836}
1837
1e739730
CH
1838bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1839 struct bio *bio)
1840{
1841 unsigned short segments = blk_rq_nr_discard_segments(req);
1842
1843 if (segments >= queue_max_discard_segments(q))
1844 goto no_merge;
1845 if (blk_rq_sectors(req) + bio_sectors(bio) >
1846 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1847 goto no_merge;
1848
1849 req->biotail->bi_next = bio;
1850 req->biotail = bio;
1851 req->__data_len += bio->bi_iter.bi_size;
1852 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1853 req->nr_phys_segments = segments + 1;
1854
1855 blk_account_io_start(req, false);
1856 return true;
1857no_merge:
1858 req_set_nomerge(q, req);
1859 return false;
1860}
1861
bd87b589 1862/**
320ae51f 1863 * blk_attempt_plug_merge - try to merge with %current's plugged list
bd87b589
TH
1864 * @q: request_queue new bio is being queued at
1865 * @bio: new bio being queued
1866 * @request_count: out parameter for number of traversed plugged requests
ccc2600b
RD
1867 * @same_queue_rq: pointer to &struct request that gets filled in when
1868 * another request associated with @q is found on the plug list
1869 * (optional, may be %NULL)
bd87b589
TH
1870 *
1871 * Determine whether @bio being queued on @q can be merged with a request
1872 * on %current's plugged list. Returns %true if merge was successful,
1873 * otherwise %false.
1874 *
07c2bd37
TH
1875 * Plugging coalesces IOs from the same issuer for the same purpose without
1876 * going through @q->queue_lock. As such it's more of an issuing mechanism
1877 * than scheduling, and the request, while may have elvpriv data, is not
1878 * added on the elevator at this point. In addition, we don't have
1879 * reliable access to the elevator outside queue lock. Only check basic
1880 * merging parameters without querying the elevator.
da41a589
RE
1881 *
1882 * Caller must ensure !blk_queue_nomerges(q) beforehand.
73c10101 1883 */
320ae51f 1884bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
5b3f341f
SL
1885 unsigned int *request_count,
1886 struct request **same_queue_rq)
73c10101
JA
1887{
1888 struct blk_plug *plug;
1889 struct request *rq;
92f399c7 1890 struct list_head *plug_list;
73c10101 1891
bd87b589 1892 plug = current->plug;
73c10101 1893 if (!plug)
34fe7c05 1894 return false;
56ebdaf2 1895 *request_count = 0;
73c10101 1896
92f399c7
SL
1897 if (q->mq_ops)
1898 plug_list = &plug->mq_list;
1899 else
1900 plug_list = &plug->list;
1901
1902 list_for_each_entry_reverse(rq, plug_list, queuelist) {
34fe7c05 1903 bool merged = false;
73c10101 1904
5b3f341f 1905 if (rq->q == q) {
1b2e19f1 1906 (*request_count)++;
5b3f341f
SL
1907 /*
1908 * Only blk-mq multiple hardware queues case checks the
1909 * rq in the same queue, there should be only one such
1910 * rq in a queue
1911 **/
1912 if (same_queue_rq)
1913 *same_queue_rq = rq;
1914 }
56ebdaf2 1915
07c2bd37 1916 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
73c10101
JA
1917 continue;
1918
34fe7c05
CH
1919 switch (blk_try_merge(rq, bio)) {
1920 case ELEVATOR_BACK_MERGE:
1921 merged = bio_attempt_back_merge(q, rq, bio);
1922 break;
1923 case ELEVATOR_FRONT_MERGE:
1924 merged = bio_attempt_front_merge(q, rq, bio);
1925 break;
1e739730
CH
1926 case ELEVATOR_DISCARD_MERGE:
1927 merged = bio_attempt_discard_merge(q, rq, bio);
1928 break;
34fe7c05
CH
1929 default:
1930 break;
73c10101 1931 }
34fe7c05
CH
1932
1933 if (merged)
1934 return true;
73c10101 1935 }
34fe7c05
CH
1936
1937 return false;
73c10101
JA
1938}
1939
0809e3ac
JM
1940unsigned int blk_plug_queued_count(struct request_queue *q)
1941{
1942 struct blk_plug *plug;
1943 struct request *rq;
1944 struct list_head *plug_list;
1945 unsigned int ret = 0;
1946
1947 plug = current->plug;
1948 if (!plug)
1949 goto out;
1950
1951 if (q->mq_ops)
1952 plug_list = &plug->mq_list;
1953 else
1954 plug_list = &plug->list;
1955
1956 list_for_each_entry(rq, plug_list, queuelist) {
1957 if (rq->q == q)
1958 ret++;
1959 }
1960out:
1961 return ret;
1962}
1963
da8d7f07 1964void blk_init_request_from_bio(struct request *req, struct bio *bio)
52d9e675 1965{
0be0dee6
BVA
1966 struct io_context *ioc = rq_ioc(bio);
1967
1eff9d32 1968 if (bio->bi_opf & REQ_RAHEAD)
a82afdfc 1969 req->cmd_flags |= REQ_FAILFAST_MASK;
b31dc66a 1970
4f024f37 1971 req->__sector = bio->bi_iter.bi_sector;
5dc8b362
AM
1972 if (ioprio_valid(bio_prio(bio)))
1973 req->ioprio = bio_prio(bio);
0be0dee6
BVA
1974 else if (ioc)
1975 req->ioprio = ioc->ioprio;
1976 else
1977 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
cb6934f8 1978 req->write_hint = bio->bi_write_hint;
bc1c56fd 1979 blk_rq_bio_prep(req->q, req, bio);
52d9e675 1980}
da8d7f07 1981EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
52d9e675 1982
dece1635 1983static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1da177e4 1984{
73c10101 1985 struct blk_plug *plug;
34fe7c05 1986 int where = ELEVATOR_INSERT_SORT;
e4d750c9 1987 struct request *req, *free;
56ebdaf2 1988 unsigned int request_count = 0;
87760e5e 1989 unsigned int wb_acct;
1da177e4 1990
1da177e4
LT
1991 /*
1992 * low level driver can indicate that it wants pages above a
1993 * certain limit bounced to low memory (ie for highmem, or even
1994 * ISA dma in theory)
1995 */
1996 blk_queue_bounce(q, &bio);
1997
af67c31f 1998 blk_queue_split(q, &bio);
23688bf4 1999
e23947bd 2000 if (!bio_integrity_prep(bio))
dece1635 2001 return BLK_QC_T_NONE;
ffecfd1a 2002
f73f44eb 2003 if (op_is_flush(bio->bi_opf)) {
73c10101 2004 spin_lock_irq(q->queue_lock);
ae1b1539 2005 where = ELEVATOR_INSERT_FLUSH;
28e7d184
TH
2006 goto get_rq;
2007 }
2008
73c10101
JA
2009 /*
2010 * Check if we can merge with the plugged list before grabbing
2011 * any locks.
2012 */
0809e3ac
JM
2013 if (!blk_queue_nomerges(q)) {
2014 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
dece1635 2015 return BLK_QC_T_NONE;
0809e3ac
JM
2016 } else
2017 request_count = blk_plug_queued_count(q);
1da177e4 2018
73c10101 2019 spin_lock_irq(q->queue_lock);
2056a782 2020
34fe7c05
CH
2021 switch (elv_merge(q, &req, bio)) {
2022 case ELEVATOR_BACK_MERGE:
2023 if (!bio_attempt_back_merge(q, req, bio))
2024 break;
2025 elv_bio_merged(q, req, bio);
2026 free = attempt_back_merge(q, req);
2027 if (free)
2028 __blk_put_request(q, free);
2029 else
2030 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2031 goto out_unlock;
2032 case ELEVATOR_FRONT_MERGE:
2033 if (!bio_attempt_front_merge(q, req, bio))
2034 break;
2035 elv_bio_merged(q, req, bio);
2036 free = attempt_front_merge(q, req);
2037 if (free)
2038 __blk_put_request(q, free);
2039 else
2040 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2041 goto out_unlock;
2042 default:
2043 break;
1da177e4
LT
2044 }
2045
450991bc 2046get_rq:
87760e5e
JA
2047 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
2048
1da177e4 2049 /*
450991bc 2050 * Grab a free request. This is might sleep but can not fail.
d6344532 2051 * Returns with the queue unlocked.
450991bc 2052 */
055f6e18 2053 blk_queue_enter_live(q);
c3036021 2054 req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO);
a492f075 2055 if (IS_ERR(req)) {
055f6e18 2056 blk_queue_exit(q);
87760e5e 2057 __wbt_done(q->rq_wb, wb_acct);
4e4cbee9
CH
2058 if (PTR_ERR(req) == -ENOMEM)
2059 bio->bi_status = BLK_STS_RESOURCE;
2060 else
2061 bio->bi_status = BLK_STS_IOERR;
4246a0b6 2062 bio_endio(bio);
da8303c6
TH
2063 goto out_unlock;
2064 }
d6344532 2065
a8a45941 2066 wbt_track(req, wb_acct);
87760e5e 2067
450991bc
NP
2068 /*
2069 * After dropping the lock and possibly sleeping here, our request
2070 * may now be mergeable after it had proven unmergeable (above).
2071 * We don't worry about that case for efficiency. It won't happen
2072 * often, and the elevators are able to handle it.
1da177e4 2073 */
da8d7f07 2074 blk_init_request_from_bio(req, bio);
1da177e4 2075
9562ad9a 2076 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
11ccf116 2077 req->cpu = raw_smp_processor_id();
73c10101
JA
2078
2079 plug = current->plug;
721a9602 2080 if (plug) {
dc6d36c9
JA
2081 /*
2082 * If this is the first request added after a plug, fire
7aef2e78 2083 * of a plug trace.
0a6219a9
ML
2084 *
2085 * @request_count may become stale because of schedule
2086 * out, so check plug list again.
dc6d36c9 2087 */
0a6219a9 2088 if (!request_count || list_empty(&plug->list))
dc6d36c9 2089 trace_block_plug(q);
3540d5e8 2090 else {
50d24c34
SL
2091 struct request *last = list_entry_rq(plug->list.prev);
2092 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2093 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
3540d5e8 2094 blk_flush_plug_list(plug, false);
019ceb7d
SL
2095 trace_block_plug(q);
2096 }
73c10101 2097 }
73c10101 2098 list_add_tail(&req->queuelist, &plug->list);
320ae51f 2099 blk_account_io_start(req, true);
73c10101
JA
2100 } else {
2101 spin_lock_irq(q->queue_lock);
2102 add_acct_request(q, req, where);
24ecfbe2 2103 __blk_run_queue(q);
73c10101
JA
2104out_unlock:
2105 spin_unlock_irq(q->queue_lock);
2106 }
dece1635
JA
2107
2108 return BLK_QC_T_NONE;
1da177e4
LT
2109}
2110
52c5e62d 2111static void handle_bad_sector(struct bio *bio, sector_t maxsector)
1da177e4
LT
2112{
2113 char b[BDEVNAME_SIZE];
2114
2115 printk(KERN_INFO "attempt to access beyond end of device\n");
6296b960 2116 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
74d46992 2117 bio_devname(bio, b), bio->bi_opf,
f73a1c7d 2118 (unsigned long long)bio_end_sector(bio),
52c5e62d 2119 (long long)maxsector);
1da177e4
LT
2120}
2121
c17bb495
AM
2122#ifdef CONFIG_FAIL_MAKE_REQUEST
2123
2124static DECLARE_FAULT_ATTR(fail_make_request);
2125
2126static int __init setup_fail_make_request(char *str)
2127{
2128 return setup_fault_attr(&fail_make_request, str);
2129}
2130__setup("fail_make_request=", setup_fail_make_request);
2131
b2c9cd37 2132static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
c17bb495 2133{
b2c9cd37 2134 return part->make_it_fail && should_fail(&fail_make_request, bytes);
c17bb495
AM
2135}
2136
2137static int __init fail_make_request_debugfs(void)
2138{
dd48c085
AM
2139 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2140 NULL, &fail_make_request);
2141
21f9fcd8 2142 return PTR_ERR_OR_ZERO(dir);
c17bb495
AM
2143}
2144
2145late_initcall(fail_make_request_debugfs);
2146
2147#else /* CONFIG_FAIL_MAKE_REQUEST */
2148
b2c9cd37
AM
2149static inline bool should_fail_request(struct hd_struct *part,
2150 unsigned int bytes)
c17bb495 2151{
b2c9cd37 2152 return false;
c17bb495
AM
2153}
2154
2155#endif /* CONFIG_FAIL_MAKE_REQUEST */
2156
721c7fc7
ID
2157static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2158{
2159 if (part->policy && op_is_write(bio_op(bio))) {
2160 char b[BDEVNAME_SIZE];
2161
2162 printk(KERN_ERR
2163 "generic_make_request: Trying to write "
2164 "to read-only block-device %s (partno %d)\n",
2165 bio_devname(bio, b), part->partno);
2166 return true;
2167 }
2168
2169 return false;
2170}
2171
30abb3a6
HM
2172static noinline int should_fail_bio(struct bio *bio)
2173{
2174 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2175 return -EIO;
2176 return 0;
2177}
2178ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2179
52c5e62d
CH
2180/*
2181 * Check whether this bio extends beyond the end of the device or partition.
2182 * This may well happen - the kernel calls bread() without checking the size of
2183 * the device, e.g., when mounting a file system.
2184 */
2185static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2186{
2187 unsigned int nr_sectors = bio_sectors(bio);
2188
2189 if (nr_sectors && maxsector &&
2190 (nr_sectors > maxsector ||
2191 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2192 handle_bad_sector(bio, maxsector);
2193 return -EIO;
2194 }
2195 return 0;
2196}
2197
74d46992
CH
2198/*
2199 * Remap block n of partition p to block n+start(p) of the disk.
2200 */
2201static inline int blk_partition_remap(struct bio *bio)
2202{
2203 struct hd_struct *p;
52c5e62d 2204 int ret = -EIO;
74d46992 2205
721c7fc7
ID
2206 rcu_read_lock();
2207 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
52c5e62d
CH
2208 if (unlikely(!p))
2209 goto out;
2210 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2211 goto out;
2212 if (unlikely(bio_check_ro(bio, p)))
721c7fc7 2213 goto out;
721c7fc7 2214
74d46992
CH
2215 /*
2216 * Zone reset does not include bi_size so bio_sectors() is always 0.
2217 * Include a test for the reset op code and perform the remap if needed.
2218 */
52c5e62d
CH
2219 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2220 if (bio_check_eod(bio, part_nr_sects_read(p)))
2221 goto out;
2222 bio->bi_iter.bi_sector += p->start_sect;
52c5e62d
CH
2223 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2224 bio->bi_iter.bi_sector - p->start_sect);
2225 }
c04fa44b 2226 bio->bi_partno = 0;
52c5e62d 2227 ret = 0;
721c7fc7
ID
2228out:
2229 rcu_read_unlock();
74d46992
CH
2230 return ret;
2231}
2232
27a84d54
CH
2233static noinline_for_stack bool
2234generic_make_request_checks(struct bio *bio)
1da177e4 2235{
165125e1 2236 struct request_queue *q;
5a7bbad2 2237 int nr_sectors = bio_sectors(bio);
4e4cbee9 2238 blk_status_t status = BLK_STS_IOERR;
5a7bbad2 2239 char b[BDEVNAME_SIZE];
1da177e4
LT
2240
2241 might_sleep();
1da177e4 2242
74d46992 2243 q = bio->bi_disk->queue;
5a7bbad2
CH
2244 if (unlikely(!q)) {
2245 printk(KERN_ERR
2246 "generic_make_request: Trying to access "
2247 "nonexistent block-device %s (%Lu)\n",
74d46992 2248 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
5a7bbad2
CH
2249 goto end_io;
2250 }
c17bb495 2251
03a07c92
GR
2252 /*
2253 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2254 * if queue is not a request based queue.
2255 */
03a07c92
GR
2256 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2257 goto not_supported;
2258
30abb3a6 2259 if (should_fail_bio(bio))
5a7bbad2 2260 goto end_io;
2056a782 2261
52c5e62d
CH
2262 if (bio->bi_partno) {
2263 if (unlikely(blk_partition_remap(bio)))
721c7fc7
ID
2264 goto end_io;
2265 } else {
52c5e62d
CH
2266 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2267 goto end_io;
2268 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
721c7fc7
ID
2269 goto end_io;
2270 }
2056a782 2271
5a7bbad2
CH
2272 /*
2273 * Filter flush bio's early so that make_request based
2274 * drivers without flush support don't have to worry
2275 * about them.
2276 */
f3a8ab7d 2277 if (op_is_flush(bio->bi_opf) &&
c888a8f9 2278 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1eff9d32 2279 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
5a7bbad2 2280 if (!nr_sectors) {
4e4cbee9 2281 status = BLK_STS_OK;
51fd77bd
JA
2282 goto end_io;
2283 }
5a7bbad2 2284 }
5ddfe969 2285
288dab8a
CH
2286 switch (bio_op(bio)) {
2287 case REQ_OP_DISCARD:
2288 if (!blk_queue_discard(q))
2289 goto not_supported;
2290 break;
2291 case REQ_OP_SECURE_ERASE:
2292 if (!blk_queue_secure_erase(q))
2293 goto not_supported;
2294 break;
2295 case REQ_OP_WRITE_SAME:
74d46992 2296 if (!q->limits.max_write_same_sectors)
288dab8a 2297 goto not_supported;
58886785 2298 break;
2d253440
ST
2299 case REQ_OP_ZONE_REPORT:
2300 case REQ_OP_ZONE_RESET:
74d46992 2301 if (!blk_queue_is_zoned(q))
2d253440 2302 goto not_supported;
288dab8a 2303 break;
a6f0788e 2304 case REQ_OP_WRITE_ZEROES:
74d46992 2305 if (!q->limits.max_write_zeroes_sectors)
a6f0788e
CK
2306 goto not_supported;
2307 break;
288dab8a
CH
2308 default:
2309 break;
5a7bbad2 2310 }
01edede4 2311
7f4b35d1
TH
2312 /*
2313 * Various block parts want %current->io_context and lazy ioc
2314 * allocation ends up trading a lot of pain for a small amount of
2315 * memory. Just allocate it upfront. This may fail and block
2316 * layer knows how to live with it.
2317 */
2318 create_io_context(GFP_ATOMIC, q->node);
2319
ae118896
TH
2320 if (!blkcg_bio_issue_check(q, bio))
2321 return false;
27a84d54 2322
fbbaf700
N
2323 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2324 trace_block_bio_queue(q, bio);
2325 /* Now that enqueuing has been traced, we need to trace
2326 * completion as well.
2327 */
2328 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2329 }
27a84d54 2330 return true;
a7384677 2331
288dab8a 2332not_supported:
4e4cbee9 2333 status = BLK_STS_NOTSUPP;
a7384677 2334end_io:
4e4cbee9 2335 bio->bi_status = status;
4246a0b6 2336 bio_endio(bio);
27a84d54 2337 return false;
1da177e4
LT
2338}
2339
27a84d54
CH
2340/**
2341 * generic_make_request - hand a buffer to its device driver for I/O
2342 * @bio: The bio describing the location in memory and on the device.
2343 *
2344 * generic_make_request() is used to make I/O requests of block
2345 * devices. It is passed a &struct bio, which describes the I/O that needs
2346 * to be done.
2347 *
2348 * generic_make_request() does not return any status. The
2349 * success/failure status of the request, along with notification of
2350 * completion, is delivered asynchronously through the bio->bi_end_io
2351 * function described (one day) else where.
2352 *
2353 * The caller of generic_make_request must make sure that bi_io_vec
2354 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2355 * set to describe the device address, and the
2356 * bi_end_io and optionally bi_private are set to describe how
2357 * completion notification should be signaled.
2358 *
2359 * generic_make_request and the drivers it calls may use bi_next if this
2360 * bio happens to be merged with someone else, and may resubmit the bio to
2361 * a lower device by calling into generic_make_request recursively, which
2362 * means the bio should NOT be touched after the call to ->make_request_fn.
d89d8796 2363 */
dece1635 2364blk_qc_t generic_make_request(struct bio *bio)
d89d8796 2365{
f5fe1b51
N
2366 /*
2367 * bio_list_on_stack[0] contains bios submitted by the current
2368 * make_request_fn.
2369 * bio_list_on_stack[1] contains bios that were submitted before
2370 * the current make_request_fn, but that haven't been processed
2371 * yet.
2372 */
2373 struct bio_list bio_list_on_stack[2];
37f9579f
BVA
2374 blk_mq_req_flags_t flags = 0;
2375 struct request_queue *q = bio->bi_disk->queue;
dece1635 2376 blk_qc_t ret = BLK_QC_T_NONE;
bddd87c7 2377
37f9579f
BVA
2378 if (bio->bi_opf & REQ_NOWAIT)
2379 flags = BLK_MQ_REQ_NOWAIT;
cd4a4ae4
JA
2380 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
2381 blk_queue_enter_live(q);
2382 else if (blk_queue_enter(q, flags) < 0) {
37f9579f
BVA
2383 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2384 bio_wouldblock_error(bio);
2385 else
2386 bio_io_error(bio);
2387 return ret;
2388 }
2389
27a84d54 2390 if (!generic_make_request_checks(bio))
dece1635 2391 goto out;
27a84d54
CH
2392
2393 /*
2394 * We only want one ->make_request_fn to be active at a time, else
2395 * stack usage with stacked devices could be a problem. So use
2396 * current->bio_list to keep a list of requests submited by a
2397 * make_request_fn function. current->bio_list is also used as a
2398 * flag to say if generic_make_request is currently active in this
2399 * task or not. If it is NULL, then no make_request is active. If
2400 * it is non-NULL, then a make_request is active, and new requests
2401 * should be added at the tail
2402 */
bddd87c7 2403 if (current->bio_list) {
f5fe1b51 2404 bio_list_add(&current->bio_list[0], bio);
dece1635 2405 goto out;
d89d8796 2406 }
27a84d54 2407
d89d8796
NB
2408 /* following loop may be a bit non-obvious, and so deserves some
2409 * explanation.
2410 * Before entering the loop, bio->bi_next is NULL (as all callers
2411 * ensure that) so we have a list with a single bio.
2412 * We pretend that we have just taken it off a longer list, so
bddd87c7
AM
2413 * we assign bio_list to a pointer to the bio_list_on_stack,
2414 * thus initialising the bio_list of new bios to be
27a84d54 2415 * added. ->make_request() may indeed add some more bios
d89d8796
NB
2416 * through a recursive call to generic_make_request. If it
2417 * did, we find a non-NULL value in bio_list and re-enter the loop
2418 * from the top. In this case we really did just take the bio
bddd87c7 2419 * of the top of the list (no pretending) and so remove it from
27a84d54 2420 * bio_list, and call into ->make_request() again.
d89d8796
NB
2421 */
2422 BUG_ON(bio->bi_next);
f5fe1b51
N
2423 bio_list_init(&bio_list_on_stack[0]);
2424 current->bio_list = bio_list_on_stack;
d89d8796 2425 do {
37f9579f
BVA
2426 bool enter_succeeded = true;
2427
2428 if (unlikely(q != bio->bi_disk->queue)) {
2429 if (q)
2430 blk_queue_exit(q);
2431 q = bio->bi_disk->queue;
2432 flags = 0;
2433 if (bio->bi_opf & REQ_NOWAIT)
2434 flags = BLK_MQ_REQ_NOWAIT;
2435 if (blk_queue_enter(q, flags) < 0) {
2436 enter_succeeded = false;
2437 q = NULL;
2438 }
2439 }
27a84d54 2440
37f9579f 2441 if (enter_succeeded) {
79bd9959
N
2442 struct bio_list lower, same;
2443
2444 /* Create a fresh bio_list for all subordinate requests */
f5fe1b51
N
2445 bio_list_on_stack[1] = bio_list_on_stack[0];
2446 bio_list_init(&bio_list_on_stack[0]);
dece1635 2447 ret = q->make_request_fn(q, bio);
3ef28e83 2448
79bd9959
N
2449 /* sort new bios into those for a lower level
2450 * and those for the same level
2451 */
2452 bio_list_init(&lower);
2453 bio_list_init(&same);
f5fe1b51 2454 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
74d46992 2455 if (q == bio->bi_disk->queue)
79bd9959
N
2456 bio_list_add(&same, bio);
2457 else
2458 bio_list_add(&lower, bio);
2459 /* now assemble so we handle the lowest level first */
f5fe1b51
N
2460 bio_list_merge(&bio_list_on_stack[0], &lower);
2461 bio_list_merge(&bio_list_on_stack[0], &same);
2462 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
3ef28e83 2463 } else {
03a07c92
GR
2464 if (unlikely(!blk_queue_dying(q) &&
2465 (bio->bi_opf & REQ_NOWAIT)))
2466 bio_wouldblock_error(bio);
2467 else
2468 bio_io_error(bio);
3ef28e83 2469 }
f5fe1b51 2470 bio = bio_list_pop(&bio_list_on_stack[0]);
d89d8796 2471 } while (bio);
bddd87c7 2472 current->bio_list = NULL; /* deactivate */
dece1635
JA
2473
2474out:
37f9579f
BVA
2475 if (q)
2476 blk_queue_exit(q);
dece1635 2477 return ret;
d89d8796 2478}
1da177e4
LT
2479EXPORT_SYMBOL(generic_make_request);
2480
f421e1d9
CH
2481/**
2482 * direct_make_request - hand a buffer directly to its device driver for I/O
2483 * @bio: The bio describing the location in memory and on the device.
2484 *
2485 * This function behaves like generic_make_request(), but does not protect
2486 * against recursion. Must only be used if the called driver is known
2487 * to not call generic_make_request (or direct_make_request) again from
2488 * its make_request function. (Calling direct_make_request again from
2489 * a workqueue is perfectly fine as that doesn't recurse).
2490 */
2491blk_qc_t direct_make_request(struct bio *bio)
2492{
2493 struct request_queue *q = bio->bi_disk->queue;
2494 bool nowait = bio->bi_opf & REQ_NOWAIT;
2495 blk_qc_t ret;
2496
2497 if (!generic_make_request_checks(bio))
2498 return BLK_QC_T_NONE;
2499
3a0a5299 2500 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
f421e1d9
CH
2501 if (nowait && !blk_queue_dying(q))
2502 bio->bi_status = BLK_STS_AGAIN;
2503 else
2504 bio->bi_status = BLK_STS_IOERR;
2505 bio_endio(bio);
2506 return BLK_QC_T_NONE;
2507 }
2508
2509 ret = q->make_request_fn(q, bio);
2510 blk_queue_exit(q);
2511 return ret;
2512}
2513EXPORT_SYMBOL_GPL(direct_make_request);
2514
1da177e4 2515/**
710027a4 2516 * submit_bio - submit a bio to the block device layer for I/O
1da177e4
LT
2517 * @bio: The &struct bio which describes the I/O
2518 *
2519 * submit_bio() is very similar in purpose to generic_make_request(), and
2520 * uses that function to do most of the work. Both are fairly rough
710027a4 2521 * interfaces; @bio must be presetup and ready for I/O.
1da177e4
LT
2522 *
2523 */
4e49ea4a 2524blk_qc_t submit_bio(struct bio *bio)
1da177e4 2525{
bf2de6f5
JA
2526 /*
2527 * If it's a regular read/write or a barrier with data attached,
2528 * go through the normal accounting stuff before submission.
2529 */
e2a60da7 2530 if (bio_has_data(bio)) {
4363ac7c
MP
2531 unsigned int count;
2532
95fe6c1a 2533 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
7c5a0dcf 2534 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
4363ac7c
MP
2535 else
2536 count = bio_sectors(bio);
2537
a8ebb056 2538 if (op_is_write(bio_op(bio))) {
bf2de6f5
JA
2539 count_vm_events(PGPGOUT, count);
2540 } else {
4f024f37 2541 task_io_account_read(bio->bi_iter.bi_size);
bf2de6f5
JA
2542 count_vm_events(PGPGIN, count);
2543 }
2544
2545 if (unlikely(block_dump)) {
2546 char b[BDEVNAME_SIZE];
8dcbdc74 2547 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
ba25f9dc 2548 current->comm, task_pid_nr(current),
a8ebb056 2549 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
4f024f37 2550 (unsigned long long)bio->bi_iter.bi_sector,
74d46992 2551 bio_devname(bio, b), count);
bf2de6f5 2552 }
1da177e4
LT
2553 }
2554
dece1635 2555 return generic_make_request(bio);
1da177e4 2556}
1da177e4
LT
2557EXPORT_SYMBOL(submit_bio);
2558
ea435e1b
CH
2559bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2560{
2561 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2562 return false;
2563
2564 if (current->plug)
2565 blk_flush_plug_list(current->plug, false);
2566 return q->poll_fn(q, cookie);
2567}
2568EXPORT_SYMBOL_GPL(blk_poll);
2569
82124d60 2570/**
bf4e6b4e
HR
2571 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2572 * for new the queue limits
82124d60
KU
2573 * @q: the queue
2574 * @rq: the request being checked
2575 *
2576 * Description:
2577 * @rq may have been made based on weaker limitations of upper-level queues
2578 * in request stacking drivers, and it may violate the limitation of @q.
2579 * Since the block layer and the underlying device driver trust @rq
2580 * after it is inserted to @q, it should be checked against @q before
2581 * the insertion using this generic function.
2582 *
82124d60 2583 * Request stacking drivers like request-based dm may change the queue
bf4e6b4e
HR
2584 * limits when retrying requests on other queues. Those requests need
2585 * to be checked against the new queue limits again during dispatch.
82124d60 2586 */
bf4e6b4e
HR
2587static int blk_cloned_rq_check_limits(struct request_queue *q,
2588 struct request *rq)
82124d60 2589{
8fe0d473 2590 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
82124d60
KU
2591 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2592 return -EIO;
2593 }
2594
2595 /*
2596 * queue's settings related to segment counting like q->bounce_pfn
2597 * may differ from that of other stacking queues.
2598 * Recalculate it to check the request correctly on this queue's
2599 * limitation.
2600 */
2601 blk_recalc_rq_segments(rq);
8a78362c 2602 if (rq->nr_phys_segments > queue_max_segments(q)) {
82124d60
KU
2603 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2604 return -EIO;
2605 }
2606
2607 return 0;
2608}
82124d60
KU
2609
2610/**
2611 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2612 * @q: the queue to submit the request
2613 * @rq: the request being queued
2614 */
2a842aca 2615blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
82124d60
KU
2616{
2617 unsigned long flags;
4853abaa 2618 int where = ELEVATOR_INSERT_BACK;
82124d60 2619
bf4e6b4e 2620 if (blk_cloned_rq_check_limits(q, rq))
2a842aca 2621 return BLK_STS_IOERR;
82124d60 2622
b2c9cd37
AM
2623 if (rq->rq_disk &&
2624 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2a842aca 2625 return BLK_STS_IOERR;
82124d60 2626
7fb4898e
KB
2627 if (q->mq_ops) {
2628 if (blk_queue_io_stat(q))
2629 blk_account_io_start(rq, true);
157f377b
JA
2630 /*
2631 * Since we have a scheduler attached on the top device,
2632 * bypass a potential scheduler on the bottom device for
2633 * insert.
2634 */
c77ff7fd 2635 return blk_mq_request_issue_directly(rq);
7fb4898e
KB
2636 }
2637
82124d60 2638 spin_lock_irqsave(q->queue_lock, flags);
3f3299d5 2639 if (unlikely(blk_queue_dying(q))) {
8ba61435 2640 spin_unlock_irqrestore(q->queue_lock, flags);
2a842aca 2641 return BLK_STS_IOERR;
8ba61435 2642 }
82124d60
KU
2643
2644 /*
2645 * Submitting request must be dequeued before calling this function
2646 * because it will be linked to another request_queue
2647 */
2648 BUG_ON(blk_queued_rq(rq));
2649
f73f44eb 2650 if (op_is_flush(rq->cmd_flags))
4853abaa
JM
2651 where = ELEVATOR_INSERT_FLUSH;
2652
2653 add_acct_request(q, rq, where);
e67b77c7
JM
2654 if (where == ELEVATOR_INSERT_FLUSH)
2655 __blk_run_queue(q);
82124d60
KU
2656 spin_unlock_irqrestore(q->queue_lock, flags);
2657
2a842aca 2658 return BLK_STS_OK;
82124d60
KU
2659}
2660EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2661
80a761fd
TH
2662/**
2663 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2664 * @rq: request to examine
2665 *
2666 * Description:
2667 * A request could be merge of IOs which require different failure
2668 * handling. This function determines the number of bytes which
2669 * can be failed from the beginning of the request without
2670 * crossing into area which need to be retried further.
2671 *
2672 * Return:
2673 * The number of bytes to fail.
80a761fd
TH
2674 */
2675unsigned int blk_rq_err_bytes(const struct request *rq)
2676{
2677 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2678 unsigned int bytes = 0;
2679 struct bio *bio;
2680
e8064021 2681 if (!(rq->rq_flags & RQF_MIXED_MERGE))
80a761fd
TH
2682 return blk_rq_bytes(rq);
2683
2684 /*
2685 * Currently the only 'mixing' which can happen is between
2686 * different fastfail types. We can safely fail portions
2687 * which have all the failfast bits that the first one has -
2688 * the ones which are at least as eager to fail as the first
2689 * one.
2690 */
2691 for (bio = rq->bio; bio; bio = bio->bi_next) {
1eff9d32 2692 if ((bio->bi_opf & ff) != ff)
80a761fd 2693 break;
4f024f37 2694 bytes += bio->bi_iter.bi_size;
80a761fd
TH
2695 }
2696
2697 /* this could lead to infinite loop */
2698 BUG_ON(blk_rq_bytes(rq) && !bytes);
2699 return bytes;
2700}
2701EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2702
320ae51f 2703void blk_account_io_completion(struct request *req, unsigned int bytes)
bc58ba94 2704{
c2553b58 2705 if (blk_do_io_stat(req)) {
bc58ba94
JA
2706 const int rw = rq_data_dir(req);
2707 struct hd_struct *part;
2708 int cpu;
2709
2710 cpu = part_stat_lock();
09e099d4 2711 part = req->part;
bc58ba94
JA
2712 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2713 part_stat_unlock();
2714 }
2715}
2716
522a7775 2717void blk_account_io_done(struct request *req, u64 now)
bc58ba94 2718{
bc58ba94 2719 /*
dd4c133f
TH
2720 * Account IO completion. flush_rq isn't accounted as a
2721 * normal IO on queueing nor completion. Accounting the
2722 * containing request is enough.
bc58ba94 2723 */
e8064021 2724 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
522a7775 2725 unsigned long duration;
bc58ba94
JA
2726 const int rw = rq_data_dir(req);
2727 struct hd_struct *part;
2728 int cpu;
2729
522a7775 2730 duration = nsecs_to_jiffies(now - req->start_time_ns);
bc58ba94 2731 cpu = part_stat_lock();
09e099d4 2732 part = req->part;
bc58ba94
JA
2733
2734 part_stat_inc(cpu, part, ios[rw]);
2735 part_stat_add(cpu, part, ticks[rw], duration);
d62e26b3
JA
2736 part_round_stats(req->q, cpu, part);
2737 part_dec_in_flight(req->q, part, rw);
bc58ba94 2738
6c23a968 2739 hd_struct_put(part);
bc58ba94
JA
2740 part_stat_unlock();
2741 }
2742}
2743
47fafbc7 2744#ifdef CONFIG_PM
c8158819
LM
2745/*
2746 * Don't process normal requests when queue is suspended
2747 * or in the process of suspending/resuming
2748 */
e4f36b24 2749static bool blk_pm_allow_request(struct request *rq)
c8158819 2750{
e4f36b24
CH
2751 switch (rq->q->rpm_status) {
2752 case RPM_RESUMING:
2753 case RPM_SUSPENDING:
2754 return rq->rq_flags & RQF_PM;
2755 case RPM_SUSPENDED:
2756 return false;
2757 }
2758
2759 return true;
c8158819
LM
2760}
2761#else
e4f36b24 2762static bool blk_pm_allow_request(struct request *rq)
c8158819 2763{
e4f36b24 2764 return true;
c8158819
LM
2765}
2766#endif
2767
320ae51f
JA
2768void blk_account_io_start(struct request *rq, bool new_io)
2769{
2770 struct hd_struct *part;
2771 int rw = rq_data_dir(rq);
2772 int cpu;
2773
2774 if (!blk_do_io_stat(rq))
2775 return;
2776
2777 cpu = part_stat_lock();
2778
2779 if (!new_io) {
2780 part = rq->part;
2781 part_stat_inc(cpu, part, merges[rw]);
2782 } else {
2783 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2784 if (!hd_struct_try_get(part)) {
2785 /*
2786 * The partition is already being removed,
2787 * the request will be accounted on the disk only
2788 *
2789 * We take a reference on disk->part0 although that
2790 * partition will never be deleted, so we can treat
2791 * it as any other partition.
2792 */
2793 part = &rq->rq_disk->part0;
2794 hd_struct_get(part);
2795 }
d62e26b3
JA
2796 part_round_stats(rq->q, cpu, part);
2797 part_inc_in_flight(rq->q, part, rw);
320ae51f
JA
2798 rq->part = part;
2799 }
2800
2801 part_stat_unlock();
2802}
2803
9c988374
CH
2804static struct request *elv_next_request(struct request_queue *q)
2805{
2806 struct request *rq;
2807 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2808
2809 WARN_ON_ONCE(q->mq_ops);
2810
2811 while (1) {
e4f36b24
CH
2812 list_for_each_entry(rq, &q->queue_head, queuelist) {
2813 if (blk_pm_allow_request(rq))
2814 return rq;
2815
2816 if (rq->rq_flags & RQF_SOFTBARRIER)
2817 break;
9c988374
CH
2818 }
2819
2820 /*
2821 * Flush request is running and flush request isn't queueable
2822 * in the drive, we can hold the queue till flush request is
2823 * finished. Even we don't do this, driver can't dispatch next
2824 * requests and will requeue them. And this can improve
2825 * throughput too. For example, we have request flush1, write1,
2826 * flush 2. flush1 is dispatched, then queue is hold, write1
2827 * isn't inserted to queue. After flush1 is finished, flush2
2828 * will be dispatched. Since disk cache is already clean,
2829 * flush2 will be finished very soon, so looks like flush2 is
2830 * folded to flush1.
2831 * Since the queue is hold, a flag is set to indicate the queue
2832 * should be restarted later. Please see flush_end_io() for
2833 * details.
2834 */
2835 if (fq->flush_pending_idx != fq->flush_running_idx &&
2836 !queue_flush_queueable(q)) {
2837 fq->flush_queue_delayed = 1;
2838 return NULL;
2839 }
2840 if (unlikely(blk_queue_bypass(q)) ||
2841 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2842 return NULL;
2843 }
2844}
2845
3bcddeac 2846/**
9934c8c0
TH
2847 * blk_peek_request - peek at the top of a request queue
2848 * @q: request queue to peek at
2849 *
2850 * Description:
2851 * Return the request at the top of @q. The returned request
2852 * should be started using blk_start_request() before LLD starts
2853 * processing it.
2854 *
2855 * Return:
2856 * Pointer to the request at the top of @q if available. Null
2857 * otherwise.
9934c8c0
TH
2858 */
2859struct request *blk_peek_request(struct request_queue *q)
158dbda0
TH
2860{
2861 struct request *rq;
2862 int ret;
2863
2fff8a92 2864 lockdep_assert_held(q->queue_lock);
332ebbf7 2865 WARN_ON_ONCE(q->mq_ops);
2fff8a92 2866
9c988374 2867 while ((rq = elv_next_request(q)) != NULL) {
e8064021 2868 if (!(rq->rq_flags & RQF_STARTED)) {
158dbda0
TH
2869 /*
2870 * This is the first time the device driver
2871 * sees this request (possibly after
2872 * requeueing). Notify IO scheduler.
2873 */
e8064021 2874 if (rq->rq_flags & RQF_SORTED)
158dbda0
TH
2875 elv_activate_rq(q, rq);
2876
2877 /*
2878 * just mark as started even if we don't start
2879 * it, a request that has been delayed should
2880 * not be passed by new incoming requests
2881 */
e8064021 2882 rq->rq_flags |= RQF_STARTED;
158dbda0
TH
2883 trace_block_rq_issue(q, rq);
2884 }
2885
2886 if (!q->boundary_rq || q->boundary_rq == rq) {
2887 q->end_sector = rq_end_sector(rq);
2888 q->boundary_rq = NULL;
2889 }
2890
e8064021 2891 if (rq->rq_flags & RQF_DONTPREP)
158dbda0
TH
2892 break;
2893
2e46e8b2 2894 if (q->dma_drain_size && blk_rq_bytes(rq)) {
158dbda0
TH
2895 /*
2896 * make sure space for the drain appears we
2897 * know we can do this because max_hw_segments
2898 * has been adjusted to be one fewer than the
2899 * device can handle
2900 */
2901 rq->nr_phys_segments++;
2902 }
2903
2904 if (!q->prep_rq_fn)
2905 break;
2906
2907 ret = q->prep_rq_fn(q, rq);
2908 if (ret == BLKPREP_OK) {
2909 break;
2910 } else if (ret == BLKPREP_DEFER) {
2911 /*
2912 * the request may have been (partially) prepped.
2913 * we need to keep this request in the front to
e8064021 2914 * avoid resource deadlock. RQF_STARTED will
158dbda0
TH
2915 * prevent other fs requests from passing this one.
2916 */
2e46e8b2 2917 if (q->dma_drain_size && blk_rq_bytes(rq) &&
e8064021 2918 !(rq->rq_flags & RQF_DONTPREP)) {
158dbda0
TH
2919 /*
2920 * remove the space for the drain we added
2921 * so that we don't add it again
2922 */
2923 --rq->nr_phys_segments;
2924 }
2925
2926 rq = NULL;
2927 break;
0fb5b1fb 2928 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
e8064021 2929 rq->rq_flags |= RQF_QUIET;
c143dc90
JB
2930 /*
2931 * Mark this request as started so we don't trigger
2932 * any debug logic in the end I/O path.
2933 */
2934 blk_start_request(rq);
2a842aca
CH
2935 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2936 BLK_STS_TARGET : BLK_STS_IOERR);
158dbda0
TH
2937 } else {
2938 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2939 break;
2940 }
2941 }
2942
2943 return rq;
2944}
9934c8c0 2945EXPORT_SYMBOL(blk_peek_request);
158dbda0 2946
5034435c 2947static void blk_dequeue_request(struct request *rq)
158dbda0 2948{
9934c8c0
TH
2949 struct request_queue *q = rq->q;
2950
158dbda0
TH
2951 BUG_ON(list_empty(&rq->queuelist));
2952 BUG_ON(ELV_ON_HASH(rq));
2953
2954 list_del_init(&rq->queuelist);
2955
2956 /*
2957 * the time frame between a request being removed from the lists
2958 * and to it is freed is accounted as io that is in progress at
2959 * the driver side.
2960 */
522a7775 2961 if (blk_account_rq(rq))
0a7ae2ff 2962 q->in_flight[rq_is_sync(rq)]++;
158dbda0
TH
2963}
2964
9934c8c0
TH
2965/**
2966 * blk_start_request - start request processing on the driver
2967 * @req: request to dequeue
2968 *
2969 * Description:
2970 * Dequeue @req and start timeout timer on it. This hands off the
2971 * request to the driver.
9934c8c0
TH
2972 */
2973void blk_start_request(struct request *req)
2974{
2fff8a92 2975 lockdep_assert_held(req->q->queue_lock);
332ebbf7 2976 WARN_ON_ONCE(req->q->mq_ops);
2fff8a92 2977
9934c8c0
TH
2978 blk_dequeue_request(req);
2979
cf43e6be 2980 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
544ccc8d
OS
2981 req->io_start_time_ns = ktime_get_ns();
2982#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2983 req->throtl_size = blk_rq_sectors(req);
2984#endif
cf43e6be 2985 req->rq_flags |= RQF_STATS;
a8a45941 2986 wbt_issue(req->q->rq_wb, req);
cf43e6be
JA
2987 }
2988
e14575b3 2989 BUG_ON(blk_rq_is_complete(req));
9934c8c0
TH
2990 blk_add_timer(req);
2991}
2992EXPORT_SYMBOL(blk_start_request);
2993
2994/**
2995 * blk_fetch_request - fetch a request from a request queue
2996 * @q: request queue to fetch a request from
2997 *
2998 * Description:
2999 * Return the request at the top of @q. The request is started on
3000 * return and LLD can start processing it immediately.
3001 *
3002 * Return:
3003 * Pointer to the request at the top of @q if available. Null
3004 * otherwise.
9934c8c0
TH
3005 */
3006struct request *blk_fetch_request(struct request_queue *q)
3007{
3008 struct request *rq;
3009
2fff8a92 3010 lockdep_assert_held(q->queue_lock);
332ebbf7 3011 WARN_ON_ONCE(q->mq_ops);
2fff8a92 3012
9934c8c0
TH
3013 rq = blk_peek_request(q);
3014 if (rq)
3015 blk_start_request(rq);
3016 return rq;
3017}
3018EXPORT_SYMBOL(blk_fetch_request);
3019
ef71de8b
CH
3020/*
3021 * Steal bios from a request and add them to a bio list.
3022 * The request must not have been partially completed before.
3023 */
3024void blk_steal_bios(struct bio_list *list, struct request *rq)
3025{
3026 if (rq->bio) {
3027 if (list->tail)
3028 list->tail->bi_next = rq->bio;
3029 else
3030 list->head = rq->bio;
3031 list->tail = rq->biotail;
3032
3033 rq->bio = NULL;
3034 rq->biotail = NULL;
3035 }
3036
3037 rq->__data_len = 0;
3038}
3039EXPORT_SYMBOL_GPL(blk_steal_bios);
3040
3bcddeac 3041/**
2e60e022 3042 * blk_update_request - Special helper function for request stacking drivers
8ebf9756 3043 * @req: the request being processed
2a842aca 3044 * @error: block status code
8ebf9756 3045 * @nr_bytes: number of bytes to complete @req
3bcddeac
KU
3046 *
3047 * Description:
8ebf9756
RD
3048 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3049 * the request structure even if @req doesn't have leftover.
3050 * If @req has leftover, sets it up for the next range of segments.
2e60e022
TH
3051 *
3052 * This special helper function is only for request stacking drivers
3053 * (e.g. request-based dm) so that they can handle partial completion.
3054 * Actual device drivers should use blk_end_request instead.
3055 *
3056 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3057 * %false return from this function.
3bcddeac
KU
3058 *
3059 * Return:
2e60e022
TH
3060 * %false - this request doesn't have any more data
3061 * %true - this request has more data
3bcddeac 3062 **/
2a842aca
CH
3063bool blk_update_request(struct request *req, blk_status_t error,
3064 unsigned int nr_bytes)
1da177e4 3065{
f79ea416 3066 int total_bytes;
1da177e4 3067
2a842aca 3068 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
4a0efdc9 3069
2e60e022
TH
3070 if (!req->bio)
3071 return false;
3072
2a842aca
CH
3073 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3074 !(req->rq_flags & RQF_QUIET)))
3075 print_req_error(req, error);
1da177e4 3076
bc58ba94 3077 blk_account_io_completion(req, nr_bytes);
d72d904a 3078
f79ea416
KO
3079 total_bytes = 0;
3080 while (req->bio) {
3081 struct bio *bio = req->bio;
4f024f37 3082 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1da177e4 3083
9c24c10a 3084 if (bio_bytes == bio->bi_iter.bi_size)
1da177e4 3085 req->bio = bio->bi_next;
1da177e4 3086
fbbaf700
N
3087 /* Completion has already been traced */
3088 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
f79ea416 3089 req_bio_endio(req, bio, bio_bytes, error);
1da177e4 3090
f79ea416
KO
3091 total_bytes += bio_bytes;
3092 nr_bytes -= bio_bytes;
1da177e4 3093
f79ea416
KO
3094 if (!nr_bytes)
3095 break;
1da177e4
LT
3096 }
3097
3098 /*
3099 * completely done
3100 */
2e60e022
TH
3101 if (!req->bio) {
3102 /*
3103 * Reset counters so that the request stacking driver
3104 * can find how many bytes remain in the request
3105 * later.
3106 */
a2dec7b3 3107 req->__data_len = 0;
2e60e022
TH
3108 return false;
3109 }
1da177e4 3110
a2dec7b3 3111 req->__data_len -= total_bytes;
2e46e8b2
TH
3112
3113 /* update sector only for requests with clear definition of sector */
57292b58 3114 if (!blk_rq_is_passthrough(req))
a2dec7b3 3115 req->__sector += total_bytes >> 9;
2e46e8b2 3116
80a761fd 3117 /* mixed attributes always follow the first bio */
e8064021 3118 if (req->rq_flags & RQF_MIXED_MERGE) {
80a761fd 3119 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1eff9d32 3120 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
80a761fd
TH
3121 }
3122
ed6565e7
CH
3123 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3124 /*
3125 * If total number of sectors is less than the first segment
3126 * size, something has gone terribly wrong.
3127 */
3128 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3129 blk_dump_rq_flags(req, "request botched");
3130 req->__data_len = blk_rq_cur_bytes(req);
3131 }
2e46e8b2 3132
ed6565e7
CH
3133 /* recalculate the number of segments */
3134 blk_recalc_rq_segments(req);
3135 }
2e46e8b2 3136
2e60e022 3137 return true;
1da177e4 3138}
2e60e022 3139EXPORT_SYMBOL_GPL(blk_update_request);
1da177e4 3140
2a842aca 3141static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2e60e022
TH
3142 unsigned int nr_bytes,
3143 unsigned int bidi_bytes)
5efccd17 3144{
2e60e022
TH
3145 if (blk_update_request(rq, error, nr_bytes))
3146 return true;
5efccd17 3147
2e60e022
TH
3148 /* Bidi request must be completed as a whole */
3149 if (unlikely(blk_bidi_rq(rq)) &&
3150 blk_update_request(rq->next_rq, error, bidi_bytes))
3151 return true;
5efccd17 3152
e2e1a148
JA
3153 if (blk_queue_add_random(rq->q))
3154 add_disk_randomness(rq->rq_disk);
2e60e022
TH
3155
3156 return false;
1da177e4
LT
3157}
3158
28018c24
JB
3159/**
3160 * blk_unprep_request - unprepare a request
3161 * @req: the request
3162 *
3163 * This function makes a request ready for complete resubmission (or
3164 * completion). It happens only after all error handling is complete,
3165 * so represents the appropriate moment to deallocate any resources
3166 * that were allocated to the request in the prep_rq_fn. The queue
3167 * lock is held when calling this.
3168 */
3169void blk_unprep_request(struct request *req)
3170{
3171 struct request_queue *q = req->q;
3172
e8064021 3173 req->rq_flags &= ~RQF_DONTPREP;
28018c24
JB
3174 if (q->unprep_rq_fn)
3175 q->unprep_rq_fn(q, req);
3176}
3177EXPORT_SYMBOL_GPL(blk_unprep_request);
3178
2a842aca 3179void blk_finish_request(struct request *req, blk_status_t error)
1da177e4 3180{
cf43e6be 3181 struct request_queue *q = req->q;
522a7775 3182 u64 now = ktime_get_ns();
cf43e6be 3183
2fff8a92 3184 lockdep_assert_held(req->q->queue_lock);
332ebbf7 3185 WARN_ON_ONCE(q->mq_ops);
2fff8a92 3186
cf43e6be 3187 if (req->rq_flags & RQF_STATS)
522a7775 3188 blk_stat_add(req, now);
cf43e6be 3189
e8064021 3190 if (req->rq_flags & RQF_QUEUED)
cf43e6be 3191 blk_queue_end_tag(q, req);
b8286239 3192
ba396a6c 3193 BUG_ON(blk_queued_rq(req));
1da177e4 3194
57292b58 3195 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
dc3b17cc 3196 laptop_io_completion(req->q->backing_dev_info);
1da177e4 3197
e78042e5
MA
3198 blk_delete_timer(req);
3199
e8064021 3200 if (req->rq_flags & RQF_DONTPREP)
28018c24
JB
3201 blk_unprep_request(req);
3202
522a7775 3203 blk_account_io_done(req, now);
b8286239 3204
87760e5e 3205 if (req->end_io) {
a8a45941 3206 wbt_done(req->q->rq_wb, req);
8ffdc655 3207 req->end_io(req, error);
87760e5e 3208 } else {
b8286239
KU
3209 if (blk_bidi_rq(req))
3210 __blk_put_request(req->next_rq->q, req->next_rq);
3211
cf43e6be 3212 __blk_put_request(q, req);
b8286239 3213 }
1da177e4 3214}
12120077 3215EXPORT_SYMBOL(blk_finish_request);
1da177e4 3216
3b11313a 3217/**
2e60e022
TH
3218 * blk_end_bidi_request - Complete a bidi request
3219 * @rq: the request to complete
2a842aca 3220 * @error: block status code
2e60e022
TH
3221 * @nr_bytes: number of bytes to complete @rq
3222 * @bidi_bytes: number of bytes to complete @rq->next_rq
a0cd1285
JA
3223 *
3224 * Description:
e3a04fe3 3225 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2e60e022
TH
3226 * Drivers that supports bidi can safely call this member for any
3227 * type of request, bidi or uni. In the later case @bidi_bytes is
3228 * just ignored.
336cdb40
KU
3229 *
3230 * Return:
2e60e022
TH
3231 * %false - we are done with this request
3232 * %true - still buffers pending for this request
a0cd1285 3233 **/
2a842aca 3234static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
32fab448
KU
3235 unsigned int nr_bytes, unsigned int bidi_bytes)
3236{
336cdb40 3237 struct request_queue *q = rq->q;
2e60e022 3238 unsigned long flags;
32fab448 3239
332ebbf7
BVA
3240 WARN_ON_ONCE(q->mq_ops);
3241
2e60e022
TH
3242 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3243 return true;
32fab448 3244
336cdb40 3245 spin_lock_irqsave(q->queue_lock, flags);
2e60e022 3246 blk_finish_request(rq, error);
336cdb40
KU
3247 spin_unlock_irqrestore(q->queue_lock, flags);
3248
2e60e022 3249 return false;
32fab448
KU
3250}
3251
336cdb40 3252/**
2e60e022
TH
3253 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3254 * @rq: the request to complete
2a842aca 3255 * @error: block status code
e3a04fe3
KU
3256 * @nr_bytes: number of bytes to complete @rq
3257 * @bidi_bytes: number of bytes to complete @rq->next_rq
336cdb40
KU
3258 *
3259 * Description:
2e60e022
TH
3260 * Identical to blk_end_bidi_request() except that queue lock is
3261 * assumed to be locked on entry and remains so on return.
336cdb40
KU
3262 *
3263 * Return:
2e60e022
TH
3264 * %false - we are done with this request
3265 * %true - still buffers pending for this request
336cdb40 3266 **/
2a842aca 3267static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
b1f74493 3268 unsigned int nr_bytes, unsigned int bidi_bytes)
336cdb40 3269{
2fff8a92 3270 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3271 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3272
2e60e022
TH
3273 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3274 return true;
336cdb40 3275
2e60e022 3276 blk_finish_request(rq, error);
336cdb40 3277
2e60e022 3278 return false;
336cdb40 3279}
e19a3ab0
KU
3280
3281/**
3282 * blk_end_request - Helper function for drivers to complete the request.
3283 * @rq: the request being processed
2a842aca 3284 * @error: block status code
e19a3ab0
KU
3285 * @nr_bytes: number of bytes to complete
3286 *
3287 * Description:
3288 * Ends I/O on a number of bytes attached to @rq.
3289 * If @rq has leftover, sets it up for the next range of segments.
3290 *
3291 * Return:
b1f74493
FT
3292 * %false - we are done with this request
3293 * %true - still buffers pending for this request
e19a3ab0 3294 **/
2a842aca
CH
3295bool blk_end_request(struct request *rq, blk_status_t error,
3296 unsigned int nr_bytes)
e19a3ab0 3297{
332ebbf7 3298 WARN_ON_ONCE(rq->q->mq_ops);
b1f74493 3299 return blk_end_bidi_request(rq, error, nr_bytes, 0);
e19a3ab0 3300}
56ad1740 3301EXPORT_SYMBOL(blk_end_request);
336cdb40
KU
3302
3303/**
b1f74493
FT
3304 * blk_end_request_all - Helper function for drives to finish the request.
3305 * @rq: the request to finish
2a842aca 3306 * @error: block status code
336cdb40
KU
3307 *
3308 * Description:
b1f74493
FT
3309 * Completely finish @rq.
3310 */
2a842aca 3311void blk_end_request_all(struct request *rq, blk_status_t error)
336cdb40 3312{
b1f74493
FT
3313 bool pending;
3314 unsigned int bidi_bytes = 0;
336cdb40 3315
b1f74493
FT
3316 if (unlikely(blk_bidi_rq(rq)))
3317 bidi_bytes = blk_rq_bytes(rq->next_rq);
336cdb40 3318
b1f74493
FT
3319 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3320 BUG_ON(pending);
3321}
56ad1740 3322EXPORT_SYMBOL(blk_end_request_all);
336cdb40 3323
e3a04fe3 3324/**
b1f74493
FT
3325 * __blk_end_request - Helper function for drivers to complete the request.
3326 * @rq: the request being processed
2a842aca 3327 * @error: block status code
b1f74493 3328 * @nr_bytes: number of bytes to complete
e3a04fe3
KU
3329 *
3330 * Description:
b1f74493 3331 * Must be called with queue lock held unlike blk_end_request().
e3a04fe3
KU
3332 *
3333 * Return:
b1f74493
FT
3334 * %false - we are done with this request
3335 * %true - still buffers pending for this request
e3a04fe3 3336 **/
2a842aca
CH
3337bool __blk_end_request(struct request *rq, blk_status_t error,
3338 unsigned int nr_bytes)
e3a04fe3 3339{
2fff8a92 3340 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3341 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3342
b1f74493 3343 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
e3a04fe3 3344}
56ad1740 3345EXPORT_SYMBOL(__blk_end_request);
e3a04fe3 3346
32fab448 3347/**
b1f74493
FT
3348 * __blk_end_request_all - Helper function for drives to finish the request.
3349 * @rq: the request to finish
2a842aca 3350 * @error: block status code
32fab448
KU
3351 *
3352 * Description:
b1f74493 3353 * Completely finish @rq. Must be called with queue lock held.
32fab448 3354 */
2a842aca 3355void __blk_end_request_all(struct request *rq, blk_status_t error)
32fab448 3356{
b1f74493
FT
3357 bool pending;
3358 unsigned int bidi_bytes = 0;
3359
2fff8a92 3360 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3361 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3362
b1f74493
FT
3363 if (unlikely(blk_bidi_rq(rq)))
3364 bidi_bytes = blk_rq_bytes(rq->next_rq);
3365
3366 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3367 BUG_ON(pending);
32fab448 3368}
56ad1740 3369EXPORT_SYMBOL(__blk_end_request_all);
32fab448 3370
e19a3ab0 3371/**
b1f74493
FT
3372 * __blk_end_request_cur - Helper function to finish the current request chunk.
3373 * @rq: the request to finish the current chunk for
2a842aca 3374 * @error: block status code
e19a3ab0
KU
3375 *
3376 * Description:
b1f74493
FT
3377 * Complete the current consecutively mapped chunk from @rq. Must
3378 * be called with queue lock held.
e19a3ab0
KU
3379 *
3380 * Return:
b1f74493
FT
3381 * %false - we are done with this request
3382 * %true - still buffers pending for this request
3383 */
2a842aca 3384bool __blk_end_request_cur(struct request *rq, blk_status_t error)
e19a3ab0 3385{
b1f74493 3386 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
e19a3ab0 3387}
56ad1740 3388EXPORT_SYMBOL(__blk_end_request_cur);
e19a3ab0 3389
86db1e29
JA
3390void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3391 struct bio *bio)
1da177e4 3392{
b4f42e28 3393 if (bio_has_data(bio))
fb2dce86 3394 rq->nr_phys_segments = bio_phys_segments(q, bio);
445251d0
JA
3395 else if (bio_op(bio) == REQ_OP_DISCARD)
3396 rq->nr_phys_segments = 1;
b4f42e28 3397
4f024f37 3398 rq->__data_len = bio->bi_iter.bi_size;
1da177e4 3399 rq->bio = rq->biotail = bio;
1da177e4 3400
74d46992
CH
3401 if (bio->bi_disk)
3402 rq->rq_disk = bio->bi_disk;
66846572 3403}
1da177e4 3404
2d4dc890
IL
3405#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3406/**
3407 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3408 * @rq: the request to be flushed
3409 *
3410 * Description:
3411 * Flush all pages in @rq.
3412 */
3413void rq_flush_dcache_pages(struct request *rq)
3414{
3415 struct req_iterator iter;
7988613b 3416 struct bio_vec bvec;
2d4dc890
IL
3417
3418 rq_for_each_segment(bvec, rq, iter)
7988613b 3419 flush_dcache_page(bvec.bv_page);
2d4dc890
IL
3420}
3421EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3422#endif
3423
ef9e3fac
KU
3424/**
3425 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3426 * @q : the queue of the device being checked
3427 *
3428 * Description:
3429 * Check if underlying low-level drivers of a device are busy.
3430 * If the drivers want to export their busy state, they must set own
3431 * exporting function using blk_queue_lld_busy() first.
3432 *
3433 * Basically, this function is used only by request stacking drivers
3434 * to stop dispatching requests to underlying devices when underlying
3435 * devices are busy. This behavior helps more I/O merging on the queue
3436 * of the request stacking driver and prevents I/O throughput regression
3437 * on burst I/O load.
3438 *
3439 * Return:
3440 * 0 - Not busy (The request stacking driver should dispatch request)
3441 * 1 - Busy (The request stacking driver should stop dispatching request)
3442 */
3443int blk_lld_busy(struct request_queue *q)
3444{
3445 if (q->lld_busy_fn)
3446 return q->lld_busy_fn(q);
3447
3448 return 0;
3449}
3450EXPORT_SYMBOL_GPL(blk_lld_busy);
3451
78d8e58a
MS
3452/**
3453 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3454 * @rq: the clone request to be cleaned up
3455 *
3456 * Description:
3457 * Free all bios in @rq for a cloned request.
3458 */
3459void blk_rq_unprep_clone(struct request *rq)
3460{
3461 struct bio *bio;
3462
3463 while ((bio = rq->bio) != NULL) {
3464 rq->bio = bio->bi_next;
3465
3466 bio_put(bio);
3467 }
3468}
3469EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3470
3471/*
3472 * Copy attributes of the original request to the clone request.
3473 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3474 */
3475static void __blk_rq_prep_clone(struct request *dst, struct request *src)
b0fd271d
KU
3476{
3477 dst->cpu = src->cpu;
b0fd271d
KU
3478 dst->__sector = blk_rq_pos(src);
3479 dst->__data_len = blk_rq_bytes(src);
297ba57d
BVA
3480 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
3481 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
3482 dst->special_vec = src->special_vec;
3483 }
b0fd271d
KU
3484 dst->nr_phys_segments = src->nr_phys_segments;
3485 dst->ioprio = src->ioprio;
3486 dst->extra_len = src->extra_len;
78d8e58a
MS
3487}
3488
3489/**
3490 * blk_rq_prep_clone - Helper function to setup clone request
3491 * @rq: the request to be setup
3492 * @rq_src: original request to be cloned
3493 * @bs: bio_set that bios for clone are allocated from
3494 * @gfp_mask: memory allocation mask for bio
3495 * @bio_ctr: setup function to be called for each clone bio.
3496 * Returns %0 for success, non %0 for failure.
3497 * @data: private data to be passed to @bio_ctr
3498 *
3499 * Description:
3500 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3501 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3502 * are not copied, and copying such parts is the caller's responsibility.
3503 * Also, pages which the original bios are pointing to are not copied
3504 * and the cloned bios just point same pages.
3505 * So cloned bios must be completed before original bios, which means
3506 * the caller must complete @rq before @rq_src.
3507 */
3508int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3509 struct bio_set *bs, gfp_t gfp_mask,
3510 int (*bio_ctr)(struct bio *, struct bio *, void *),
3511 void *data)
3512{
3513 struct bio *bio, *bio_src;
3514
3515 if (!bs)
f4f8154a 3516 bs = &fs_bio_set;
78d8e58a
MS
3517
3518 __rq_for_each_bio(bio_src, rq_src) {
3519 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3520 if (!bio)
3521 goto free_and_out;
3522
3523 if (bio_ctr && bio_ctr(bio, bio_src, data))
3524 goto free_and_out;
3525
3526 if (rq->bio) {
3527 rq->biotail->bi_next = bio;
3528 rq->biotail = bio;
3529 } else
3530 rq->bio = rq->biotail = bio;
3531 }
3532
3533 __blk_rq_prep_clone(rq, rq_src);
3534
3535 return 0;
3536
3537free_and_out:
3538 if (bio)
3539 bio_put(bio);
3540 blk_rq_unprep_clone(rq);
3541
3542 return -ENOMEM;
b0fd271d
KU
3543}
3544EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3545
59c3d45e 3546int kblockd_schedule_work(struct work_struct *work)
1da177e4
LT
3547{
3548 return queue_work(kblockd_workqueue, work);
3549}
1da177e4
LT
3550EXPORT_SYMBOL(kblockd_schedule_work);
3551
ee63cfa7
JA
3552int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3553{
3554 return queue_work_on(cpu, kblockd_workqueue, work);
3555}
3556EXPORT_SYMBOL(kblockd_schedule_work_on);
3557
818cd1cb
JA
3558int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3559 unsigned long delay)
3560{
3561 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3562}
3563EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3564
75df7136
SJ
3565/**
3566 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3567 * @plug: The &struct blk_plug that needs to be initialized
3568 *
3569 * Description:
3570 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3571 * pending I/O should the task end up blocking between blk_start_plug() and
3572 * blk_finish_plug(). This is important from a performance perspective, but
3573 * also ensures that we don't deadlock. For instance, if the task is blocking
3574 * for a memory allocation, memory reclaim could end up wanting to free a
3575 * page belonging to that request that is currently residing in our private
3576 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3577 * this kind of deadlock.
3578 */
73c10101
JA
3579void blk_start_plug(struct blk_plug *plug)
3580{
3581 struct task_struct *tsk = current;
3582
dd6cf3e1
SL
3583 /*
3584 * If this is a nested plug, don't actually assign it.
3585 */
3586 if (tsk->plug)
3587 return;
3588
73c10101 3589 INIT_LIST_HEAD(&plug->list);
320ae51f 3590 INIT_LIST_HEAD(&plug->mq_list);
048c9374 3591 INIT_LIST_HEAD(&plug->cb_list);
73c10101 3592 /*
dd6cf3e1
SL
3593 * Store ordering should not be needed here, since a potential
3594 * preempt will imply a full memory barrier
73c10101 3595 */
dd6cf3e1 3596 tsk->plug = plug;
73c10101
JA
3597}
3598EXPORT_SYMBOL(blk_start_plug);
3599
3600static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3601{
3602 struct request *rqa = container_of(a, struct request, queuelist);
3603 struct request *rqb = container_of(b, struct request, queuelist);
3604
975927b9
JM
3605 return !(rqa->q < rqb->q ||
3606 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
73c10101
JA
3607}
3608
49cac01e
JA
3609/*
3610 * If 'from_schedule' is true, then postpone the dispatch of requests
3611 * until a safe kblockd context. We due this to avoid accidental big
3612 * additional stack usage in driver dispatch, in places where the originally
3613 * plugger did not intend it.
3614 */
f6603783 3615static void queue_unplugged(struct request_queue *q, unsigned int depth,
49cac01e 3616 bool from_schedule)
99e22598 3617 __releases(q->queue_lock)
94b5eb28 3618{
2fff8a92
BVA
3619 lockdep_assert_held(q->queue_lock);
3620
49cac01e 3621 trace_block_unplug(q, depth, !from_schedule);
99e22598 3622
70460571 3623 if (from_schedule)
24ecfbe2 3624 blk_run_queue_async(q);
70460571 3625 else
24ecfbe2 3626 __blk_run_queue(q);
50864670 3627 spin_unlock_irq(q->queue_lock);
94b5eb28
JA
3628}
3629
74018dc3 3630static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
048c9374
N
3631{
3632 LIST_HEAD(callbacks);
3633
2a7d5559
SL
3634 while (!list_empty(&plug->cb_list)) {
3635 list_splice_init(&plug->cb_list, &callbacks);
048c9374 3636
2a7d5559
SL
3637 while (!list_empty(&callbacks)) {
3638 struct blk_plug_cb *cb = list_first_entry(&callbacks,
048c9374
N
3639 struct blk_plug_cb,
3640 list);
2a7d5559 3641 list_del(&cb->list);
74018dc3 3642 cb->callback(cb, from_schedule);
2a7d5559 3643 }
048c9374
N
3644 }
3645}
3646
9cbb1750
N
3647struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3648 int size)
3649{
3650 struct blk_plug *plug = current->plug;
3651 struct blk_plug_cb *cb;
3652
3653 if (!plug)
3654 return NULL;
3655
3656 list_for_each_entry(cb, &plug->cb_list, list)
3657 if (cb->callback == unplug && cb->data == data)
3658 return cb;
3659
3660 /* Not currently on the callback list */
3661 BUG_ON(size < sizeof(*cb));
3662 cb = kzalloc(size, GFP_ATOMIC);
3663 if (cb) {
3664 cb->data = data;
3665 cb->callback = unplug;
3666 list_add(&cb->list, &plug->cb_list);
3667 }
3668 return cb;
3669}
3670EXPORT_SYMBOL(blk_check_plugged);
3671
49cac01e 3672void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
73c10101
JA
3673{
3674 struct request_queue *q;
73c10101 3675 struct request *rq;
109b8129 3676 LIST_HEAD(list);
94b5eb28 3677 unsigned int depth;
73c10101 3678
74018dc3 3679 flush_plug_callbacks(plug, from_schedule);
320ae51f
JA
3680
3681 if (!list_empty(&plug->mq_list))
3682 blk_mq_flush_plug_list(plug, from_schedule);
3683
73c10101
JA
3684 if (list_empty(&plug->list))
3685 return;
3686
109b8129
N
3687 list_splice_init(&plug->list, &list);
3688
422765c2 3689 list_sort(NULL, &list, plug_rq_cmp);
73c10101
JA
3690
3691 q = NULL;
94b5eb28 3692 depth = 0;
18811272 3693
109b8129
N
3694 while (!list_empty(&list)) {
3695 rq = list_entry_rq(list.next);
73c10101 3696 list_del_init(&rq->queuelist);
73c10101
JA
3697 BUG_ON(!rq->q);
3698 if (rq->q != q) {
99e22598
JA
3699 /*
3700 * This drops the queue lock
3701 */
3702 if (q)
49cac01e 3703 queue_unplugged(q, depth, from_schedule);
73c10101 3704 q = rq->q;
94b5eb28 3705 depth = 0;
50864670 3706 spin_lock_irq(q->queue_lock);
73c10101 3707 }
8ba61435
TH
3708
3709 /*
3710 * Short-circuit if @q is dead
3711 */
3f3299d5 3712 if (unlikely(blk_queue_dying(q))) {
2a842aca 3713 __blk_end_request_all(rq, BLK_STS_IOERR);
8ba61435
TH
3714 continue;
3715 }
3716
73c10101
JA
3717 /*
3718 * rq is already accounted, so use raw insert
3719 */
f73f44eb 3720 if (op_is_flush(rq->cmd_flags))
401a18e9
JA
3721 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3722 else
3723 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
94b5eb28
JA
3724
3725 depth++;
73c10101
JA
3726 }
3727
99e22598
JA
3728 /*
3729 * This drops the queue lock
3730 */
3731 if (q)
49cac01e 3732 queue_unplugged(q, depth, from_schedule);
73c10101 3733}
73c10101
JA
3734
3735void blk_finish_plug(struct blk_plug *plug)
3736{
dd6cf3e1
SL
3737 if (plug != current->plug)
3738 return;
f6603783 3739 blk_flush_plug_list(plug, false);
73c10101 3740
dd6cf3e1 3741 current->plug = NULL;
73c10101 3742}
88b996cd 3743EXPORT_SYMBOL(blk_finish_plug);
73c10101 3744
47fafbc7 3745#ifdef CONFIG_PM
6c954667
LM
3746/**
3747 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3748 * @q: the queue of the device
3749 * @dev: the device the queue belongs to
3750 *
3751 * Description:
3752 * Initialize runtime-PM-related fields for @q and start auto suspend for
3753 * @dev. Drivers that want to take advantage of request-based runtime PM
3754 * should call this function after @dev has been initialized, and its
3755 * request queue @q has been allocated, and runtime PM for it can not happen
3756 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3757 * cases, driver should call this function before any I/O has taken place.
3758 *
3759 * This function takes care of setting up using auto suspend for the device,
3760 * the autosuspend delay is set to -1 to make runtime suspend impossible
3761 * until an updated value is either set by user or by driver. Drivers do
3762 * not need to touch other autosuspend settings.
3763 *
3764 * The block layer runtime PM is request based, so only works for drivers
3765 * that use request as their IO unit instead of those directly use bio's.
3766 */
3767void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3768{
765e40b6
CH
3769 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3770 if (q->mq_ops)
3771 return;
3772
6c954667
LM
3773 q->dev = dev;
3774 q->rpm_status = RPM_ACTIVE;
3775 pm_runtime_set_autosuspend_delay(q->dev, -1);
3776 pm_runtime_use_autosuspend(q->dev);
3777}
3778EXPORT_SYMBOL(blk_pm_runtime_init);
3779
3780/**
3781 * blk_pre_runtime_suspend - Pre runtime suspend check
3782 * @q: the queue of the device
3783 *
3784 * Description:
3785 * This function will check if runtime suspend is allowed for the device
3786 * by examining if there are any requests pending in the queue. If there
3787 * are requests pending, the device can not be runtime suspended; otherwise,
3788 * the queue's status will be updated to SUSPENDING and the driver can
3789 * proceed to suspend the device.
3790 *
3791 * For the not allowed case, we mark last busy for the device so that
3792 * runtime PM core will try to autosuspend it some time later.
3793 *
3794 * This function should be called near the start of the device's
3795 * runtime_suspend callback.
3796 *
3797 * Return:
3798 * 0 - OK to runtime suspend the device
3799 * -EBUSY - Device should not be runtime suspended
3800 */
3801int blk_pre_runtime_suspend(struct request_queue *q)
3802{
3803 int ret = 0;
3804
4fd41a85
KX
3805 if (!q->dev)
3806 return ret;
3807
6c954667
LM
3808 spin_lock_irq(q->queue_lock);
3809 if (q->nr_pending) {
3810 ret = -EBUSY;
3811 pm_runtime_mark_last_busy(q->dev);
3812 } else {
3813 q->rpm_status = RPM_SUSPENDING;
3814 }
3815 spin_unlock_irq(q->queue_lock);
3816 return ret;
3817}
3818EXPORT_SYMBOL(blk_pre_runtime_suspend);
3819
3820/**
3821 * blk_post_runtime_suspend - Post runtime suspend processing
3822 * @q: the queue of the device
3823 * @err: return value of the device's runtime_suspend function
3824 *
3825 * Description:
3826 * Update the queue's runtime status according to the return value of the
3827 * device's runtime suspend function and mark last busy for the device so
3828 * that PM core will try to auto suspend the device at a later time.
3829 *
3830 * This function should be called near the end of the device's
3831 * runtime_suspend callback.
3832 */
3833void blk_post_runtime_suspend(struct request_queue *q, int err)
3834{
4fd41a85
KX
3835 if (!q->dev)
3836 return;
3837
6c954667
LM
3838 spin_lock_irq(q->queue_lock);
3839 if (!err) {
3840 q->rpm_status = RPM_SUSPENDED;
3841 } else {
3842 q->rpm_status = RPM_ACTIVE;
3843 pm_runtime_mark_last_busy(q->dev);
3844 }
3845 spin_unlock_irq(q->queue_lock);
3846}
3847EXPORT_SYMBOL(blk_post_runtime_suspend);
3848
3849/**
3850 * blk_pre_runtime_resume - Pre runtime resume processing
3851 * @q: the queue of the device
3852 *
3853 * Description:
3854 * Update the queue's runtime status to RESUMING in preparation for the
3855 * runtime resume of the device.
3856 *
3857 * This function should be called near the start of the device's
3858 * runtime_resume callback.
3859 */
3860void blk_pre_runtime_resume(struct request_queue *q)
3861{
4fd41a85
KX
3862 if (!q->dev)
3863 return;
3864
6c954667
LM
3865 spin_lock_irq(q->queue_lock);
3866 q->rpm_status = RPM_RESUMING;
3867 spin_unlock_irq(q->queue_lock);
3868}
3869EXPORT_SYMBOL(blk_pre_runtime_resume);
3870
3871/**
3872 * blk_post_runtime_resume - Post runtime resume processing
3873 * @q: the queue of the device
3874 * @err: return value of the device's runtime_resume function
3875 *
3876 * Description:
3877 * Update the queue's runtime status according to the return value of the
3878 * device's runtime_resume function. If it is successfully resumed, process
3879 * the requests that are queued into the device's queue when it is resuming
3880 * and then mark last busy and initiate autosuspend for it.
3881 *
3882 * This function should be called near the end of the device's
3883 * runtime_resume callback.
3884 */
3885void blk_post_runtime_resume(struct request_queue *q, int err)
3886{
4fd41a85
KX
3887 if (!q->dev)
3888 return;
3889
6c954667
LM
3890 spin_lock_irq(q->queue_lock);
3891 if (!err) {
3892 q->rpm_status = RPM_ACTIVE;
3893 __blk_run_queue(q);
3894 pm_runtime_mark_last_busy(q->dev);
c60855cd 3895 pm_request_autosuspend(q->dev);
6c954667
LM
3896 } else {
3897 q->rpm_status = RPM_SUSPENDED;
3898 }
3899 spin_unlock_irq(q->queue_lock);
3900}
3901EXPORT_SYMBOL(blk_post_runtime_resume);
d07ab6d1
MW
3902
3903/**
3904 * blk_set_runtime_active - Force runtime status of the queue to be active
3905 * @q: the queue of the device
3906 *
3907 * If the device is left runtime suspended during system suspend the resume
3908 * hook typically resumes the device and corrects runtime status
3909 * accordingly. However, that does not affect the queue runtime PM status
3910 * which is still "suspended". This prevents processing requests from the
3911 * queue.
3912 *
3913 * This function can be used in driver's resume hook to correct queue
3914 * runtime PM status and re-enable peeking requests from the queue. It
3915 * should be called before first request is added to the queue.
3916 */
3917void blk_set_runtime_active(struct request_queue *q)
3918{
3919 spin_lock_irq(q->queue_lock);
3920 q->rpm_status = RPM_ACTIVE;
3921 pm_runtime_mark_last_busy(q->dev);
3922 pm_request_autosuspend(q->dev);
3923 spin_unlock_irq(q->queue_lock);
3924}
3925EXPORT_SYMBOL(blk_set_runtime_active);
6c954667
LM
3926#endif
3927
1da177e4
LT
3928int __init blk_dev_init(void)
3929{
ef295ecf
CH
3930 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3931 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
0762b23d 3932 FIELD_SIZEOF(struct request, cmd_flags));
ef295ecf
CH
3933 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3934 FIELD_SIZEOF(struct bio, bi_opf));
9eb55b03 3935
89b90be2
TH
3936 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3937 kblockd_workqueue = alloc_workqueue("kblockd",
28747fcd 3938 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1da177e4
LT
3939 if (!kblockd_workqueue)
3940 panic("Failed to create kblockd\n");
3941
3942 request_cachep = kmem_cache_create("blkdev_requests",
20c2df83 3943 sizeof(struct request), 0, SLAB_PANIC, NULL);
1da177e4 3944
c2789bd4 3945 blk_requestq_cachep = kmem_cache_create("request_queue",
165125e1 3946 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1da177e4 3947
18fbda91
OS
3948#ifdef CONFIG_DEBUG_FS
3949 blk_debugfs_root = debugfs_create_dir("block", NULL);
3950#endif
3951
d38ecf93 3952 return 0;
1da177e4 3953}