block, bfq: prevent soft_rt_next_start from being stuck at infinity
[linux-block.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 */
0ba99ca4
KO
276 /*
277 * XXX this code looks suspicious - it's not consistent with advancing
278 * req->bio in caller
279 */
e8064021 280 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
4246a0b6 281 bio_endio(bio);
1da177e4 282}
1da177e4 283
1da177e4
LT
284void blk_dump_rq_flags(struct request *rq, char *msg)
285{
aebf526b
CH
286 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
287 rq->rq_disk ? rq->rq_disk->disk_name : "?",
5953316d 288 (unsigned long long) rq->cmd_flags);
1da177e4 289
83096ebf
TH
290 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
291 (unsigned long long)blk_rq_pos(rq),
292 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
b4f42e28
JA
293 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
294 rq->bio, rq->biotail, blk_rq_bytes(rq));
1da177e4 295}
1da177e4
LT
296EXPORT_SYMBOL(blk_dump_rq_flags);
297
3cca6dc1 298static void blk_delay_work(struct work_struct *work)
1da177e4 299{
3cca6dc1 300 struct request_queue *q;
1da177e4 301
3cca6dc1
JA
302 q = container_of(work, struct request_queue, delay_work.work);
303 spin_lock_irq(q->queue_lock);
24ecfbe2 304 __blk_run_queue(q);
3cca6dc1 305 spin_unlock_irq(q->queue_lock);
1da177e4 306}
1da177e4
LT
307
308/**
3cca6dc1
JA
309 * blk_delay_queue - restart queueing after defined interval
310 * @q: The &struct request_queue in question
311 * @msecs: Delay in msecs
1da177e4
LT
312 *
313 * Description:
3cca6dc1
JA
314 * Sometimes queueing needs to be postponed for a little while, to allow
315 * resources to come back. This function will make sure that queueing is
2fff8a92 316 * restarted around the specified time.
3cca6dc1
JA
317 */
318void blk_delay_queue(struct request_queue *q, unsigned long msecs)
2ad8b1ef 319{
2fff8a92 320 lockdep_assert_held(q->queue_lock);
332ebbf7 321 WARN_ON_ONCE(q->mq_ops);
2fff8a92 322
70460571
BVA
323 if (likely(!blk_queue_dead(q)))
324 queue_delayed_work(kblockd_workqueue, &q->delay_work,
325 msecs_to_jiffies(msecs));
2ad8b1ef 326}
3cca6dc1 327EXPORT_SYMBOL(blk_delay_queue);
2ad8b1ef 328
21491412
JA
329/**
330 * blk_start_queue_async - asynchronously restart a previously stopped queue
331 * @q: The &struct request_queue in question
332 *
333 * Description:
334 * blk_start_queue_async() will clear the stop flag on the queue, and
335 * ensure that the request_fn for the queue is run from an async
336 * context.
337 **/
338void blk_start_queue_async(struct request_queue *q)
339{
2fff8a92 340 lockdep_assert_held(q->queue_lock);
332ebbf7 341 WARN_ON_ONCE(q->mq_ops);
2fff8a92 342
21491412
JA
343 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
344 blk_run_queue_async(q);
345}
346EXPORT_SYMBOL(blk_start_queue_async);
347
1da177e4
LT
348/**
349 * blk_start_queue - restart a previously stopped queue
165125e1 350 * @q: The &struct request_queue in question
1da177e4
LT
351 *
352 * Description:
353 * blk_start_queue() will clear the stop flag on the queue, and call
354 * the request_fn for the queue if it was in a stopped state when
2fff8a92 355 * entered. Also see blk_stop_queue().
1da177e4 356 **/
165125e1 357void blk_start_queue(struct request_queue *q)
1da177e4 358{
2fff8a92 359 lockdep_assert_held(q->queue_lock);
332ebbf7 360 WARN_ON_ONCE(q->mq_ops);
a038e253 361
75ad23bc 362 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
24ecfbe2 363 __blk_run_queue(q);
1da177e4 364}
1da177e4
LT
365EXPORT_SYMBOL(blk_start_queue);
366
367/**
368 * blk_stop_queue - stop a queue
165125e1 369 * @q: The &struct request_queue in question
1da177e4
LT
370 *
371 * Description:
372 * The Linux block layer assumes that a block driver will consume all
373 * entries on the request queue when the request_fn strategy is called.
374 * Often this will not happen, because of hardware limitations (queue
375 * depth settings). If a device driver gets a 'queue full' response,
376 * or if it simply chooses not to queue more I/O at one point, it can
377 * call this function to prevent the request_fn from being called until
378 * the driver has signalled it's ready to go again. This happens by calling
2fff8a92 379 * blk_start_queue() to restart queue operations.
1da177e4 380 **/
165125e1 381void blk_stop_queue(struct request_queue *q)
1da177e4 382{
2fff8a92 383 lockdep_assert_held(q->queue_lock);
332ebbf7 384 WARN_ON_ONCE(q->mq_ops);
2fff8a92 385
136b5721 386 cancel_delayed_work(&q->delay_work);
75ad23bc 387 queue_flag_set(QUEUE_FLAG_STOPPED, q);
1da177e4
LT
388}
389EXPORT_SYMBOL(blk_stop_queue);
390
391/**
392 * blk_sync_queue - cancel any pending callbacks on a queue
393 * @q: the queue
394 *
395 * Description:
396 * The block layer may perform asynchronous callback activity
397 * on a queue, such as calling the unplug function after a timeout.
398 * A block device may call blk_sync_queue to ensure that any
399 * such activity is cancelled, thus allowing it to release resources
59c51591 400 * that the callbacks might use. The caller must already have made sure
1da177e4
LT
401 * that its ->make_request_fn will not re-add plugging prior to calling
402 * this function.
403 *
da527770 404 * This function does not cancel any asynchronous activity arising
da3dae54 405 * out of elevator or throttling code. That would require elevator_exit()
5efd6113 406 * and blkcg_exit_queue() to be called with queue lock initialized.
da527770 407 *
1da177e4
LT
408 */
409void blk_sync_queue(struct request_queue *q)
410{
70ed28b9 411 del_timer_sync(&q->timeout);
4e9b6f20 412 cancel_work_sync(&q->timeout_work);
f04c1fe7
ML
413
414 if (q->mq_ops) {
415 struct blk_mq_hw_ctx *hctx;
416 int i;
417
aba7afc5 418 cancel_delayed_work_sync(&q->requeue_work);
21c6e939 419 queue_for_each_hw_ctx(q, hctx, i)
9f993737 420 cancel_delayed_work_sync(&hctx->run_work);
f04c1fe7
ML
421 } else {
422 cancel_delayed_work_sync(&q->delay_work);
423 }
1da177e4
LT
424}
425EXPORT_SYMBOL(blk_sync_queue);
426
c9254f2d
BVA
427/**
428 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
429 * @q: request queue pointer
430 *
431 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
432 * set and 1 if the flag was already set.
433 */
434int blk_set_preempt_only(struct request_queue *q)
435{
8814ce8a 436 return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
c9254f2d
BVA
437}
438EXPORT_SYMBOL_GPL(blk_set_preempt_only);
439
440void blk_clear_preempt_only(struct request_queue *q)
441{
8814ce8a 442 blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
3a0a5299 443 wake_up_all(&q->mq_freeze_wq);
c9254f2d
BVA
444}
445EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
446
c246e80d
BVA
447/**
448 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
449 * @q: The queue to run
450 *
451 * Description:
452 * Invoke request handling on a queue if there are any pending requests.
453 * May be used to restart request handling after a request has completed.
454 * This variant runs the queue whether or not the queue has been
455 * stopped. Must be called with the queue lock held and interrupts
456 * disabled. See also @blk_run_queue.
457 */
458inline void __blk_run_queue_uncond(struct request_queue *q)
459{
2fff8a92 460 lockdep_assert_held(q->queue_lock);
332ebbf7 461 WARN_ON_ONCE(q->mq_ops);
2fff8a92 462
c246e80d
BVA
463 if (unlikely(blk_queue_dead(q)))
464 return;
465
24faf6f6
BVA
466 /*
467 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
468 * the queue lock internally. As a result multiple threads may be
469 * running such a request function concurrently. Keep track of the
470 * number of active request_fn invocations such that blk_drain_queue()
471 * can wait until all these request_fn calls have finished.
472 */
473 q->request_fn_active++;
c246e80d 474 q->request_fn(q);
24faf6f6 475 q->request_fn_active--;
c246e80d 476}
a7928c15 477EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
c246e80d 478
1da177e4 479/**
80a4b58e 480 * __blk_run_queue - run a single device queue
1da177e4 481 * @q: The queue to run
80a4b58e
JA
482 *
483 * Description:
2fff8a92 484 * See @blk_run_queue.
1da177e4 485 */
24ecfbe2 486void __blk_run_queue(struct request_queue *q)
1da177e4 487{
2fff8a92 488 lockdep_assert_held(q->queue_lock);
332ebbf7 489 WARN_ON_ONCE(q->mq_ops);
2fff8a92 490
a538cd03
TH
491 if (unlikely(blk_queue_stopped(q)))
492 return;
493
c246e80d 494 __blk_run_queue_uncond(q);
75ad23bc
NP
495}
496EXPORT_SYMBOL(__blk_run_queue);
dac07ec1 497
24ecfbe2
CH
498/**
499 * blk_run_queue_async - run a single device queue in workqueue context
500 * @q: The queue to run
501 *
502 * Description:
503 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
2fff8a92
BVA
504 * of us.
505 *
506 * Note:
507 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
508 * has canceled q->delay_work, callers must hold the queue lock to avoid
509 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
24ecfbe2
CH
510 */
511void blk_run_queue_async(struct request_queue *q)
512{
2fff8a92 513 lockdep_assert_held(q->queue_lock);
332ebbf7 514 WARN_ON_ONCE(q->mq_ops);
2fff8a92 515
70460571 516 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
e7c2f967 517 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
24ecfbe2 518}
c21e6beb 519EXPORT_SYMBOL(blk_run_queue_async);
24ecfbe2 520
75ad23bc
NP
521/**
522 * blk_run_queue - run a single device queue
523 * @q: The queue to run
80a4b58e
JA
524 *
525 * Description:
526 * Invoke request handling on this queue, if it has pending work to do.
a7f55792 527 * May be used to restart queueing when a request has completed.
75ad23bc
NP
528 */
529void blk_run_queue(struct request_queue *q)
530{
531 unsigned long flags;
532
332ebbf7
BVA
533 WARN_ON_ONCE(q->mq_ops);
534
75ad23bc 535 spin_lock_irqsave(q->queue_lock, flags);
24ecfbe2 536 __blk_run_queue(q);
1da177e4
LT
537 spin_unlock_irqrestore(q->queue_lock, flags);
538}
539EXPORT_SYMBOL(blk_run_queue);
540
165125e1 541void blk_put_queue(struct request_queue *q)
483f4afc
AV
542{
543 kobject_put(&q->kobj);
544}
d86e0e83 545EXPORT_SYMBOL(blk_put_queue);
483f4afc 546
e3c78ca5 547/**
807592a4 548 * __blk_drain_queue - drain requests from request_queue
e3c78ca5 549 * @q: queue to drain
c9a929dd 550 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
e3c78ca5 551 *
c9a929dd
TH
552 * Drain requests from @q. If @drain_all is set, all requests are drained.
553 * If not, only ELVPRIV requests are drained. The caller is responsible
554 * for ensuring that no new requests which need to be drained are queued.
e3c78ca5 555 */
807592a4
BVA
556static void __blk_drain_queue(struct request_queue *q, bool drain_all)
557 __releases(q->queue_lock)
558 __acquires(q->queue_lock)
e3c78ca5 559{
458f27a9
AH
560 int i;
561
807592a4 562 lockdep_assert_held(q->queue_lock);
332ebbf7 563 WARN_ON_ONCE(q->mq_ops);
807592a4 564
e3c78ca5 565 while (true) {
481a7d64 566 bool drain = false;
e3c78ca5 567
b855b04a
TH
568 /*
569 * The caller might be trying to drain @q before its
570 * elevator is initialized.
571 */
572 if (q->elevator)
573 elv_drain_elevator(q);
574
5efd6113 575 blkcg_drain_queue(q);
e3c78ca5 576
4eabc941
TH
577 /*
578 * This function might be called on a queue which failed
b855b04a
TH
579 * driver init after queue creation or is not yet fully
580 * active yet. Some drivers (e.g. fd and loop) get unhappy
581 * in such cases. Kick queue iff dispatch queue has
582 * something on it and @q has request_fn set.
4eabc941 583 */
b855b04a 584 if (!list_empty(&q->queue_head) && q->request_fn)
4eabc941 585 __blk_run_queue(q);
c9a929dd 586
8a5ecdd4 587 drain |= q->nr_rqs_elvpriv;
24faf6f6 588 drain |= q->request_fn_active;
481a7d64
TH
589
590 /*
591 * Unfortunately, requests are queued at and tracked from
592 * multiple places and there's no single counter which can
593 * be drained. Check all the queues and counters.
594 */
595 if (drain_all) {
e97c293c 596 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
481a7d64
TH
597 drain |= !list_empty(&q->queue_head);
598 for (i = 0; i < 2; i++) {
8a5ecdd4 599 drain |= q->nr_rqs[i];
481a7d64 600 drain |= q->in_flight[i];
7c94e1c1
ML
601 if (fq)
602 drain |= !list_empty(&fq->flush_queue[i]);
481a7d64
TH
603 }
604 }
e3c78ca5 605
481a7d64 606 if (!drain)
e3c78ca5 607 break;
807592a4
BVA
608
609 spin_unlock_irq(q->queue_lock);
610
e3c78ca5 611 msleep(10);
807592a4
BVA
612
613 spin_lock_irq(q->queue_lock);
e3c78ca5 614 }
458f27a9
AH
615
616 /*
617 * With queue marked dead, any woken up waiter will fail the
618 * allocation path, so the wakeup chaining is lost and we're
619 * left with hung waiters. We need to wake up those waiters.
620 */
621 if (q->request_fn) {
a051661c
TH
622 struct request_list *rl;
623
a051661c
TH
624 blk_queue_for_each_rl(rl, q)
625 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
626 wake_up_all(&rl->wait[i]);
458f27a9 627 }
e3c78ca5
TH
628}
629
454be724
ML
630void blk_drain_queue(struct request_queue *q)
631{
632 spin_lock_irq(q->queue_lock);
633 __blk_drain_queue(q, true);
634 spin_unlock_irq(q->queue_lock);
635}
636
d732580b
TH
637/**
638 * blk_queue_bypass_start - enter queue bypass mode
639 * @q: queue of interest
640 *
641 * In bypass mode, only the dispatch FIFO queue of @q is used. This
642 * function makes @q enter bypass mode and drains all requests which were
6ecf23af 643 * throttled or issued before. On return, it's guaranteed that no request
80fd9979
TH
644 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
645 * inside queue or RCU read lock.
d732580b
TH
646 */
647void blk_queue_bypass_start(struct request_queue *q)
648{
332ebbf7
BVA
649 WARN_ON_ONCE(q->mq_ops);
650
d732580b 651 spin_lock_irq(q->queue_lock);
776687bc 652 q->bypass_depth++;
d732580b
TH
653 queue_flag_set(QUEUE_FLAG_BYPASS, q);
654 spin_unlock_irq(q->queue_lock);
655
776687bc
TH
656 /*
657 * Queues start drained. Skip actual draining till init is
658 * complete. This avoids lenghty delays during queue init which
659 * can happen many times during boot.
660 */
661 if (blk_queue_init_done(q)) {
807592a4
BVA
662 spin_lock_irq(q->queue_lock);
663 __blk_drain_queue(q, false);
664 spin_unlock_irq(q->queue_lock);
665
b82d4b19
TH
666 /* ensure blk_queue_bypass() is %true inside RCU read lock */
667 synchronize_rcu();
668 }
d732580b
TH
669}
670EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
671
672/**
673 * blk_queue_bypass_end - leave queue bypass mode
674 * @q: queue of interest
675 *
676 * Leave bypass mode and restore the normal queueing behavior.
332ebbf7
BVA
677 *
678 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
679 * this function is called for both blk-sq and blk-mq queues.
d732580b
TH
680 */
681void blk_queue_bypass_end(struct request_queue *q)
682{
683 spin_lock_irq(q->queue_lock);
684 if (!--q->bypass_depth)
685 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
686 WARN_ON_ONCE(q->bypass_depth < 0);
687 spin_unlock_irq(q->queue_lock);
688}
689EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
690
aed3ea94
JA
691void blk_set_queue_dying(struct request_queue *q)
692{
8814ce8a 693 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
aed3ea94 694
d3cfb2a0
ML
695 /*
696 * When queue DYING flag is set, we need to block new req
697 * entering queue, so we call blk_freeze_queue_start() to
698 * prevent I/O from crossing blk_queue_enter().
699 */
700 blk_freeze_queue_start(q);
701
aed3ea94
JA
702 if (q->mq_ops)
703 blk_mq_wake_waiters(q);
704 else {
705 struct request_list *rl;
706
bbfc3c5d 707 spin_lock_irq(q->queue_lock);
aed3ea94
JA
708 blk_queue_for_each_rl(rl, q) {
709 if (rl->rq_pool) {
34d9715a
ML
710 wake_up_all(&rl->wait[BLK_RW_SYNC]);
711 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
aed3ea94
JA
712 }
713 }
bbfc3c5d 714 spin_unlock_irq(q->queue_lock);
aed3ea94 715 }
055f6e18
ML
716
717 /* Make blk_queue_enter() reexamine the DYING flag. */
718 wake_up_all(&q->mq_freeze_wq);
aed3ea94
JA
719}
720EXPORT_SYMBOL_GPL(blk_set_queue_dying);
721
c9a929dd
TH
722/**
723 * blk_cleanup_queue - shutdown a request queue
724 * @q: request queue to shutdown
725 *
c246e80d
BVA
726 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
727 * put it. All future requests will be failed immediately with -ENODEV.
c94a96ac 728 */
6728cb0e 729void blk_cleanup_queue(struct request_queue *q)
483f4afc 730{
c9a929dd 731 spinlock_t *lock = q->queue_lock;
e3335de9 732
3f3299d5 733 /* mark @q DYING, no new request or merges will be allowed afterwards */
483f4afc 734 mutex_lock(&q->sysfs_lock);
aed3ea94 735 blk_set_queue_dying(q);
c9a929dd 736 spin_lock_irq(lock);
6ecf23af 737
80fd9979 738 /*
3f3299d5 739 * A dying queue is permanently in bypass mode till released. Note
80fd9979
TH
740 * that, unlike blk_queue_bypass_start(), we aren't performing
741 * synchronize_rcu() after entering bypass mode to avoid the delay
742 * as some drivers create and destroy a lot of queues while
743 * probing. This is still safe because blk_release_queue() will be
744 * called only after the queue refcnt drops to zero and nothing,
745 * RCU or not, would be traversing the queue by then.
746 */
6ecf23af
TH
747 q->bypass_depth++;
748 queue_flag_set(QUEUE_FLAG_BYPASS, q);
749
c9a929dd
TH
750 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
751 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
3f3299d5 752 queue_flag_set(QUEUE_FLAG_DYING, q);
c9a929dd
TH
753 spin_unlock_irq(lock);
754 mutex_unlock(&q->sysfs_lock);
755
c246e80d
BVA
756 /*
757 * Drain all requests queued before DYING marking. Set DEAD flag to
758 * prevent that q->request_fn() gets invoked after draining finished.
759 */
3ef28e83 760 blk_freeze_queue(q);
9c1051aa 761 spin_lock_irq(lock);
c246e80d 762 queue_flag_set(QUEUE_FLAG_DEAD, q);
807592a4 763 spin_unlock_irq(lock);
c9a929dd 764
c2856ae2
ML
765 /*
766 * make sure all in-progress dispatch are completed because
767 * blk_freeze_queue() can only complete all requests, and
768 * dispatch may still be in-progress since we dispatch requests
769 * from more than one contexts
770 */
771 if (q->mq_ops)
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
00380a40 1002 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
a73f730d 1003 if (q->id < 0)
3d2936f4 1004 goto fail_q;
a73f730d 1005
338aa96d
KO
1006 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1007 if (ret)
54efd50b
KO
1008 goto fail_id;
1009
d03f6cdc
JK
1010 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1011 if (!q->backing_dev_info)
1012 goto fail_split;
1013
a83b576c
JA
1014 q->stats = blk_alloc_queue_stats();
1015 if (!q->stats)
1016 goto fail_stats;
1017
dc3b17cc 1018 q->backing_dev_info->ra_pages =
09cbfeaf 1019 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
dc3b17cc
JK
1020 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1021 q->backing_dev_info->name = "block";
5151412d 1022 q->node = node_id;
0989a025 1023
bca237a5
KC
1024 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1025 laptop_mode_timer_fn, 0);
1026 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
4e9b6f20 1027 INIT_WORK(&q->timeout_work, NULL);
b855b04a 1028 INIT_LIST_HEAD(&q->queue_head);
242f9dcb 1029 INIT_LIST_HEAD(&q->timeout_list);
a612fddf 1030 INIT_LIST_HEAD(&q->icq_list);
4eef3049 1031#ifdef CONFIG_BLK_CGROUP
e8989fae 1032 INIT_LIST_HEAD(&q->blkg_list);
4eef3049 1033#endif
3cca6dc1 1034 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
483f4afc 1035
8324aa91 1036 kobject_init(&q->kobj, &blk_queue_ktype);
1da177e4 1037
5acb3cc2
WL
1038#ifdef CONFIG_BLK_DEV_IO_TRACE
1039 mutex_init(&q->blk_trace_mutex);
1040#endif
483f4afc 1041 mutex_init(&q->sysfs_lock);
e7e72bf6 1042 spin_lock_init(&q->__queue_lock);
483f4afc 1043
498f6650
BVA
1044 if (!q->mq_ops)
1045 q->queue_lock = lock ? : &q->__queue_lock;
c94a96ac 1046
b82d4b19
TH
1047 /*
1048 * A queue starts its life with bypass turned on to avoid
1049 * unnecessary bypass on/off overhead and nasty surprises during
749fefe6
TH
1050 * init. The initial bypass will be finished when the queue is
1051 * registered by blk_register_queue().
b82d4b19
TH
1052 */
1053 q->bypass_depth = 1;
f78bac2c 1054 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
b82d4b19 1055
320ae51f
JA
1056 init_waitqueue_head(&q->mq_freeze_wq);
1057
3ef28e83
DW
1058 /*
1059 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1060 * See blk_register_queue() for details.
1061 */
1062 if (percpu_ref_init(&q->q_usage_counter,
1063 blk_queue_usage_counter_release,
1064 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
fff4996b 1065 goto fail_bdi;
f51b802c 1066
3ef28e83
DW
1067 if (blkcg_init_queue(q))
1068 goto fail_ref;
1069
1da177e4 1070 return q;
a73f730d 1071
3ef28e83
DW
1072fail_ref:
1073 percpu_ref_exit(&q->q_usage_counter);
fff4996b 1074fail_bdi:
a83b576c
JA
1075 blk_free_queue_stats(q->stats);
1076fail_stats:
d03f6cdc 1077 bdi_put(q->backing_dev_info);
54efd50b 1078fail_split:
338aa96d 1079 bioset_exit(&q->bio_split);
a73f730d
TH
1080fail_id:
1081 ida_simple_remove(&blk_queue_ida, q->id);
1082fail_q:
1083 kmem_cache_free(blk_requestq_cachep, q);
1084 return NULL;
1da177e4 1085}
1946089a 1086EXPORT_SYMBOL(blk_alloc_queue_node);
1da177e4
LT
1087
1088/**
1089 * blk_init_queue - prepare a request queue for use with a block device
1090 * @rfn: The function to be called to process requests that have been
1091 * placed on the queue.
1092 * @lock: Request queue spin lock
1093 *
1094 * Description:
1095 * If a block device wishes to use the standard request handling procedures,
1096 * which sorts requests and coalesces adjacent requests, then it must
1097 * call blk_init_queue(). The function @rfn will be called when there
1098 * are requests on the queue that need to be processed. If the device
1099 * supports plugging, then @rfn may not be called immediately when requests
1100 * are available on the queue, but may be called at some time later instead.
1101 * Plugged queues are generally unplugged when a buffer belonging to one
1102 * of the requests on the queue is needed, or due to memory pressure.
1103 *
1104 * @rfn is not required, or even expected, to remove all requests off the
1105 * queue, but only as many as it can handle at a time. If it does leave
1106 * requests on the queue, it is responsible for arranging that the requests
1107 * get dealt with eventually.
1108 *
1109 * The queue spin lock must be held while manipulating the requests on the
a038e253
PBG
1110 * request queue; this lock will be taken also from interrupt context, so irq
1111 * disabling is needed for it.
1da177e4 1112 *
710027a4 1113 * Function returns a pointer to the initialized request queue, or %NULL if
1da177e4
LT
1114 * it didn't succeed.
1115 *
1116 * Note:
1117 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1118 * when the block device is deactivated (such as at module unload).
1119 **/
1946089a 1120
165125e1 1121struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1da177e4 1122{
c304a51b 1123 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1946089a
CL
1124}
1125EXPORT_SYMBOL(blk_init_queue);
1126
165125e1 1127struct request_queue *
1946089a
CL
1128blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1129{
5ea708d1 1130 struct request_queue *q;
1da177e4 1131
498f6650 1132 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
5ea708d1 1133 if (!q)
c86d1b8a
MS
1134 return NULL;
1135
5ea708d1 1136 q->request_fn = rfn;
5ea708d1
CH
1137 if (blk_init_allocated_queue(q) < 0) {
1138 blk_cleanup_queue(q);
1139 return NULL;
1140 }
18741986 1141
7982e90c 1142 return q;
01effb0d
MS
1143}
1144EXPORT_SYMBOL(blk_init_queue_node);
1145
dece1635 1146static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
336b7e1f 1147
1da177e4 1148
5ea708d1
CH
1149int blk_init_allocated_queue(struct request_queue *q)
1150{
332ebbf7
BVA
1151 WARN_ON_ONCE(q->mq_ops);
1152
6d247d7f 1153 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
ba483388 1154 if (!q->fq)
5ea708d1 1155 return -ENOMEM;
7982e90c 1156
6d247d7f
CH
1157 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1158 goto out_free_flush_queue;
7982e90c 1159
a051661c 1160 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
6d247d7f 1161 goto out_exit_flush_rq;
1da177e4 1162
287922eb 1163 INIT_WORK(&q->timeout_work, blk_timeout_work);
60ea8226 1164 q->queue_flags |= QUEUE_FLAG_DEFAULT;
c94a96ac 1165
f3b144aa
JA
1166 /*
1167 * This also sets hw/phys segments, boundary and size
1168 */
c20e8de2 1169 blk_queue_make_request(q, blk_queue_bio);
1da177e4 1170
44ec9542
AS
1171 q->sg_reserved_size = INT_MAX;
1172
eb1c160b
TS
1173 /* Protect q->elevator from elevator_change */
1174 mutex_lock(&q->sysfs_lock);
1175
b82d4b19 1176 /* init elevator */
eb1c160b
TS
1177 if (elevator_init(q, NULL)) {
1178 mutex_unlock(&q->sysfs_lock);
6d247d7f 1179 goto out_exit_flush_rq;
eb1c160b
TS
1180 }
1181
1182 mutex_unlock(&q->sysfs_lock);
5ea708d1 1183 return 0;
eb1c160b 1184
6d247d7f
CH
1185out_exit_flush_rq:
1186 if (q->exit_rq_fn)
1187 q->exit_rq_fn(q, q->fq->flush_rq);
1188out_free_flush_queue:
ba483388 1189 blk_free_flush_queue(q->fq);
5ea708d1 1190 return -ENOMEM;
1da177e4 1191}
5151412d 1192EXPORT_SYMBOL(blk_init_allocated_queue);
1da177e4 1193
09ac46c4 1194bool blk_get_queue(struct request_queue *q)
1da177e4 1195{
3f3299d5 1196 if (likely(!blk_queue_dying(q))) {
09ac46c4
TH
1197 __blk_get_queue(q);
1198 return true;
1da177e4
LT
1199 }
1200
09ac46c4 1201 return false;
1da177e4 1202}
d86e0e83 1203EXPORT_SYMBOL(blk_get_queue);
1da177e4 1204
5b788ce3 1205static inline void blk_free_request(struct request_list *rl, struct request *rq)
1da177e4 1206{
e8064021 1207 if (rq->rq_flags & RQF_ELVPRIV) {
5b788ce3 1208 elv_put_request(rl->q, rq);
f1f8cc94 1209 if (rq->elv.icq)
11a3122f 1210 put_io_context(rq->elv.icq->ioc);
f1f8cc94
TH
1211 }
1212
5b788ce3 1213 mempool_free(rq, rl->rq_pool);
1da177e4
LT
1214}
1215
1da177e4
LT
1216/*
1217 * ioc_batching returns true if the ioc is a valid batching request and
1218 * should be given priority access to a request.
1219 */
165125e1 1220static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1221{
1222 if (!ioc)
1223 return 0;
1224
1225 /*
1226 * Make sure the process is able to allocate at least 1 request
1227 * even if the batch times out, otherwise we could theoretically
1228 * lose wakeups.
1229 */
1230 return ioc->nr_batch_requests == q->nr_batching ||
1231 (ioc->nr_batch_requests > 0
1232 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1233}
1234
1235/*
1236 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1237 * will cause the process to be a "batcher" on all queues in the system. This
1238 * is the behaviour we want though - once it gets a wakeup it should be given
1239 * a nice run.
1240 */
165125e1 1241static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1242{
1243 if (!ioc || ioc_batching(q, ioc))
1244 return;
1245
1246 ioc->nr_batch_requests = q->nr_batching;
1247 ioc->last_waited = jiffies;
1248}
1249
5b788ce3 1250static void __freed_request(struct request_list *rl, int sync)
1da177e4 1251{
5b788ce3 1252 struct request_queue *q = rl->q;
1da177e4 1253
d40f75a0
TH
1254 if (rl->count[sync] < queue_congestion_off_threshold(q))
1255 blk_clear_congested(rl, sync);
1da177e4 1256
1faa16d2
JA
1257 if (rl->count[sync] + 1 <= q->nr_requests) {
1258 if (waitqueue_active(&rl->wait[sync]))
1259 wake_up(&rl->wait[sync]);
1da177e4 1260
5b788ce3 1261 blk_clear_rl_full(rl, sync);
1da177e4
LT
1262 }
1263}
1264
1265/*
1266 * A request has just been released. Account for it, update the full and
1267 * congestion status, wake up any waiters. Called under q->queue_lock.
1268 */
e8064021
CH
1269static void freed_request(struct request_list *rl, bool sync,
1270 req_flags_t rq_flags)
1da177e4 1271{
5b788ce3 1272 struct request_queue *q = rl->q;
1da177e4 1273
8a5ecdd4 1274 q->nr_rqs[sync]--;
1faa16d2 1275 rl->count[sync]--;
e8064021 1276 if (rq_flags & RQF_ELVPRIV)
8a5ecdd4 1277 q->nr_rqs_elvpriv--;
1da177e4 1278
5b788ce3 1279 __freed_request(rl, sync);
1da177e4 1280
1faa16d2 1281 if (unlikely(rl->starved[sync ^ 1]))
5b788ce3 1282 __freed_request(rl, sync ^ 1);
1da177e4
LT
1283}
1284
e3a2b3f9
JA
1285int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1286{
1287 struct request_list *rl;
d40f75a0 1288 int on_thresh, off_thresh;
e3a2b3f9 1289
332ebbf7
BVA
1290 WARN_ON_ONCE(q->mq_ops);
1291
e3a2b3f9
JA
1292 spin_lock_irq(q->queue_lock);
1293 q->nr_requests = nr;
1294 blk_queue_congestion_threshold(q);
d40f75a0
TH
1295 on_thresh = queue_congestion_on_threshold(q);
1296 off_thresh = queue_congestion_off_threshold(q);
e3a2b3f9 1297
d40f75a0
TH
1298 blk_queue_for_each_rl(rl, q) {
1299 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1300 blk_set_congested(rl, BLK_RW_SYNC);
1301 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1302 blk_clear_congested(rl, BLK_RW_SYNC);
e3a2b3f9 1303
d40f75a0
TH
1304 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1305 blk_set_congested(rl, BLK_RW_ASYNC);
1306 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1307 blk_clear_congested(rl, BLK_RW_ASYNC);
e3a2b3f9 1308
e3a2b3f9
JA
1309 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1310 blk_set_rl_full(rl, BLK_RW_SYNC);
1311 } else {
1312 blk_clear_rl_full(rl, BLK_RW_SYNC);
1313 wake_up(&rl->wait[BLK_RW_SYNC]);
1314 }
1315
1316 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1317 blk_set_rl_full(rl, BLK_RW_ASYNC);
1318 } else {
1319 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1320 wake_up(&rl->wait[BLK_RW_ASYNC]);
1321 }
1322 }
1323
1324 spin_unlock_irq(q->queue_lock);
1325 return 0;
1326}
1327
da8303c6 1328/**
a06e05e6 1329 * __get_request - get a free request
5b788ce3 1330 * @rl: request list to allocate from
ef295ecf 1331 * @op: operation and flags
da8303c6 1332 * @bio: bio to allocate request for (can be %NULL)
6a15674d 1333 * @flags: BLQ_MQ_REQ_* flags
4accf5fc 1334 * @gfp_mask: allocator flags
da8303c6
TH
1335 *
1336 * Get a free request from @q. This function may fail under memory
1337 * pressure or if @q is dead.
1338 *
da3dae54 1339 * Must be called with @q->queue_lock held and,
a492f075
JL
1340 * Returns ERR_PTR on failure, with @q->queue_lock held.
1341 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1342 */
ef295ecf 1343static struct request *__get_request(struct request_list *rl, unsigned int op,
4accf5fc 1344 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1da177e4 1345{
5b788ce3 1346 struct request_queue *q = rl->q;
b679281a 1347 struct request *rq;
7f4b35d1
TH
1348 struct elevator_type *et = q->elevator->type;
1349 struct io_context *ioc = rq_ioc(bio);
f1f8cc94 1350 struct io_cq *icq = NULL;
ef295ecf 1351 const bool is_sync = op_is_sync(op);
75eb6c37 1352 int may_queue;
e8064021 1353 req_flags_t rq_flags = RQF_ALLOCED;
88ee5ef1 1354
2fff8a92
BVA
1355 lockdep_assert_held(q->queue_lock);
1356
3f3299d5 1357 if (unlikely(blk_queue_dying(q)))
a492f075 1358 return ERR_PTR(-ENODEV);
da8303c6 1359
ef295ecf 1360 may_queue = elv_may_queue(q, op);
88ee5ef1
JA
1361 if (may_queue == ELV_MQUEUE_NO)
1362 goto rq_starved;
1363
1faa16d2
JA
1364 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1365 if (rl->count[is_sync]+1 >= q->nr_requests) {
88ee5ef1
JA
1366 /*
1367 * The queue will fill after this allocation, so set
1368 * it as full, and mark this process as "batching".
1369 * This process will be allowed to complete a batch of
1370 * requests, others will be blocked.
1371 */
5b788ce3 1372 if (!blk_rl_full(rl, is_sync)) {
88ee5ef1 1373 ioc_set_batching(q, ioc);
5b788ce3 1374 blk_set_rl_full(rl, is_sync);
88ee5ef1
JA
1375 } else {
1376 if (may_queue != ELV_MQUEUE_MUST
1377 && !ioc_batching(q, ioc)) {
1378 /*
1379 * The queue is full and the allocating
1380 * process is not a "batcher", and not
1381 * exempted by the IO scheduler
1382 */
a492f075 1383 return ERR_PTR(-ENOMEM);
88ee5ef1
JA
1384 }
1385 }
1da177e4 1386 }
d40f75a0 1387 blk_set_congested(rl, is_sync);
1da177e4
LT
1388 }
1389
082cf69e
JA
1390 /*
1391 * Only allow batching queuers to allocate up to 50% over the defined
1392 * limit of requests, otherwise we could have thousands of requests
1393 * allocated with any setting of ->nr_requests
1394 */
1faa16d2 1395 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
a492f075 1396 return ERR_PTR(-ENOMEM);
fd782a4a 1397
8a5ecdd4 1398 q->nr_rqs[is_sync]++;
1faa16d2
JA
1399 rl->count[is_sync]++;
1400 rl->starved[is_sync] = 0;
cb98fc8b 1401
f1f8cc94
TH
1402 /*
1403 * Decide whether the new request will be managed by elevator. If
e8064021 1404 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
f1f8cc94
TH
1405 * prevent the current elevator from being destroyed until the new
1406 * request is freed. This guarantees icq's won't be destroyed and
1407 * makes creating new ones safe.
1408 *
e6f7f93d
CH
1409 * Flush requests do not use the elevator so skip initialization.
1410 * This allows a request to share the flush and elevator data.
1411 *
f1f8cc94
TH
1412 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1413 * it will be created after releasing queue_lock.
1414 */
e6f7f93d 1415 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
e8064021 1416 rq_flags |= RQF_ELVPRIV;
8a5ecdd4 1417 q->nr_rqs_elvpriv++;
f1f8cc94
TH
1418 if (et->icq_cache && ioc)
1419 icq = ioc_lookup_icq(ioc, q);
9d5a4e94 1420 }
cb98fc8b 1421
f253b86b 1422 if (blk_queue_io_stat(q))
e8064021 1423 rq_flags |= RQF_IO_STAT;
1da177e4
LT
1424 spin_unlock_irq(q->queue_lock);
1425
29e2b09a 1426 /* allocate and init request */
5b788ce3 1427 rq = mempool_alloc(rl->rq_pool, gfp_mask);
29e2b09a 1428 if (!rq)
b679281a 1429 goto fail_alloc;
1da177e4 1430
29e2b09a 1431 blk_rq_init(q, rq);
a051661c 1432 blk_rq_set_rl(rq, rl);
ef295ecf 1433 rq->cmd_flags = op;
e8064021 1434 rq->rq_flags = rq_flags;
1b6d65a0
BVA
1435 if (flags & BLK_MQ_REQ_PREEMPT)
1436 rq->rq_flags |= RQF_PREEMPT;
29e2b09a 1437
aaf7c680 1438 /* init elvpriv */
e8064021 1439 if (rq_flags & RQF_ELVPRIV) {
aaf7c680 1440 if (unlikely(et->icq_cache && !icq)) {
7f4b35d1
TH
1441 if (ioc)
1442 icq = ioc_create_icq(ioc, q, gfp_mask);
aaf7c680
TH
1443 if (!icq)
1444 goto fail_elvpriv;
29e2b09a 1445 }
aaf7c680
TH
1446
1447 rq->elv.icq = icq;
1448 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1449 goto fail_elvpriv;
1450
1451 /* @rq->elv.icq holds io_context until @rq is freed */
29e2b09a
TH
1452 if (icq)
1453 get_io_context(icq->ioc);
1454 }
aaf7c680 1455out:
88ee5ef1
JA
1456 /*
1457 * ioc may be NULL here, and ioc_batching will be false. That's
1458 * OK, if the queue is under the request limit then requests need
1459 * not count toward the nr_batch_requests limit. There will always
1460 * be some limit enforced by BLK_BATCH_TIME.
1461 */
1da177e4
LT
1462 if (ioc_batching(q, ioc))
1463 ioc->nr_batch_requests--;
6728cb0e 1464
e6a40b09 1465 trace_block_getrq(q, bio, op);
1da177e4 1466 return rq;
b679281a 1467
aaf7c680
TH
1468fail_elvpriv:
1469 /*
1470 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1471 * and may fail indefinitely under memory pressure and thus
1472 * shouldn't stall IO. Treat this request as !elvpriv. This will
1473 * disturb iosched and blkcg but weird is bettern than dead.
1474 */
7b2b10e0 1475 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
dc3b17cc 1476 __func__, dev_name(q->backing_dev_info->dev));
aaf7c680 1477
e8064021 1478 rq->rq_flags &= ~RQF_ELVPRIV;
aaf7c680
TH
1479 rq->elv.icq = NULL;
1480
1481 spin_lock_irq(q->queue_lock);
8a5ecdd4 1482 q->nr_rqs_elvpriv--;
aaf7c680
TH
1483 spin_unlock_irq(q->queue_lock);
1484 goto out;
1485
b679281a
TH
1486fail_alloc:
1487 /*
1488 * Allocation failed presumably due to memory. Undo anything we
1489 * might have messed up.
1490 *
1491 * Allocating task should really be put onto the front of the wait
1492 * queue, but this is pretty rare.
1493 */
1494 spin_lock_irq(q->queue_lock);
e8064021 1495 freed_request(rl, is_sync, rq_flags);
b679281a
TH
1496
1497 /*
1498 * in the very unlikely event that allocation failed and no
1499 * requests for this direction was pending, mark us starved so that
1500 * freeing of a request in the other direction will notice
1501 * us. another possible fix would be to split the rq mempool into
1502 * READ and WRITE
1503 */
1504rq_starved:
1505 if (unlikely(rl->count[is_sync] == 0))
1506 rl->starved[is_sync] = 1;
a492f075 1507 return ERR_PTR(-ENOMEM);
1da177e4
LT
1508}
1509
da8303c6 1510/**
a06e05e6 1511 * get_request - get a free request
da8303c6 1512 * @q: request_queue to allocate request from
ef295ecf 1513 * @op: operation and flags
da8303c6 1514 * @bio: bio to allocate request for (can be %NULL)
6a15674d 1515 * @flags: BLK_MQ_REQ_* flags.
4accf5fc 1516 * @gfp: allocator flags
da8303c6 1517 *
a9a14d36 1518 * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags,
d0164adc 1519 * this function keeps retrying under memory pressure and fails iff @q is dead.
d6344532 1520 *
da3dae54 1521 * Must be called with @q->queue_lock held and,
a492f075
JL
1522 * Returns ERR_PTR on failure, with @q->queue_lock held.
1523 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1524 */
ef295ecf 1525static struct request *get_request(struct request_queue *q, unsigned int op,
4accf5fc 1526 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1da177e4 1527{
ef295ecf 1528 const bool is_sync = op_is_sync(op);
a06e05e6 1529 DEFINE_WAIT(wait);
a051661c 1530 struct request_list *rl;
1da177e4 1531 struct request *rq;
a051661c 1532
2fff8a92 1533 lockdep_assert_held(q->queue_lock);
332ebbf7 1534 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1535
a051661c 1536 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
a06e05e6 1537retry:
4accf5fc 1538 rq = __get_request(rl, op, bio, flags, gfp);
a492f075 1539 if (!IS_ERR(rq))
a06e05e6 1540 return rq;
1da177e4 1541
03a07c92
GR
1542 if (op & REQ_NOWAIT) {
1543 blk_put_rl(rl);
1544 return ERR_PTR(-EAGAIN);
1545 }
1546
6a15674d 1547 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
a051661c 1548 blk_put_rl(rl);
a492f075 1549 return rq;
a051661c 1550 }
1da177e4 1551
a06e05e6
TH
1552 /* wait on @rl and retry */
1553 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1554 TASK_UNINTERRUPTIBLE);
1da177e4 1555
e6a40b09 1556 trace_block_sleeprq(q, bio, op);
1da177e4 1557
a06e05e6
TH
1558 spin_unlock_irq(q->queue_lock);
1559 io_schedule();
d6344532 1560
a06e05e6
TH
1561 /*
1562 * After sleeping, we become a "batching" process and will be able
1563 * to allocate at least one request, and up to a big batch of them
1564 * for a small period time. See ioc_batching, ioc_set_batching
1565 */
a06e05e6 1566 ioc_set_batching(q, current->io_context);
05caf8db 1567
a06e05e6
TH
1568 spin_lock_irq(q->queue_lock);
1569 finish_wait(&rl->wait[is_sync], &wait);
1da177e4 1570
a06e05e6 1571 goto retry;
1da177e4
LT
1572}
1573
6a15674d 1574/* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
cd6ce148 1575static struct request *blk_old_get_request(struct request_queue *q,
9a95e4ef 1576 unsigned int op, blk_mq_req_flags_t flags)
1da177e4
LT
1577{
1578 struct request *rq;
c3036021 1579 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
055f6e18 1580 int ret = 0;
1da177e4 1581
332ebbf7
BVA
1582 WARN_ON_ONCE(q->mq_ops);
1583
7f4b35d1
TH
1584 /* create ioc upfront */
1585 create_io_context(gfp_mask, q->node);
1586
3a0a5299 1587 ret = blk_queue_enter(q, flags);
055f6e18
ML
1588 if (ret)
1589 return ERR_PTR(ret);
d6344532 1590 spin_lock_irq(q->queue_lock);
4accf5fc 1591 rq = get_request(q, op, NULL, flags, gfp_mask);
0c4de0f3 1592 if (IS_ERR(rq)) {
da8303c6 1593 spin_unlock_irq(q->queue_lock);
055f6e18 1594 blk_queue_exit(q);
0c4de0f3
CH
1595 return rq;
1596 }
1da177e4 1597
0c4de0f3
CH
1598 /* q->queue_lock is unlocked at this point */
1599 rq->__data_len = 0;
1600 rq->__sector = (sector_t) -1;
1601 rq->bio = rq->biotail = NULL;
1da177e4
LT
1602 return rq;
1603}
320ae51f 1604
6a15674d 1605/**
ff005a06 1606 * blk_get_request - allocate a request
6a15674d
BVA
1607 * @q: request queue to allocate a request for
1608 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1609 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1610 */
ff005a06
CH
1611struct request *blk_get_request(struct request_queue *q, unsigned int op,
1612 blk_mq_req_flags_t flags)
320ae51f 1613{
d280bab3
BVA
1614 struct request *req;
1615
6a15674d 1616 WARN_ON_ONCE(op & REQ_NOWAIT);
1b6d65a0 1617 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
6a15674d 1618
d280bab3 1619 if (q->mq_ops) {
6a15674d 1620 req = blk_mq_alloc_request(q, op, flags);
d280bab3
BVA
1621 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1622 q->mq_ops->initialize_rq_fn(req);
1623 } else {
6a15674d 1624 req = blk_old_get_request(q, op, flags);
d280bab3
BVA
1625 if (!IS_ERR(req) && q->initialize_rq_fn)
1626 q->initialize_rq_fn(req);
1627 }
1628
1629 return req;
320ae51f 1630}
1da177e4
LT
1631EXPORT_SYMBOL(blk_get_request);
1632
1633/**
1634 * blk_requeue_request - put a request back on queue
1635 * @q: request queue where request should be inserted
1636 * @rq: request to be inserted
1637 *
1638 * Description:
1639 * Drivers often keep queueing requests until the hardware cannot accept
1640 * more, when that condition happens we need to put the request back
1641 * on the queue. Must be called with queue lock held.
1642 */
165125e1 1643void blk_requeue_request(struct request_queue *q, struct request *rq)
1da177e4 1644{
2fff8a92 1645 lockdep_assert_held(q->queue_lock);
332ebbf7 1646 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1647
242f9dcb
JA
1648 blk_delete_timer(rq);
1649 blk_clear_rq_complete(rq);
5f3ea37c 1650 trace_block_rq_requeue(q, rq);
a8a45941 1651 wbt_requeue(q->rq_wb, rq);
2056a782 1652
e8064021 1653 if (rq->rq_flags & RQF_QUEUED)
1da177e4
LT
1654 blk_queue_end_tag(q, rq);
1655
ba396a6c
JB
1656 BUG_ON(blk_queued_rq(rq));
1657
1da177e4
LT
1658 elv_requeue_request(q, rq);
1659}
1da177e4
LT
1660EXPORT_SYMBOL(blk_requeue_request);
1661
73c10101
JA
1662static void add_acct_request(struct request_queue *q, struct request *rq,
1663 int where)
1664{
320ae51f 1665 blk_account_io_start(rq, true);
7eaceacc 1666 __elv_add_request(q, rq, where);
73c10101
JA
1667}
1668
d62e26b3 1669static void part_round_stats_single(struct request_queue *q, int cpu,
b8d62b3a
JA
1670 struct hd_struct *part, unsigned long now,
1671 unsigned int inflight)
074a7aca 1672{
b8d62b3a 1673 if (inflight) {
074a7aca 1674 __part_stat_add(cpu, part, time_in_queue,
b8d62b3a 1675 inflight * (now - part->stamp));
074a7aca
TH
1676 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1677 }
1678 part->stamp = now;
1679}
1680
1681/**
496aa8a9 1682 * part_round_stats() - Round off the performance stats on a struct disk_stats.
d62e26b3 1683 * @q: target block queue
496aa8a9
RD
1684 * @cpu: cpu number for stats access
1685 * @part: target partition
1da177e4
LT
1686 *
1687 * The average IO queue length and utilisation statistics are maintained
1688 * by observing the current state of the queue length and the amount of
1689 * time it has been in this state for.
1690 *
1691 * Normally, that accounting is done on IO completion, but that can result
1692 * in more than a second's worth of IO being accounted for within any one
1693 * second, leading to >100% utilisation. To deal with that, we call this
1694 * function to do a round-off before returning the results when reading
1695 * /proc/diskstats. This accounts immediately for all queue usage up to
1696 * the current jiffies and restarts the counters again.
1697 */
d62e26b3 1698void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
6f2576af 1699{
b8d62b3a 1700 struct hd_struct *part2 = NULL;
6f2576af 1701 unsigned long now = jiffies;
b8d62b3a
JA
1702 unsigned int inflight[2];
1703 int stats = 0;
1704
1705 if (part->stamp != now)
1706 stats |= 1;
1707
1708 if (part->partno) {
1709 part2 = &part_to_disk(part)->part0;
1710 if (part2->stamp != now)
1711 stats |= 2;
1712 }
1713
1714 if (!stats)
1715 return;
1716
1717 part_in_flight(q, part, inflight);
6f2576af 1718
b8d62b3a
JA
1719 if (stats & 2)
1720 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1721 if (stats & 1)
1722 part_round_stats_single(q, cpu, part, now, inflight[0]);
6f2576af 1723}
074a7aca 1724EXPORT_SYMBOL_GPL(part_round_stats);
6f2576af 1725
47fafbc7 1726#ifdef CONFIG_PM
c8158819
LM
1727static void blk_pm_put_request(struct request *rq)
1728{
e8064021 1729 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
c8158819
LM
1730 pm_runtime_mark_last_busy(rq->q->dev);
1731}
1732#else
1733static inline void blk_pm_put_request(struct request *rq) {}
1734#endif
1735
165125e1 1736void __blk_put_request(struct request_queue *q, struct request *req)
1da177e4 1737{
e8064021
CH
1738 req_flags_t rq_flags = req->rq_flags;
1739
1da177e4
LT
1740 if (unlikely(!q))
1741 return;
1da177e4 1742
6f5ba581
CH
1743 if (q->mq_ops) {
1744 blk_mq_free_request(req);
1745 return;
1746 }
1747
2fff8a92
BVA
1748 lockdep_assert_held(q->queue_lock);
1749
6cc77e9c 1750 blk_req_zone_write_unlock(req);
c8158819
LM
1751 blk_pm_put_request(req);
1752
8922e16c
TH
1753 elv_completed_request(q, req);
1754
1cd96c24
BH
1755 /* this is a bio leak */
1756 WARN_ON(req->bio != NULL);
1757
a8a45941 1758 wbt_done(q->rq_wb, req);
87760e5e 1759
1da177e4
LT
1760 /*
1761 * Request may not have originated from ll_rw_blk. if not,
1762 * it didn't come out of our reserved rq pools
1763 */
e8064021 1764 if (rq_flags & RQF_ALLOCED) {
a051661c 1765 struct request_list *rl = blk_rq_rl(req);
ef295ecf 1766 bool sync = op_is_sync(req->cmd_flags);
1da177e4 1767
1da177e4 1768 BUG_ON(!list_empty(&req->queuelist));
360f92c2 1769 BUG_ON(ELV_ON_HASH(req));
1da177e4 1770
a051661c 1771 blk_free_request(rl, req);
e8064021 1772 freed_request(rl, sync, rq_flags);
a051661c 1773 blk_put_rl(rl);
055f6e18 1774 blk_queue_exit(q);
1da177e4
LT
1775 }
1776}
6e39b69e
MC
1777EXPORT_SYMBOL_GPL(__blk_put_request);
1778
1da177e4
LT
1779void blk_put_request(struct request *req)
1780{
165125e1 1781 struct request_queue *q = req->q;
8922e16c 1782
320ae51f
JA
1783 if (q->mq_ops)
1784 blk_mq_free_request(req);
1785 else {
1786 unsigned long flags;
1787
1788 spin_lock_irqsave(q->queue_lock, flags);
1789 __blk_put_request(q, req);
1790 spin_unlock_irqrestore(q->queue_lock, flags);
1791 }
1da177e4 1792}
1da177e4
LT
1793EXPORT_SYMBOL(blk_put_request);
1794
320ae51f
JA
1795bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1796 struct bio *bio)
73c10101 1797{
1eff9d32 1798 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1799
73c10101
JA
1800 if (!ll_back_merge_fn(q, req, bio))
1801 return false;
1802
8c1cf6bb 1803 trace_block_bio_backmerge(q, req, bio);
73c10101
JA
1804
1805 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1806 blk_rq_set_mixed_merge(req);
1807
1808 req->biotail->bi_next = bio;
1809 req->biotail = bio;
4f024f37 1810 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1811 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1812
320ae51f 1813 blk_account_io_start(req, false);
73c10101
JA
1814 return true;
1815}
1816
320ae51f
JA
1817bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1818 struct bio *bio)
73c10101 1819{
1eff9d32 1820 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1821
73c10101
JA
1822 if (!ll_front_merge_fn(q, req, bio))
1823 return false;
1824
8c1cf6bb 1825 trace_block_bio_frontmerge(q, req, bio);
73c10101
JA
1826
1827 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1828 blk_rq_set_mixed_merge(req);
1829
73c10101
JA
1830 bio->bi_next = req->bio;
1831 req->bio = bio;
1832
4f024f37
KO
1833 req->__sector = bio->bi_iter.bi_sector;
1834 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1835 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1836
320ae51f 1837 blk_account_io_start(req, false);
73c10101
JA
1838 return true;
1839}
1840
1e739730
CH
1841bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1842 struct bio *bio)
1843{
1844 unsigned short segments = blk_rq_nr_discard_segments(req);
1845
1846 if (segments >= queue_max_discard_segments(q))
1847 goto no_merge;
1848 if (blk_rq_sectors(req) + bio_sectors(bio) >
1849 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1850 goto no_merge;
1851
1852 req->biotail->bi_next = bio;
1853 req->biotail = bio;
1854 req->__data_len += bio->bi_iter.bi_size;
1855 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1856 req->nr_phys_segments = segments + 1;
1857
1858 blk_account_io_start(req, false);
1859 return true;
1860no_merge:
1861 req_set_nomerge(q, req);
1862 return false;
1863}
1864
bd87b589 1865/**
320ae51f 1866 * blk_attempt_plug_merge - try to merge with %current's plugged list
bd87b589
TH
1867 * @q: request_queue new bio is being queued at
1868 * @bio: new bio being queued
1869 * @request_count: out parameter for number of traversed plugged requests
ccc2600b
RD
1870 * @same_queue_rq: pointer to &struct request that gets filled in when
1871 * another request associated with @q is found on the plug list
1872 * (optional, may be %NULL)
bd87b589
TH
1873 *
1874 * Determine whether @bio being queued on @q can be merged with a request
1875 * on %current's plugged list. Returns %true if merge was successful,
1876 * otherwise %false.
1877 *
07c2bd37
TH
1878 * Plugging coalesces IOs from the same issuer for the same purpose without
1879 * going through @q->queue_lock. As such it's more of an issuing mechanism
1880 * than scheduling, and the request, while may have elvpriv data, is not
1881 * added on the elevator at this point. In addition, we don't have
1882 * reliable access to the elevator outside queue lock. Only check basic
1883 * merging parameters without querying the elevator.
da41a589
RE
1884 *
1885 * Caller must ensure !blk_queue_nomerges(q) beforehand.
73c10101 1886 */
320ae51f 1887bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
5b3f341f
SL
1888 unsigned int *request_count,
1889 struct request **same_queue_rq)
73c10101
JA
1890{
1891 struct blk_plug *plug;
1892 struct request *rq;
92f399c7 1893 struct list_head *plug_list;
73c10101 1894
bd87b589 1895 plug = current->plug;
73c10101 1896 if (!plug)
34fe7c05 1897 return false;
56ebdaf2 1898 *request_count = 0;
73c10101 1899
92f399c7
SL
1900 if (q->mq_ops)
1901 plug_list = &plug->mq_list;
1902 else
1903 plug_list = &plug->list;
1904
1905 list_for_each_entry_reverse(rq, plug_list, queuelist) {
34fe7c05 1906 bool merged = false;
73c10101 1907
5b3f341f 1908 if (rq->q == q) {
1b2e19f1 1909 (*request_count)++;
5b3f341f
SL
1910 /*
1911 * Only blk-mq multiple hardware queues case checks the
1912 * rq in the same queue, there should be only one such
1913 * rq in a queue
1914 **/
1915 if (same_queue_rq)
1916 *same_queue_rq = rq;
1917 }
56ebdaf2 1918
07c2bd37 1919 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
73c10101
JA
1920 continue;
1921
34fe7c05
CH
1922 switch (blk_try_merge(rq, bio)) {
1923 case ELEVATOR_BACK_MERGE:
1924 merged = bio_attempt_back_merge(q, rq, bio);
1925 break;
1926 case ELEVATOR_FRONT_MERGE:
1927 merged = bio_attempt_front_merge(q, rq, bio);
1928 break;
1e739730
CH
1929 case ELEVATOR_DISCARD_MERGE:
1930 merged = bio_attempt_discard_merge(q, rq, bio);
1931 break;
34fe7c05
CH
1932 default:
1933 break;
73c10101 1934 }
34fe7c05
CH
1935
1936 if (merged)
1937 return true;
73c10101 1938 }
34fe7c05
CH
1939
1940 return false;
73c10101
JA
1941}
1942
0809e3ac
JM
1943unsigned int blk_plug_queued_count(struct request_queue *q)
1944{
1945 struct blk_plug *plug;
1946 struct request *rq;
1947 struct list_head *plug_list;
1948 unsigned int ret = 0;
1949
1950 plug = current->plug;
1951 if (!plug)
1952 goto out;
1953
1954 if (q->mq_ops)
1955 plug_list = &plug->mq_list;
1956 else
1957 plug_list = &plug->list;
1958
1959 list_for_each_entry(rq, plug_list, queuelist) {
1960 if (rq->q == q)
1961 ret++;
1962 }
1963out:
1964 return ret;
1965}
1966
da8d7f07 1967void blk_init_request_from_bio(struct request *req, struct bio *bio)
52d9e675 1968{
0be0dee6
BVA
1969 struct io_context *ioc = rq_ioc(bio);
1970
1eff9d32 1971 if (bio->bi_opf & REQ_RAHEAD)
a82afdfc 1972 req->cmd_flags |= REQ_FAILFAST_MASK;
b31dc66a 1973
4f024f37 1974 req->__sector = bio->bi_iter.bi_sector;
5dc8b362
AM
1975 if (ioprio_valid(bio_prio(bio)))
1976 req->ioprio = bio_prio(bio);
0be0dee6
BVA
1977 else if (ioc)
1978 req->ioprio = ioc->ioprio;
1979 else
1980 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
cb6934f8 1981 req->write_hint = bio->bi_write_hint;
bc1c56fd 1982 blk_rq_bio_prep(req->q, req, bio);
52d9e675 1983}
da8d7f07 1984EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
52d9e675 1985
dece1635 1986static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1da177e4 1987{
73c10101 1988 struct blk_plug *plug;
34fe7c05 1989 int where = ELEVATOR_INSERT_SORT;
e4d750c9 1990 struct request *req, *free;
56ebdaf2 1991 unsigned int request_count = 0;
87760e5e 1992 unsigned int wb_acct;
1da177e4 1993
1da177e4
LT
1994 /*
1995 * low level driver can indicate that it wants pages above a
1996 * certain limit bounced to low memory (ie for highmem, or even
1997 * ISA dma in theory)
1998 */
1999 blk_queue_bounce(q, &bio);
2000
af67c31f 2001 blk_queue_split(q, &bio);
23688bf4 2002
e23947bd 2003 if (!bio_integrity_prep(bio))
dece1635 2004 return BLK_QC_T_NONE;
ffecfd1a 2005
f73f44eb 2006 if (op_is_flush(bio->bi_opf)) {
73c10101 2007 spin_lock_irq(q->queue_lock);
ae1b1539 2008 where = ELEVATOR_INSERT_FLUSH;
28e7d184
TH
2009 goto get_rq;
2010 }
2011
73c10101
JA
2012 /*
2013 * Check if we can merge with the plugged list before grabbing
2014 * any locks.
2015 */
0809e3ac
JM
2016 if (!blk_queue_nomerges(q)) {
2017 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
dece1635 2018 return BLK_QC_T_NONE;
0809e3ac
JM
2019 } else
2020 request_count = blk_plug_queued_count(q);
1da177e4 2021
73c10101 2022 spin_lock_irq(q->queue_lock);
2056a782 2023
34fe7c05
CH
2024 switch (elv_merge(q, &req, bio)) {
2025 case ELEVATOR_BACK_MERGE:
2026 if (!bio_attempt_back_merge(q, req, bio))
2027 break;
2028 elv_bio_merged(q, req, bio);
2029 free = attempt_back_merge(q, req);
2030 if (free)
2031 __blk_put_request(q, free);
2032 else
2033 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2034 goto out_unlock;
2035 case ELEVATOR_FRONT_MERGE:
2036 if (!bio_attempt_front_merge(q, req, bio))
2037 break;
2038 elv_bio_merged(q, req, bio);
2039 free = attempt_front_merge(q, req);
2040 if (free)
2041 __blk_put_request(q, free);
2042 else
2043 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2044 goto out_unlock;
2045 default:
2046 break;
1da177e4
LT
2047 }
2048
450991bc 2049get_rq:
87760e5e
JA
2050 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
2051
1da177e4 2052 /*
450991bc 2053 * Grab a free request. This is might sleep but can not fail.
d6344532 2054 * Returns with the queue unlocked.
450991bc 2055 */
055f6e18 2056 blk_queue_enter_live(q);
c3036021 2057 req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO);
a492f075 2058 if (IS_ERR(req)) {
055f6e18 2059 blk_queue_exit(q);
87760e5e 2060 __wbt_done(q->rq_wb, wb_acct);
4e4cbee9
CH
2061 if (PTR_ERR(req) == -ENOMEM)
2062 bio->bi_status = BLK_STS_RESOURCE;
2063 else
2064 bio->bi_status = BLK_STS_IOERR;
4246a0b6 2065 bio_endio(bio);
da8303c6
TH
2066 goto out_unlock;
2067 }
d6344532 2068
a8a45941 2069 wbt_track(req, wb_acct);
87760e5e 2070
450991bc
NP
2071 /*
2072 * After dropping the lock and possibly sleeping here, our request
2073 * may now be mergeable after it had proven unmergeable (above).
2074 * We don't worry about that case for efficiency. It won't happen
2075 * often, and the elevators are able to handle it.
1da177e4 2076 */
da8d7f07 2077 blk_init_request_from_bio(req, bio);
1da177e4 2078
9562ad9a 2079 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
11ccf116 2080 req->cpu = raw_smp_processor_id();
73c10101
JA
2081
2082 plug = current->plug;
721a9602 2083 if (plug) {
dc6d36c9
JA
2084 /*
2085 * If this is the first request added after a plug, fire
7aef2e78 2086 * of a plug trace.
0a6219a9
ML
2087 *
2088 * @request_count may become stale because of schedule
2089 * out, so check plug list again.
dc6d36c9 2090 */
0a6219a9 2091 if (!request_count || list_empty(&plug->list))
dc6d36c9 2092 trace_block_plug(q);
3540d5e8 2093 else {
50d24c34
SL
2094 struct request *last = list_entry_rq(plug->list.prev);
2095 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2096 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
3540d5e8 2097 blk_flush_plug_list(plug, false);
019ceb7d
SL
2098 trace_block_plug(q);
2099 }
73c10101 2100 }
73c10101 2101 list_add_tail(&req->queuelist, &plug->list);
320ae51f 2102 blk_account_io_start(req, true);
73c10101
JA
2103 } else {
2104 spin_lock_irq(q->queue_lock);
2105 add_acct_request(q, req, where);
24ecfbe2 2106 __blk_run_queue(q);
73c10101
JA
2107out_unlock:
2108 spin_unlock_irq(q->queue_lock);
2109 }
dece1635
JA
2110
2111 return BLK_QC_T_NONE;
1da177e4
LT
2112}
2113
52c5e62d 2114static void handle_bad_sector(struct bio *bio, sector_t maxsector)
1da177e4
LT
2115{
2116 char b[BDEVNAME_SIZE];
2117
2118 printk(KERN_INFO "attempt to access beyond end of device\n");
6296b960 2119 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
74d46992 2120 bio_devname(bio, b), bio->bi_opf,
f73a1c7d 2121 (unsigned long long)bio_end_sector(bio),
52c5e62d 2122 (long long)maxsector);
1da177e4
LT
2123}
2124
c17bb495
AM
2125#ifdef CONFIG_FAIL_MAKE_REQUEST
2126
2127static DECLARE_FAULT_ATTR(fail_make_request);
2128
2129static int __init setup_fail_make_request(char *str)
2130{
2131 return setup_fault_attr(&fail_make_request, str);
2132}
2133__setup("fail_make_request=", setup_fail_make_request);
2134
b2c9cd37 2135static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
c17bb495 2136{
b2c9cd37 2137 return part->make_it_fail && should_fail(&fail_make_request, bytes);
c17bb495
AM
2138}
2139
2140static int __init fail_make_request_debugfs(void)
2141{
dd48c085
AM
2142 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2143 NULL, &fail_make_request);
2144
21f9fcd8 2145 return PTR_ERR_OR_ZERO(dir);
c17bb495
AM
2146}
2147
2148late_initcall(fail_make_request_debugfs);
2149
2150#else /* CONFIG_FAIL_MAKE_REQUEST */
2151
b2c9cd37
AM
2152static inline bool should_fail_request(struct hd_struct *part,
2153 unsigned int bytes)
c17bb495 2154{
b2c9cd37 2155 return false;
c17bb495
AM
2156}
2157
2158#endif /* CONFIG_FAIL_MAKE_REQUEST */
2159
721c7fc7
ID
2160static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2161{
2162 if (part->policy && op_is_write(bio_op(bio))) {
2163 char b[BDEVNAME_SIZE];
2164
2165 printk(KERN_ERR
2166 "generic_make_request: Trying to write "
2167 "to read-only block-device %s (partno %d)\n",
2168 bio_devname(bio, b), part->partno);
2169 return true;
2170 }
2171
2172 return false;
2173}
2174
30abb3a6
HM
2175static noinline int should_fail_bio(struct bio *bio)
2176{
2177 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2178 return -EIO;
2179 return 0;
2180}
2181ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2182
52c5e62d
CH
2183/*
2184 * Check whether this bio extends beyond the end of the device or partition.
2185 * This may well happen - the kernel calls bread() without checking the size of
2186 * the device, e.g., when mounting a file system.
2187 */
2188static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2189{
2190 unsigned int nr_sectors = bio_sectors(bio);
2191
2192 if (nr_sectors && maxsector &&
2193 (nr_sectors > maxsector ||
2194 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2195 handle_bad_sector(bio, maxsector);
2196 return -EIO;
2197 }
2198 return 0;
2199}
2200
74d46992
CH
2201/*
2202 * Remap block n of partition p to block n+start(p) of the disk.
2203 */
2204static inline int blk_partition_remap(struct bio *bio)
2205{
2206 struct hd_struct *p;
52c5e62d 2207 int ret = -EIO;
74d46992 2208
721c7fc7
ID
2209 rcu_read_lock();
2210 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
52c5e62d
CH
2211 if (unlikely(!p))
2212 goto out;
2213 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2214 goto out;
2215 if (unlikely(bio_check_ro(bio, p)))
721c7fc7 2216 goto out;
721c7fc7 2217
74d46992
CH
2218 /*
2219 * Zone reset does not include bi_size so bio_sectors() is always 0.
2220 * Include a test for the reset op code and perform the remap if needed.
2221 */
52c5e62d
CH
2222 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2223 if (bio_check_eod(bio, part_nr_sects_read(p)))
2224 goto out;
2225 bio->bi_iter.bi_sector += p->start_sect;
2226 bio->bi_partno = 0;
2227 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2228 bio->bi_iter.bi_sector - p->start_sect);
2229 }
2230 ret = 0;
721c7fc7
ID
2231out:
2232 rcu_read_unlock();
74d46992
CH
2233 return ret;
2234}
2235
27a84d54
CH
2236static noinline_for_stack bool
2237generic_make_request_checks(struct bio *bio)
1da177e4 2238{
165125e1 2239 struct request_queue *q;
5a7bbad2 2240 int nr_sectors = bio_sectors(bio);
4e4cbee9 2241 blk_status_t status = BLK_STS_IOERR;
5a7bbad2 2242 char b[BDEVNAME_SIZE];
1da177e4
LT
2243
2244 might_sleep();
1da177e4 2245
74d46992 2246 q = bio->bi_disk->queue;
5a7bbad2
CH
2247 if (unlikely(!q)) {
2248 printk(KERN_ERR
2249 "generic_make_request: Trying to access "
2250 "nonexistent block-device %s (%Lu)\n",
74d46992 2251 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
5a7bbad2
CH
2252 goto end_io;
2253 }
c17bb495 2254
03a07c92
GR
2255 /*
2256 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2257 * if queue is not a request based queue.
2258 */
03a07c92
GR
2259 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2260 goto not_supported;
2261
30abb3a6 2262 if (should_fail_bio(bio))
5a7bbad2 2263 goto end_io;
2056a782 2264
52c5e62d
CH
2265 if (bio->bi_partno) {
2266 if (unlikely(blk_partition_remap(bio)))
721c7fc7
ID
2267 goto end_io;
2268 } else {
52c5e62d
CH
2269 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2270 goto end_io;
2271 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
721c7fc7
ID
2272 goto end_io;
2273 }
2056a782 2274
5a7bbad2
CH
2275 /*
2276 * Filter flush bio's early so that make_request based
2277 * drivers without flush support don't have to worry
2278 * about them.
2279 */
f3a8ab7d 2280 if (op_is_flush(bio->bi_opf) &&
c888a8f9 2281 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1eff9d32 2282 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
5a7bbad2 2283 if (!nr_sectors) {
4e4cbee9 2284 status = BLK_STS_OK;
51fd77bd
JA
2285 goto end_io;
2286 }
5a7bbad2 2287 }
5ddfe969 2288
288dab8a
CH
2289 switch (bio_op(bio)) {
2290 case REQ_OP_DISCARD:
2291 if (!blk_queue_discard(q))
2292 goto not_supported;
2293 break;
2294 case REQ_OP_SECURE_ERASE:
2295 if (!blk_queue_secure_erase(q))
2296 goto not_supported;
2297 break;
2298 case REQ_OP_WRITE_SAME:
74d46992 2299 if (!q->limits.max_write_same_sectors)
288dab8a 2300 goto not_supported;
58886785 2301 break;
2d253440
ST
2302 case REQ_OP_ZONE_REPORT:
2303 case REQ_OP_ZONE_RESET:
74d46992 2304 if (!blk_queue_is_zoned(q))
2d253440 2305 goto not_supported;
288dab8a 2306 break;
a6f0788e 2307 case REQ_OP_WRITE_ZEROES:
74d46992 2308 if (!q->limits.max_write_zeroes_sectors)
a6f0788e
CK
2309 goto not_supported;
2310 break;
288dab8a
CH
2311 default:
2312 break;
5a7bbad2 2313 }
01edede4 2314
7f4b35d1
TH
2315 /*
2316 * Various block parts want %current->io_context and lazy ioc
2317 * allocation ends up trading a lot of pain for a small amount of
2318 * memory. Just allocate it upfront. This may fail and block
2319 * layer knows how to live with it.
2320 */
2321 create_io_context(GFP_ATOMIC, q->node);
2322
ae118896
TH
2323 if (!blkcg_bio_issue_check(q, bio))
2324 return false;
27a84d54 2325
fbbaf700
N
2326 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2327 trace_block_bio_queue(q, bio);
2328 /* Now that enqueuing has been traced, we need to trace
2329 * completion as well.
2330 */
2331 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2332 }
27a84d54 2333 return true;
a7384677 2334
288dab8a 2335not_supported:
4e4cbee9 2336 status = BLK_STS_NOTSUPP;
a7384677 2337end_io:
4e4cbee9 2338 bio->bi_status = status;
4246a0b6 2339 bio_endio(bio);
27a84d54 2340 return false;
1da177e4
LT
2341}
2342
27a84d54
CH
2343/**
2344 * generic_make_request - hand a buffer to its device driver for I/O
2345 * @bio: The bio describing the location in memory and on the device.
2346 *
2347 * generic_make_request() is used to make I/O requests of block
2348 * devices. It is passed a &struct bio, which describes the I/O that needs
2349 * to be done.
2350 *
2351 * generic_make_request() does not return any status. The
2352 * success/failure status of the request, along with notification of
2353 * completion, is delivered asynchronously through the bio->bi_end_io
2354 * function described (one day) else where.
2355 *
2356 * The caller of generic_make_request must make sure that bi_io_vec
2357 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2358 * set to describe the device address, and the
2359 * bi_end_io and optionally bi_private are set to describe how
2360 * completion notification should be signaled.
2361 *
2362 * generic_make_request and the drivers it calls may use bi_next if this
2363 * bio happens to be merged with someone else, and may resubmit the bio to
2364 * a lower device by calling into generic_make_request recursively, which
2365 * means the bio should NOT be touched after the call to ->make_request_fn.
d89d8796 2366 */
dece1635 2367blk_qc_t generic_make_request(struct bio *bio)
d89d8796 2368{
f5fe1b51
N
2369 /*
2370 * bio_list_on_stack[0] contains bios submitted by the current
2371 * make_request_fn.
2372 * bio_list_on_stack[1] contains bios that were submitted before
2373 * the current make_request_fn, but that haven't been processed
2374 * yet.
2375 */
2376 struct bio_list bio_list_on_stack[2];
37f9579f
BVA
2377 blk_mq_req_flags_t flags = 0;
2378 struct request_queue *q = bio->bi_disk->queue;
dece1635 2379 blk_qc_t ret = BLK_QC_T_NONE;
bddd87c7 2380
37f9579f
BVA
2381 if (bio->bi_opf & REQ_NOWAIT)
2382 flags = BLK_MQ_REQ_NOWAIT;
2383 if (blk_queue_enter(q, flags) < 0) {
2384 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2385 bio_wouldblock_error(bio);
2386 else
2387 bio_io_error(bio);
2388 return ret;
2389 }
2390
27a84d54 2391 if (!generic_make_request_checks(bio))
dece1635 2392 goto out;
27a84d54
CH
2393
2394 /*
2395 * We only want one ->make_request_fn to be active at a time, else
2396 * stack usage with stacked devices could be a problem. So use
2397 * current->bio_list to keep a list of requests submited by a
2398 * make_request_fn function. current->bio_list is also used as a
2399 * flag to say if generic_make_request is currently active in this
2400 * task or not. If it is NULL, then no make_request is active. If
2401 * it is non-NULL, then a make_request is active, and new requests
2402 * should be added at the tail
2403 */
bddd87c7 2404 if (current->bio_list) {
f5fe1b51 2405 bio_list_add(&current->bio_list[0], bio);
dece1635 2406 goto out;
d89d8796 2407 }
27a84d54 2408
d89d8796
NB
2409 /* following loop may be a bit non-obvious, and so deserves some
2410 * explanation.
2411 * Before entering the loop, bio->bi_next is NULL (as all callers
2412 * ensure that) so we have a list with a single bio.
2413 * We pretend that we have just taken it off a longer list, so
bddd87c7
AM
2414 * we assign bio_list to a pointer to the bio_list_on_stack,
2415 * thus initialising the bio_list of new bios to be
27a84d54 2416 * added. ->make_request() may indeed add some more bios
d89d8796
NB
2417 * through a recursive call to generic_make_request. If it
2418 * did, we find a non-NULL value in bio_list and re-enter the loop
2419 * from the top. In this case we really did just take the bio
bddd87c7 2420 * of the top of the list (no pretending) and so remove it from
27a84d54 2421 * bio_list, and call into ->make_request() again.
d89d8796
NB
2422 */
2423 BUG_ON(bio->bi_next);
f5fe1b51
N
2424 bio_list_init(&bio_list_on_stack[0]);
2425 current->bio_list = bio_list_on_stack;
d89d8796 2426 do {
37f9579f
BVA
2427 bool enter_succeeded = true;
2428
2429 if (unlikely(q != bio->bi_disk->queue)) {
2430 if (q)
2431 blk_queue_exit(q);
2432 q = bio->bi_disk->queue;
2433 flags = 0;
2434 if (bio->bi_opf & REQ_NOWAIT)
2435 flags = BLK_MQ_REQ_NOWAIT;
2436 if (blk_queue_enter(q, flags) < 0) {
2437 enter_succeeded = false;
2438 q = NULL;
2439 }
2440 }
27a84d54 2441
37f9579f 2442 if (enter_succeeded) {
79bd9959
N
2443 struct bio_list lower, same;
2444
2445 /* Create a fresh bio_list for all subordinate requests */
f5fe1b51
N
2446 bio_list_on_stack[1] = bio_list_on_stack[0];
2447 bio_list_init(&bio_list_on_stack[0]);
dece1635 2448 ret = q->make_request_fn(q, bio);
3ef28e83 2449
79bd9959
N
2450 /* sort new bios into those for a lower level
2451 * and those for the same level
2452 */
2453 bio_list_init(&lower);
2454 bio_list_init(&same);
f5fe1b51 2455 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
74d46992 2456 if (q == bio->bi_disk->queue)
79bd9959
N
2457 bio_list_add(&same, bio);
2458 else
2459 bio_list_add(&lower, bio);
2460 /* now assemble so we handle the lowest level first */
f5fe1b51
N
2461 bio_list_merge(&bio_list_on_stack[0], &lower);
2462 bio_list_merge(&bio_list_on_stack[0], &same);
2463 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
3ef28e83 2464 } else {
03a07c92
GR
2465 if (unlikely(!blk_queue_dying(q) &&
2466 (bio->bi_opf & REQ_NOWAIT)))
2467 bio_wouldblock_error(bio);
2468 else
2469 bio_io_error(bio);
3ef28e83 2470 }
f5fe1b51 2471 bio = bio_list_pop(&bio_list_on_stack[0]);
d89d8796 2472 } while (bio);
bddd87c7 2473 current->bio_list = NULL; /* deactivate */
dece1635
JA
2474
2475out:
37f9579f
BVA
2476 if (q)
2477 blk_queue_exit(q);
dece1635 2478 return ret;
d89d8796 2479}
1da177e4
LT
2480EXPORT_SYMBOL(generic_make_request);
2481
f421e1d9
CH
2482/**
2483 * direct_make_request - hand a buffer directly to its device driver for I/O
2484 * @bio: The bio describing the location in memory and on the device.
2485 *
2486 * This function behaves like generic_make_request(), but does not protect
2487 * against recursion. Must only be used if the called driver is known
2488 * to not call generic_make_request (or direct_make_request) again from
2489 * its make_request function. (Calling direct_make_request again from
2490 * a workqueue is perfectly fine as that doesn't recurse).
2491 */
2492blk_qc_t direct_make_request(struct bio *bio)
2493{
2494 struct request_queue *q = bio->bi_disk->queue;
2495 bool nowait = bio->bi_opf & REQ_NOWAIT;
2496 blk_qc_t ret;
2497
2498 if (!generic_make_request_checks(bio))
2499 return BLK_QC_T_NONE;
2500
3a0a5299 2501 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
f421e1d9
CH
2502 if (nowait && !blk_queue_dying(q))
2503 bio->bi_status = BLK_STS_AGAIN;
2504 else
2505 bio->bi_status = BLK_STS_IOERR;
2506 bio_endio(bio);
2507 return BLK_QC_T_NONE;
2508 }
2509
2510 ret = q->make_request_fn(q, bio);
2511 blk_queue_exit(q);
2512 return ret;
2513}
2514EXPORT_SYMBOL_GPL(direct_make_request);
2515
1da177e4 2516/**
710027a4 2517 * submit_bio - submit a bio to the block device layer for I/O
1da177e4
LT
2518 * @bio: The &struct bio which describes the I/O
2519 *
2520 * submit_bio() is very similar in purpose to generic_make_request(), and
2521 * uses that function to do most of the work. Both are fairly rough
710027a4 2522 * interfaces; @bio must be presetup and ready for I/O.
1da177e4
LT
2523 *
2524 */
4e49ea4a 2525blk_qc_t submit_bio(struct bio *bio)
1da177e4 2526{
bf2de6f5
JA
2527 /*
2528 * If it's a regular read/write or a barrier with data attached,
2529 * go through the normal accounting stuff before submission.
2530 */
e2a60da7 2531 if (bio_has_data(bio)) {
4363ac7c
MP
2532 unsigned int count;
2533
95fe6c1a 2534 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
7c5a0dcf 2535 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
4363ac7c
MP
2536 else
2537 count = bio_sectors(bio);
2538
a8ebb056 2539 if (op_is_write(bio_op(bio))) {
bf2de6f5
JA
2540 count_vm_events(PGPGOUT, count);
2541 } else {
4f024f37 2542 task_io_account_read(bio->bi_iter.bi_size);
bf2de6f5
JA
2543 count_vm_events(PGPGIN, count);
2544 }
2545
2546 if (unlikely(block_dump)) {
2547 char b[BDEVNAME_SIZE];
8dcbdc74 2548 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
ba25f9dc 2549 current->comm, task_pid_nr(current),
a8ebb056 2550 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
4f024f37 2551 (unsigned long long)bio->bi_iter.bi_sector,
74d46992 2552 bio_devname(bio, b), count);
bf2de6f5 2553 }
1da177e4
LT
2554 }
2555
dece1635 2556 return generic_make_request(bio);
1da177e4 2557}
1da177e4
LT
2558EXPORT_SYMBOL(submit_bio);
2559
ea435e1b
CH
2560bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2561{
2562 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2563 return false;
2564
2565 if (current->plug)
2566 blk_flush_plug_list(current->plug, false);
2567 return q->poll_fn(q, cookie);
2568}
2569EXPORT_SYMBOL_GPL(blk_poll);
2570
82124d60 2571/**
bf4e6b4e
HR
2572 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2573 * for new the queue limits
82124d60
KU
2574 * @q: the queue
2575 * @rq: the request being checked
2576 *
2577 * Description:
2578 * @rq may have been made based on weaker limitations of upper-level queues
2579 * in request stacking drivers, and it may violate the limitation of @q.
2580 * Since the block layer and the underlying device driver trust @rq
2581 * after it is inserted to @q, it should be checked against @q before
2582 * the insertion using this generic function.
2583 *
82124d60 2584 * Request stacking drivers like request-based dm may change the queue
bf4e6b4e
HR
2585 * limits when retrying requests on other queues. Those requests need
2586 * to be checked against the new queue limits again during dispatch.
82124d60 2587 */
bf4e6b4e
HR
2588static int blk_cloned_rq_check_limits(struct request_queue *q,
2589 struct request *rq)
82124d60 2590{
8fe0d473 2591 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
82124d60
KU
2592 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2593 return -EIO;
2594 }
2595
2596 /*
2597 * queue's settings related to segment counting like q->bounce_pfn
2598 * may differ from that of other stacking queues.
2599 * Recalculate it to check the request correctly on this queue's
2600 * limitation.
2601 */
2602 blk_recalc_rq_segments(rq);
8a78362c 2603 if (rq->nr_phys_segments > queue_max_segments(q)) {
82124d60
KU
2604 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2605 return -EIO;
2606 }
2607
2608 return 0;
2609}
82124d60
KU
2610
2611/**
2612 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2613 * @q: the queue to submit the request
2614 * @rq: the request being queued
2615 */
2a842aca 2616blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
82124d60
KU
2617{
2618 unsigned long flags;
4853abaa 2619 int where = ELEVATOR_INSERT_BACK;
82124d60 2620
bf4e6b4e 2621 if (blk_cloned_rq_check_limits(q, rq))
2a842aca 2622 return BLK_STS_IOERR;
82124d60 2623
b2c9cd37
AM
2624 if (rq->rq_disk &&
2625 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2a842aca 2626 return BLK_STS_IOERR;
82124d60 2627
7fb4898e
KB
2628 if (q->mq_ops) {
2629 if (blk_queue_io_stat(q))
2630 blk_account_io_start(rq, true);
157f377b
JA
2631 /*
2632 * Since we have a scheduler attached on the top device,
2633 * bypass a potential scheduler on the bottom device for
2634 * insert.
2635 */
c77ff7fd 2636 return blk_mq_request_issue_directly(rq);
7fb4898e
KB
2637 }
2638
82124d60 2639 spin_lock_irqsave(q->queue_lock, flags);
3f3299d5 2640 if (unlikely(blk_queue_dying(q))) {
8ba61435 2641 spin_unlock_irqrestore(q->queue_lock, flags);
2a842aca 2642 return BLK_STS_IOERR;
8ba61435 2643 }
82124d60
KU
2644
2645 /*
2646 * Submitting request must be dequeued before calling this function
2647 * because it will be linked to another request_queue
2648 */
2649 BUG_ON(blk_queued_rq(rq));
2650
f73f44eb 2651 if (op_is_flush(rq->cmd_flags))
4853abaa
JM
2652 where = ELEVATOR_INSERT_FLUSH;
2653
2654 add_acct_request(q, rq, where);
e67b77c7
JM
2655 if (where == ELEVATOR_INSERT_FLUSH)
2656 __blk_run_queue(q);
82124d60
KU
2657 spin_unlock_irqrestore(q->queue_lock, flags);
2658
2a842aca 2659 return BLK_STS_OK;
82124d60
KU
2660}
2661EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2662
80a761fd
TH
2663/**
2664 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2665 * @rq: request to examine
2666 *
2667 * Description:
2668 * A request could be merge of IOs which require different failure
2669 * handling. This function determines the number of bytes which
2670 * can be failed from the beginning of the request without
2671 * crossing into area which need to be retried further.
2672 *
2673 * Return:
2674 * The number of bytes to fail.
80a761fd
TH
2675 */
2676unsigned int blk_rq_err_bytes(const struct request *rq)
2677{
2678 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2679 unsigned int bytes = 0;
2680 struct bio *bio;
2681
e8064021 2682 if (!(rq->rq_flags & RQF_MIXED_MERGE))
80a761fd
TH
2683 return blk_rq_bytes(rq);
2684
2685 /*
2686 * Currently the only 'mixing' which can happen is between
2687 * different fastfail types. We can safely fail portions
2688 * which have all the failfast bits that the first one has -
2689 * the ones which are at least as eager to fail as the first
2690 * one.
2691 */
2692 for (bio = rq->bio; bio; bio = bio->bi_next) {
1eff9d32 2693 if ((bio->bi_opf & ff) != ff)
80a761fd 2694 break;
4f024f37 2695 bytes += bio->bi_iter.bi_size;
80a761fd
TH
2696 }
2697
2698 /* this could lead to infinite loop */
2699 BUG_ON(blk_rq_bytes(rq) && !bytes);
2700 return bytes;
2701}
2702EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2703
320ae51f 2704void blk_account_io_completion(struct request *req, unsigned int bytes)
bc58ba94 2705{
c2553b58 2706 if (blk_do_io_stat(req)) {
bc58ba94
JA
2707 const int rw = rq_data_dir(req);
2708 struct hd_struct *part;
2709 int cpu;
2710
2711 cpu = part_stat_lock();
09e099d4 2712 part = req->part;
bc58ba94
JA
2713 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2714 part_stat_unlock();
2715 }
2716}
2717
522a7775 2718void blk_account_io_done(struct request *req, u64 now)
bc58ba94 2719{
bc58ba94 2720 /*
dd4c133f
TH
2721 * Account IO completion. flush_rq isn't accounted as a
2722 * normal IO on queueing nor completion. Accounting the
2723 * containing request is enough.
bc58ba94 2724 */
e8064021 2725 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
522a7775 2726 unsigned long duration;
bc58ba94
JA
2727 const int rw = rq_data_dir(req);
2728 struct hd_struct *part;
2729 int cpu;
2730
522a7775 2731 duration = nsecs_to_jiffies(now - req->start_time_ns);
bc58ba94 2732 cpu = part_stat_lock();
09e099d4 2733 part = req->part;
bc58ba94
JA
2734
2735 part_stat_inc(cpu, part, ios[rw]);
2736 part_stat_add(cpu, part, ticks[rw], duration);
d62e26b3
JA
2737 part_round_stats(req->q, cpu, part);
2738 part_dec_in_flight(req->q, part, rw);
bc58ba94 2739
6c23a968 2740 hd_struct_put(part);
bc58ba94
JA
2741 part_stat_unlock();
2742 }
2743}
2744
47fafbc7 2745#ifdef CONFIG_PM
c8158819
LM
2746/*
2747 * Don't process normal requests when queue is suspended
2748 * or in the process of suspending/resuming
2749 */
e4f36b24 2750static bool blk_pm_allow_request(struct request *rq)
c8158819 2751{
e4f36b24
CH
2752 switch (rq->q->rpm_status) {
2753 case RPM_RESUMING:
2754 case RPM_SUSPENDING:
2755 return rq->rq_flags & RQF_PM;
2756 case RPM_SUSPENDED:
2757 return false;
2758 }
2759
2760 return true;
c8158819
LM
2761}
2762#else
e4f36b24 2763static bool blk_pm_allow_request(struct request *rq)
c8158819 2764{
e4f36b24 2765 return true;
c8158819
LM
2766}
2767#endif
2768
320ae51f
JA
2769void blk_account_io_start(struct request *rq, bool new_io)
2770{
2771 struct hd_struct *part;
2772 int rw = rq_data_dir(rq);
2773 int cpu;
2774
2775 if (!blk_do_io_stat(rq))
2776 return;
2777
2778 cpu = part_stat_lock();
2779
2780 if (!new_io) {
2781 part = rq->part;
2782 part_stat_inc(cpu, part, merges[rw]);
2783 } else {
2784 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2785 if (!hd_struct_try_get(part)) {
2786 /*
2787 * The partition is already being removed,
2788 * the request will be accounted on the disk only
2789 *
2790 * We take a reference on disk->part0 although that
2791 * partition will never be deleted, so we can treat
2792 * it as any other partition.
2793 */
2794 part = &rq->rq_disk->part0;
2795 hd_struct_get(part);
2796 }
d62e26b3
JA
2797 part_round_stats(rq->q, cpu, part);
2798 part_inc_in_flight(rq->q, part, rw);
320ae51f
JA
2799 rq->part = part;
2800 }
2801
2802 part_stat_unlock();
2803}
2804
9c988374
CH
2805static struct request *elv_next_request(struct request_queue *q)
2806{
2807 struct request *rq;
2808 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2809
2810 WARN_ON_ONCE(q->mq_ops);
2811
2812 while (1) {
e4f36b24
CH
2813 list_for_each_entry(rq, &q->queue_head, queuelist) {
2814 if (blk_pm_allow_request(rq))
2815 return rq;
2816
2817 if (rq->rq_flags & RQF_SOFTBARRIER)
2818 break;
9c988374
CH
2819 }
2820
2821 /*
2822 * Flush request is running and flush request isn't queueable
2823 * in the drive, we can hold the queue till flush request is
2824 * finished. Even we don't do this, driver can't dispatch next
2825 * requests and will requeue them. And this can improve
2826 * throughput too. For example, we have request flush1, write1,
2827 * flush 2. flush1 is dispatched, then queue is hold, write1
2828 * isn't inserted to queue. After flush1 is finished, flush2
2829 * will be dispatched. Since disk cache is already clean,
2830 * flush2 will be finished very soon, so looks like flush2 is
2831 * folded to flush1.
2832 * Since the queue is hold, a flag is set to indicate the queue
2833 * should be restarted later. Please see flush_end_io() for
2834 * details.
2835 */
2836 if (fq->flush_pending_idx != fq->flush_running_idx &&
2837 !queue_flush_queueable(q)) {
2838 fq->flush_queue_delayed = 1;
2839 return NULL;
2840 }
2841 if (unlikely(blk_queue_bypass(q)) ||
2842 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2843 return NULL;
2844 }
2845}
2846
3bcddeac 2847/**
9934c8c0
TH
2848 * blk_peek_request - peek at the top of a request queue
2849 * @q: request queue to peek at
2850 *
2851 * Description:
2852 * Return the request at the top of @q. The returned request
2853 * should be started using blk_start_request() before LLD starts
2854 * processing it.
2855 *
2856 * Return:
2857 * Pointer to the request at the top of @q if available. Null
2858 * otherwise.
9934c8c0
TH
2859 */
2860struct request *blk_peek_request(struct request_queue *q)
158dbda0
TH
2861{
2862 struct request *rq;
2863 int ret;
2864
2fff8a92 2865 lockdep_assert_held(q->queue_lock);
332ebbf7 2866 WARN_ON_ONCE(q->mq_ops);
2fff8a92 2867
9c988374 2868 while ((rq = elv_next_request(q)) != NULL) {
e8064021 2869 if (!(rq->rq_flags & RQF_STARTED)) {
158dbda0
TH
2870 /*
2871 * This is the first time the device driver
2872 * sees this request (possibly after
2873 * requeueing). Notify IO scheduler.
2874 */
e8064021 2875 if (rq->rq_flags & RQF_SORTED)
158dbda0
TH
2876 elv_activate_rq(q, rq);
2877
2878 /*
2879 * just mark as started even if we don't start
2880 * it, a request that has been delayed should
2881 * not be passed by new incoming requests
2882 */
e8064021 2883 rq->rq_flags |= RQF_STARTED;
158dbda0
TH
2884 trace_block_rq_issue(q, rq);
2885 }
2886
2887 if (!q->boundary_rq || q->boundary_rq == rq) {
2888 q->end_sector = rq_end_sector(rq);
2889 q->boundary_rq = NULL;
2890 }
2891
e8064021 2892 if (rq->rq_flags & RQF_DONTPREP)
158dbda0
TH
2893 break;
2894
2e46e8b2 2895 if (q->dma_drain_size && blk_rq_bytes(rq)) {
158dbda0
TH
2896 /*
2897 * make sure space for the drain appears we
2898 * know we can do this because max_hw_segments
2899 * has been adjusted to be one fewer than the
2900 * device can handle
2901 */
2902 rq->nr_phys_segments++;
2903 }
2904
2905 if (!q->prep_rq_fn)
2906 break;
2907
2908 ret = q->prep_rq_fn(q, rq);
2909 if (ret == BLKPREP_OK) {
2910 break;
2911 } else if (ret == BLKPREP_DEFER) {
2912 /*
2913 * the request may have been (partially) prepped.
2914 * we need to keep this request in the front to
e8064021 2915 * avoid resource deadlock. RQF_STARTED will
158dbda0
TH
2916 * prevent other fs requests from passing this one.
2917 */
2e46e8b2 2918 if (q->dma_drain_size && blk_rq_bytes(rq) &&
e8064021 2919 !(rq->rq_flags & RQF_DONTPREP)) {
158dbda0
TH
2920 /*
2921 * remove the space for the drain we added
2922 * so that we don't add it again
2923 */
2924 --rq->nr_phys_segments;
2925 }
2926
2927 rq = NULL;
2928 break;
0fb5b1fb 2929 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
e8064021 2930 rq->rq_flags |= RQF_QUIET;
c143dc90
JB
2931 /*
2932 * Mark this request as started so we don't trigger
2933 * any debug logic in the end I/O path.
2934 */
2935 blk_start_request(rq);
2a842aca
CH
2936 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2937 BLK_STS_TARGET : BLK_STS_IOERR);
158dbda0
TH
2938 } else {
2939 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2940 break;
2941 }
2942 }
2943
2944 return rq;
2945}
9934c8c0 2946EXPORT_SYMBOL(blk_peek_request);
158dbda0 2947
5034435c 2948static void blk_dequeue_request(struct request *rq)
158dbda0 2949{
9934c8c0
TH
2950 struct request_queue *q = rq->q;
2951
158dbda0
TH
2952 BUG_ON(list_empty(&rq->queuelist));
2953 BUG_ON(ELV_ON_HASH(rq));
2954
2955 list_del_init(&rq->queuelist);
2956
2957 /*
2958 * the time frame between a request being removed from the lists
2959 * and to it is freed is accounted as io that is in progress at
2960 * the driver side.
2961 */
522a7775 2962 if (blk_account_rq(rq))
0a7ae2ff 2963 q->in_flight[rq_is_sync(rq)]++;
158dbda0
TH
2964}
2965
9934c8c0
TH
2966/**
2967 * blk_start_request - start request processing on the driver
2968 * @req: request to dequeue
2969 *
2970 * Description:
2971 * Dequeue @req and start timeout timer on it. This hands off the
2972 * request to the driver.
9934c8c0
TH
2973 */
2974void blk_start_request(struct request *req)
2975{
2fff8a92 2976 lockdep_assert_held(req->q->queue_lock);
332ebbf7 2977 WARN_ON_ONCE(req->q->mq_ops);
2fff8a92 2978
9934c8c0
TH
2979 blk_dequeue_request(req);
2980
cf43e6be 2981 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
544ccc8d
OS
2982 req->io_start_time_ns = ktime_get_ns();
2983#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2984 req->throtl_size = blk_rq_sectors(req);
2985#endif
cf43e6be 2986 req->rq_flags |= RQF_STATS;
a8a45941 2987 wbt_issue(req->q->rq_wb, req);
cf43e6be
JA
2988 }
2989
e14575b3 2990 BUG_ON(blk_rq_is_complete(req));
9934c8c0
TH
2991 blk_add_timer(req);
2992}
2993EXPORT_SYMBOL(blk_start_request);
2994
2995/**
2996 * blk_fetch_request - fetch a request from a request queue
2997 * @q: request queue to fetch a request from
2998 *
2999 * Description:
3000 * Return the request at the top of @q. The request is started on
3001 * return and LLD can start processing it immediately.
3002 *
3003 * Return:
3004 * Pointer to the request at the top of @q if available. Null
3005 * otherwise.
9934c8c0
TH
3006 */
3007struct request *blk_fetch_request(struct request_queue *q)
3008{
3009 struct request *rq;
3010
2fff8a92 3011 lockdep_assert_held(q->queue_lock);
332ebbf7 3012 WARN_ON_ONCE(q->mq_ops);
2fff8a92 3013
9934c8c0
TH
3014 rq = blk_peek_request(q);
3015 if (rq)
3016 blk_start_request(rq);
3017 return rq;
3018}
3019EXPORT_SYMBOL(blk_fetch_request);
3020
ef71de8b
CH
3021/*
3022 * Steal bios from a request and add them to a bio list.
3023 * The request must not have been partially completed before.
3024 */
3025void blk_steal_bios(struct bio_list *list, struct request *rq)
3026{
3027 if (rq->bio) {
3028 if (list->tail)
3029 list->tail->bi_next = rq->bio;
3030 else
3031 list->head = rq->bio;
3032 list->tail = rq->biotail;
3033
3034 rq->bio = NULL;
3035 rq->biotail = NULL;
3036 }
3037
3038 rq->__data_len = 0;
3039}
3040EXPORT_SYMBOL_GPL(blk_steal_bios);
3041
3bcddeac 3042/**
2e60e022 3043 * blk_update_request - Special helper function for request stacking drivers
8ebf9756 3044 * @req: the request being processed
2a842aca 3045 * @error: block status code
8ebf9756 3046 * @nr_bytes: number of bytes to complete @req
3bcddeac
KU
3047 *
3048 * Description:
8ebf9756
RD
3049 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3050 * the request structure even if @req doesn't have leftover.
3051 * If @req has leftover, sets it up for the next range of segments.
2e60e022
TH
3052 *
3053 * This special helper function is only for request stacking drivers
3054 * (e.g. request-based dm) so that they can handle partial completion.
3055 * Actual device drivers should use blk_end_request instead.
3056 *
3057 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3058 * %false return from this function.
3bcddeac
KU
3059 *
3060 * Return:
2e60e022
TH
3061 * %false - this request doesn't have any more data
3062 * %true - this request has more data
3bcddeac 3063 **/
2a842aca
CH
3064bool blk_update_request(struct request *req, blk_status_t error,
3065 unsigned int nr_bytes)
1da177e4 3066{
f79ea416 3067 int total_bytes;
1da177e4 3068
2a842aca 3069 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
4a0efdc9 3070
2e60e022
TH
3071 if (!req->bio)
3072 return false;
3073
2a842aca
CH
3074 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3075 !(req->rq_flags & RQF_QUIET)))
3076 print_req_error(req, error);
1da177e4 3077
bc58ba94 3078 blk_account_io_completion(req, nr_bytes);
d72d904a 3079
f79ea416
KO
3080 total_bytes = 0;
3081 while (req->bio) {
3082 struct bio *bio = req->bio;
4f024f37 3083 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1da177e4 3084
0ba99ca4 3085 if (bio_bytes == bio->bi_iter.bi_size) {
1da177e4 3086 req->bio = bio->bi_next;
0ba99ca4
KO
3087 bio->bi_next = NULL;
3088 }
1da177e4 3089
fbbaf700
N
3090 /* Completion has already been traced */
3091 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
f79ea416 3092 req_bio_endio(req, bio, bio_bytes, error);
1da177e4 3093
f79ea416
KO
3094 total_bytes += bio_bytes;
3095 nr_bytes -= bio_bytes;
1da177e4 3096
f79ea416
KO
3097 if (!nr_bytes)
3098 break;
1da177e4
LT
3099 }
3100
3101 /*
3102 * completely done
3103 */
2e60e022
TH
3104 if (!req->bio) {
3105 /*
3106 * Reset counters so that the request stacking driver
3107 * can find how many bytes remain in the request
3108 * later.
3109 */
a2dec7b3 3110 req->__data_len = 0;
2e60e022
TH
3111 return false;
3112 }
1da177e4 3113
a2dec7b3 3114 req->__data_len -= total_bytes;
2e46e8b2
TH
3115
3116 /* update sector only for requests with clear definition of sector */
57292b58 3117 if (!blk_rq_is_passthrough(req))
a2dec7b3 3118 req->__sector += total_bytes >> 9;
2e46e8b2 3119
80a761fd 3120 /* mixed attributes always follow the first bio */
e8064021 3121 if (req->rq_flags & RQF_MIXED_MERGE) {
80a761fd 3122 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1eff9d32 3123 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
80a761fd
TH
3124 }
3125
ed6565e7
CH
3126 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3127 /*
3128 * If total number of sectors is less than the first segment
3129 * size, something has gone terribly wrong.
3130 */
3131 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3132 blk_dump_rq_flags(req, "request botched");
3133 req->__data_len = blk_rq_cur_bytes(req);
3134 }
2e46e8b2 3135
ed6565e7
CH
3136 /* recalculate the number of segments */
3137 blk_recalc_rq_segments(req);
3138 }
2e46e8b2 3139
2e60e022 3140 return true;
1da177e4 3141}
2e60e022 3142EXPORT_SYMBOL_GPL(blk_update_request);
1da177e4 3143
2a842aca 3144static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2e60e022
TH
3145 unsigned int nr_bytes,
3146 unsigned int bidi_bytes)
5efccd17 3147{
2e60e022
TH
3148 if (blk_update_request(rq, error, nr_bytes))
3149 return true;
5efccd17 3150
2e60e022
TH
3151 /* Bidi request must be completed as a whole */
3152 if (unlikely(blk_bidi_rq(rq)) &&
3153 blk_update_request(rq->next_rq, error, bidi_bytes))
3154 return true;
5efccd17 3155
e2e1a148
JA
3156 if (blk_queue_add_random(rq->q))
3157 add_disk_randomness(rq->rq_disk);
2e60e022
TH
3158
3159 return false;
1da177e4
LT
3160}
3161
28018c24
JB
3162/**
3163 * blk_unprep_request - unprepare a request
3164 * @req: the request
3165 *
3166 * This function makes a request ready for complete resubmission (or
3167 * completion). It happens only after all error handling is complete,
3168 * so represents the appropriate moment to deallocate any resources
3169 * that were allocated to the request in the prep_rq_fn. The queue
3170 * lock is held when calling this.
3171 */
3172void blk_unprep_request(struct request *req)
3173{
3174 struct request_queue *q = req->q;
3175
e8064021 3176 req->rq_flags &= ~RQF_DONTPREP;
28018c24
JB
3177 if (q->unprep_rq_fn)
3178 q->unprep_rq_fn(q, req);
3179}
3180EXPORT_SYMBOL_GPL(blk_unprep_request);
3181
2a842aca 3182void blk_finish_request(struct request *req, blk_status_t error)
1da177e4 3183{
cf43e6be 3184 struct request_queue *q = req->q;
522a7775 3185 u64 now = ktime_get_ns();
cf43e6be 3186
2fff8a92 3187 lockdep_assert_held(req->q->queue_lock);
332ebbf7 3188 WARN_ON_ONCE(q->mq_ops);
2fff8a92 3189
cf43e6be 3190 if (req->rq_flags & RQF_STATS)
522a7775 3191 blk_stat_add(req, now);
cf43e6be 3192
e8064021 3193 if (req->rq_flags & RQF_QUEUED)
cf43e6be 3194 blk_queue_end_tag(q, req);
b8286239 3195
ba396a6c 3196 BUG_ON(blk_queued_rq(req));
1da177e4 3197
57292b58 3198 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
dc3b17cc 3199 laptop_io_completion(req->q->backing_dev_info);
1da177e4 3200
e78042e5
MA
3201 blk_delete_timer(req);
3202
e8064021 3203 if (req->rq_flags & RQF_DONTPREP)
28018c24
JB
3204 blk_unprep_request(req);
3205
522a7775 3206 blk_account_io_done(req, now);
b8286239 3207
87760e5e 3208 if (req->end_io) {
a8a45941 3209 wbt_done(req->q->rq_wb, req);
8ffdc655 3210 req->end_io(req, error);
87760e5e 3211 } else {
b8286239
KU
3212 if (blk_bidi_rq(req))
3213 __blk_put_request(req->next_rq->q, req->next_rq);
3214
cf43e6be 3215 __blk_put_request(q, req);
b8286239 3216 }
1da177e4 3217}
12120077 3218EXPORT_SYMBOL(blk_finish_request);
1da177e4 3219
3b11313a 3220/**
2e60e022
TH
3221 * blk_end_bidi_request - Complete a bidi request
3222 * @rq: the request to complete
2a842aca 3223 * @error: block status code
2e60e022
TH
3224 * @nr_bytes: number of bytes to complete @rq
3225 * @bidi_bytes: number of bytes to complete @rq->next_rq
a0cd1285
JA
3226 *
3227 * Description:
e3a04fe3 3228 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2e60e022
TH
3229 * Drivers that supports bidi can safely call this member for any
3230 * type of request, bidi or uni. In the later case @bidi_bytes is
3231 * just ignored.
336cdb40
KU
3232 *
3233 * Return:
2e60e022
TH
3234 * %false - we are done with this request
3235 * %true - still buffers pending for this request
a0cd1285 3236 **/
2a842aca 3237static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
32fab448
KU
3238 unsigned int nr_bytes, unsigned int bidi_bytes)
3239{
336cdb40 3240 struct request_queue *q = rq->q;
2e60e022 3241 unsigned long flags;
32fab448 3242
332ebbf7
BVA
3243 WARN_ON_ONCE(q->mq_ops);
3244
2e60e022
TH
3245 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3246 return true;
32fab448 3247
336cdb40 3248 spin_lock_irqsave(q->queue_lock, flags);
2e60e022 3249 blk_finish_request(rq, error);
336cdb40
KU
3250 spin_unlock_irqrestore(q->queue_lock, flags);
3251
2e60e022 3252 return false;
32fab448
KU
3253}
3254
336cdb40 3255/**
2e60e022
TH
3256 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3257 * @rq: the request to complete
2a842aca 3258 * @error: block status code
e3a04fe3
KU
3259 * @nr_bytes: number of bytes to complete @rq
3260 * @bidi_bytes: number of bytes to complete @rq->next_rq
336cdb40
KU
3261 *
3262 * Description:
2e60e022
TH
3263 * Identical to blk_end_bidi_request() except that queue lock is
3264 * assumed to be locked on entry and remains so on return.
336cdb40
KU
3265 *
3266 * Return:
2e60e022
TH
3267 * %false - we are done with this request
3268 * %true - still buffers pending for this request
336cdb40 3269 **/
2a842aca 3270static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
b1f74493 3271 unsigned int nr_bytes, unsigned int bidi_bytes)
336cdb40 3272{
2fff8a92 3273 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3274 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3275
2e60e022
TH
3276 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3277 return true;
336cdb40 3278
2e60e022 3279 blk_finish_request(rq, error);
336cdb40 3280
2e60e022 3281 return false;
336cdb40 3282}
e19a3ab0
KU
3283
3284/**
3285 * blk_end_request - Helper function for drivers to complete the request.
3286 * @rq: the request being processed
2a842aca 3287 * @error: block status code
e19a3ab0
KU
3288 * @nr_bytes: number of bytes to complete
3289 *
3290 * Description:
3291 * Ends I/O on a number of bytes attached to @rq.
3292 * If @rq has leftover, sets it up for the next range of segments.
3293 *
3294 * Return:
b1f74493
FT
3295 * %false - we are done with this request
3296 * %true - still buffers pending for this request
e19a3ab0 3297 **/
2a842aca
CH
3298bool blk_end_request(struct request *rq, blk_status_t error,
3299 unsigned int nr_bytes)
e19a3ab0 3300{
332ebbf7 3301 WARN_ON_ONCE(rq->q->mq_ops);
b1f74493 3302 return blk_end_bidi_request(rq, error, nr_bytes, 0);
e19a3ab0 3303}
56ad1740 3304EXPORT_SYMBOL(blk_end_request);
336cdb40
KU
3305
3306/**
b1f74493
FT
3307 * blk_end_request_all - Helper function for drives to finish the request.
3308 * @rq: the request to finish
2a842aca 3309 * @error: block status code
336cdb40
KU
3310 *
3311 * Description:
b1f74493
FT
3312 * Completely finish @rq.
3313 */
2a842aca 3314void blk_end_request_all(struct request *rq, blk_status_t error)
336cdb40 3315{
b1f74493
FT
3316 bool pending;
3317 unsigned int bidi_bytes = 0;
336cdb40 3318
b1f74493
FT
3319 if (unlikely(blk_bidi_rq(rq)))
3320 bidi_bytes = blk_rq_bytes(rq->next_rq);
336cdb40 3321
b1f74493
FT
3322 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3323 BUG_ON(pending);
3324}
56ad1740 3325EXPORT_SYMBOL(blk_end_request_all);
336cdb40 3326
e3a04fe3 3327/**
b1f74493
FT
3328 * __blk_end_request - Helper function for drivers to complete the request.
3329 * @rq: the request being processed
2a842aca 3330 * @error: block status code
b1f74493 3331 * @nr_bytes: number of bytes to complete
e3a04fe3
KU
3332 *
3333 * Description:
b1f74493 3334 * Must be called with queue lock held unlike blk_end_request().
e3a04fe3
KU
3335 *
3336 * Return:
b1f74493
FT
3337 * %false - we are done with this request
3338 * %true - still buffers pending for this request
e3a04fe3 3339 **/
2a842aca
CH
3340bool __blk_end_request(struct request *rq, blk_status_t error,
3341 unsigned int nr_bytes)
e3a04fe3 3342{
2fff8a92 3343 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3344 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3345
b1f74493 3346 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
e3a04fe3 3347}
56ad1740 3348EXPORT_SYMBOL(__blk_end_request);
e3a04fe3 3349
32fab448 3350/**
b1f74493
FT
3351 * __blk_end_request_all - Helper function for drives to finish the request.
3352 * @rq: the request to finish
2a842aca 3353 * @error: block status code
32fab448
KU
3354 *
3355 * Description:
b1f74493 3356 * Completely finish @rq. Must be called with queue lock held.
32fab448 3357 */
2a842aca 3358void __blk_end_request_all(struct request *rq, blk_status_t error)
32fab448 3359{
b1f74493
FT
3360 bool pending;
3361 unsigned int bidi_bytes = 0;
3362
2fff8a92 3363 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3364 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3365
b1f74493
FT
3366 if (unlikely(blk_bidi_rq(rq)))
3367 bidi_bytes = blk_rq_bytes(rq->next_rq);
3368
3369 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3370 BUG_ON(pending);
32fab448 3371}
56ad1740 3372EXPORT_SYMBOL(__blk_end_request_all);
32fab448 3373
e19a3ab0 3374/**
b1f74493
FT
3375 * __blk_end_request_cur - Helper function to finish the current request chunk.
3376 * @rq: the request to finish the current chunk for
2a842aca 3377 * @error: block status code
e19a3ab0
KU
3378 *
3379 * Description:
b1f74493
FT
3380 * Complete the current consecutively mapped chunk from @rq. Must
3381 * be called with queue lock held.
e19a3ab0
KU
3382 *
3383 * Return:
b1f74493
FT
3384 * %false - we are done with this request
3385 * %true - still buffers pending for this request
3386 */
2a842aca 3387bool __blk_end_request_cur(struct request *rq, blk_status_t error)
e19a3ab0 3388{
b1f74493 3389 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
e19a3ab0 3390}
56ad1740 3391EXPORT_SYMBOL(__blk_end_request_cur);
e19a3ab0 3392
86db1e29
JA
3393void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3394 struct bio *bio)
1da177e4 3395{
b4f42e28 3396 if (bio_has_data(bio))
fb2dce86 3397 rq->nr_phys_segments = bio_phys_segments(q, bio);
445251d0
JA
3398 else if (bio_op(bio) == REQ_OP_DISCARD)
3399 rq->nr_phys_segments = 1;
b4f42e28 3400
4f024f37 3401 rq->__data_len = bio->bi_iter.bi_size;
1da177e4 3402 rq->bio = rq->biotail = bio;
1da177e4 3403
74d46992
CH
3404 if (bio->bi_disk)
3405 rq->rq_disk = bio->bi_disk;
66846572 3406}
1da177e4 3407
2d4dc890
IL
3408#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3409/**
3410 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3411 * @rq: the request to be flushed
3412 *
3413 * Description:
3414 * Flush all pages in @rq.
3415 */
3416void rq_flush_dcache_pages(struct request *rq)
3417{
3418 struct req_iterator iter;
7988613b 3419 struct bio_vec bvec;
2d4dc890
IL
3420
3421 rq_for_each_segment(bvec, rq, iter)
7988613b 3422 flush_dcache_page(bvec.bv_page);
2d4dc890
IL
3423}
3424EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3425#endif
3426
ef9e3fac
KU
3427/**
3428 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3429 * @q : the queue of the device being checked
3430 *
3431 * Description:
3432 * Check if underlying low-level drivers of a device are busy.
3433 * If the drivers want to export their busy state, they must set own
3434 * exporting function using blk_queue_lld_busy() first.
3435 *
3436 * Basically, this function is used only by request stacking drivers
3437 * to stop dispatching requests to underlying devices when underlying
3438 * devices are busy. This behavior helps more I/O merging on the queue
3439 * of the request stacking driver and prevents I/O throughput regression
3440 * on burst I/O load.
3441 *
3442 * Return:
3443 * 0 - Not busy (The request stacking driver should dispatch request)
3444 * 1 - Busy (The request stacking driver should stop dispatching request)
3445 */
3446int blk_lld_busy(struct request_queue *q)
3447{
3448 if (q->lld_busy_fn)
3449 return q->lld_busy_fn(q);
3450
3451 return 0;
3452}
3453EXPORT_SYMBOL_GPL(blk_lld_busy);
3454
78d8e58a
MS
3455/**
3456 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3457 * @rq: the clone request to be cleaned up
3458 *
3459 * Description:
3460 * Free all bios in @rq for a cloned request.
3461 */
3462void blk_rq_unprep_clone(struct request *rq)
3463{
3464 struct bio *bio;
3465
3466 while ((bio = rq->bio) != NULL) {
3467 rq->bio = bio->bi_next;
3468
3469 bio_put(bio);
3470 }
3471}
3472EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3473
3474/*
3475 * Copy attributes of the original request to the clone request.
3476 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3477 */
3478static void __blk_rq_prep_clone(struct request *dst, struct request *src)
b0fd271d
KU
3479{
3480 dst->cpu = src->cpu;
b0fd271d
KU
3481 dst->__sector = blk_rq_pos(src);
3482 dst->__data_len = blk_rq_bytes(src);
3483 dst->nr_phys_segments = src->nr_phys_segments;
3484 dst->ioprio = src->ioprio;
3485 dst->extra_len = src->extra_len;
78d8e58a
MS
3486}
3487
3488/**
3489 * blk_rq_prep_clone - Helper function to setup clone request
3490 * @rq: the request to be setup
3491 * @rq_src: original request to be cloned
3492 * @bs: bio_set that bios for clone are allocated from
3493 * @gfp_mask: memory allocation mask for bio
3494 * @bio_ctr: setup function to be called for each clone bio.
3495 * Returns %0 for success, non %0 for failure.
3496 * @data: private data to be passed to @bio_ctr
3497 *
3498 * Description:
3499 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3500 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3501 * are not copied, and copying such parts is the caller's responsibility.
3502 * Also, pages which the original bios are pointing to are not copied
3503 * and the cloned bios just point same pages.
3504 * So cloned bios must be completed before original bios, which means
3505 * the caller must complete @rq before @rq_src.
3506 */
3507int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3508 struct bio_set *bs, gfp_t gfp_mask,
3509 int (*bio_ctr)(struct bio *, struct bio *, void *),
3510 void *data)
3511{
3512 struct bio *bio, *bio_src;
3513
3514 if (!bs)
f4f8154a 3515 bs = &fs_bio_set;
78d8e58a
MS
3516
3517 __rq_for_each_bio(bio_src, rq_src) {
3518 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3519 if (!bio)
3520 goto free_and_out;
3521
3522 if (bio_ctr && bio_ctr(bio, bio_src, data))
3523 goto free_and_out;
3524
3525 if (rq->bio) {
3526 rq->biotail->bi_next = bio;
3527 rq->biotail = bio;
3528 } else
3529 rq->bio = rq->biotail = bio;
3530 }
3531
3532 __blk_rq_prep_clone(rq, rq_src);
3533
3534 return 0;
3535
3536free_and_out:
3537 if (bio)
3538 bio_put(bio);
3539 blk_rq_unprep_clone(rq);
3540
3541 return -ENOMEM;
b0fd271d
KU
3542}
3543EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3544
59c3d45e 3545int kblockd_schedule_work(struct work_struct *work)
1da177e4
LT
3546{
3547 return queue_work(kblockd_workqueue, work);
3548}
1da177e4
LT
3549EXPORT_SYMBOL(kblockd_schedule_work);
3550
ee63cfa7
JA
3551int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3552{
3553 return queue_work_on(cpu, kblockd_workqueue, work);
3554}
3555EXPORT_SYMBOL(kblockd_schedule_work_on);
3556
818cd1cb
JA
3557int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3558 unsigned long delay)
3559{
3560 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3561}
3562EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3563
75df7136
SJ
3564/**
3565 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3566 * @plug: The &struct blk_plug that needs to be initialized
3567 *
3568 * Description:
3569 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3570 * pending I/O should the task end up blocking between blk_start_plug() and
3571 * blk_finish_plug(). This is important from a performance perspective, but
3572 * also ensures that we don't deadlock. For instance, if the task is blocking
3573 * for a memory allocation, memory reclaim could end up wanting to free a
3574 * page belonging to that request that is currently residing in our private
3575 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3576 * this kind of deadlock.
3577 */
73c10101
JA
3578void blk_start_plug(struct blk_plug *plug)
3579{
3580 struct task_struct *tsk = current;
3581
dd6cf3e1
SL
3582 /*
3583 * If this is a nested plug, don't actually assign it.
3584 */
3585 if (tsk->plug)
3586 return;
3587
73c10101 3588 INIT_LIST_HEAD(&plug->list);
320ae51f 3589 INIT_LIST_HEAD(&plug->mq_list);
048c9374 3590 INIT_LIST_HEAD(&plug->cb_list);
73c10101 3591 /*
dd6cf3e1
SL
3592 * Store ordering should not be needed here, since a potential
3593 * preempt will imply a full memory barrier
73c10101 3594 */
dd6cf3e1 3595 tsk->plug = plug;
73c10101
JA
3596}
3597EXPORT_SYMBOL(blk_start_plug);
3598
3599static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3600{
3601 struct request *rqa = container_of(a, struct request, queuelist);
3602 struct request *rqb = container_of(b, struct request, queuelist);
3603
975927b9
JM
3604 return !(rqa->q < rqb->q ||
3605 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
73c10101
JA
3606}
3607
49cac01e
JA
3608/*
3609 * If 'from_schedule' is true, then postpone the dispatch of requests
3610 * until a safe kblockd context. We due this to avoid accidental big
3611 * additional stack usage in driver dispatch, in places where the originally
3612 * plugger did not intend it.
3613 */
f6603783 3614static void queue_unplugged(struct request_queue *q, unsigned int depth,
49cac01e 3615 bool from_schedule)
99e22598 3616 __releases(q->queue_lock)
94b5eb28 3617{
2fff8a92
BVA
3618 lockdep_assert_held(q->queue_lock);
3619
49cac01e 3620 trace_block_unplug(q, depth, !from_schedule);
99e22598 3621
70460571 3622 if (from_schedule)
24ecfbe2 3623 blk_run_queue_async(q);
70460571 3624 else
24ecfbe2 3625 __blk_run_queue(q);
50864670 3626 spin_unlock_irq(q->queue_lock);
94b5eb28
JA
3627}
3628
74018dc3 3629static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
048c9374
N
3630{
3631 LIST_HEAD(callbacks);
3632
2a7d5559
SL
3633 while (!list_empty(&plug->cb_list)) {
3634 list_splice_init(&plug->cb_list, &callbacks);
048c9374 3635
2a7d5559
SL
3636 while (!list_empty(&callbacks)) {
3637 struct blk_plug_cb *cb = list_first_entry(&callbacks,
048c9374
N
3638 struct blk_plug_cb,
3639 list);
2a7d5559 3640 list_del(&cb->list);
74018dc3 3641 cb->callback(cb, from_schedule);
2a7d5559 3642 }
048c9374
N
3643 }
3644}
3645
9cbb1750
N
3646struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3647 int size)
3648{
3649 struct blk_plug *plug = current->plug;
3650 struct blk_plug_cb *cb;
3651
3652 if (!plug)
3653 return NULL;
3654
3655 list_for_each_entry(cb, &plug->cb_list, list)
3656 if (cb->callback == unplug && cb->data == data)
3657 return cb;
3658
3659 /* Not currently on the callback list */
3660 BUG_ON(size < sizeof(*cb));
3661 cb = kzalloc(size, GFP_ATOMIC);
3662 if (cb) {
3663 cb->data = data;
3664 cb->callback = unplug;
3665 list_add(&cb->list, &plug->cb_list);
3666 }
3667 return cb;
3668}
3669EXPORT_SYMBOL(blk_check_plugged);
3670
49cac01e 3671void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
73c10101
JA
3672{
3673 struct request_queue *q;
73c10101 3674 struct request *rq;
109b8129 3675 LIST_HEAD(list);
94b5eb28 3676 unsigned int depth;
73c10101 3677
74018dc3 3678 flush_plug_callbacks(plug, from_schedule);
320ae51f
JA
3679
3680 if (!list_empty(&plug->mq_list))
3681 blk_mq_flush_plug_list(plug, from_schedule);
3682
73c10101
JA
3683 if (list_empty(&plug->list))
3684 return;
3685
109b8129
N
3686 list_splice_init(&plug->list, &list);
3687
422765c2 3688 list_sort(NULL, &list, plug_rq_cmp);
73c10101
JA
3689
3690 q = NULL;
94b5eb28 3691 depth = 0;
18811272 3692
109b8129
N
3693 while (!list_empty(&list)) {
3694 rq = list_entry_rq(list.next);
73c10101 3695 list_del_init(&rq->queuelist);
73c10101
JA
3696 BUG_ON(!rq->q);
3697 if (rq->q != q) {
99e22598
JA
3698 /*
3699 * This drops the queue lock
3700 */
3701 if (q)
49cac01e 3702 queue_unplugged(q, depth, from_schedule);
73c10101 3703 q = rq->q;
94b5eb28 3704 depth = 0;
50864670 3705 spin_lock_irq(q->queue_lock);
73c10101 3706 }
8ba61435
TH
3707
3708 /*
3709 * Short-circuit if @q is dead
3710 */
3f3299d5 3711 if (unlikely(blk_queue_dying(q))) {
2a842aca 3712 __blk_end_request_all(rq, BLK_STS_IOERR);
8ba61435
TH
3713 continue;
3714 }
3715
73c10101
JA
3716 /*
3717 * rq is already accounted, so use raw insert
3718 */
f73f44eb 3719 if (op_is_flush(rq->cmd_flags))
401a18e9
JA
3720 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3721 else
3722 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
94b5eb28
JA
3723
3724 depth++;
73c10101
JA
3725 }
3726
99e22598
JA
3727 /*
3728 * This drops the queue lock
3729 */
3730 if (q)
49cac01e 3731 queue_unplugged(q, depth, from_schedule);
73c10101 3732}
73c10101
JA
3733
3734void blk_finish_plug(struct blk_plug *plug)
3735{
dd6cf3e1
SL
3736 if (plug != current->plug)
3737 return;
f6603783 3738 blk_flush_plug_list(plug, false);
73c10101 3739
dd6cf3e1 3740 current->plug = NULL;
73c10101 3741}
88b996cd 3742EXPORT_SYMBOL(blk_finish_plug);
73c10101 3743
47fafbc7 3744#ifdef CONFIG_PM
6c954667
LM
3745/**
3746 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3747 * @q: the queue of the device
3748 * @dev: the device the queue belongs to
3749 *
3750 * Description:
3751 * Initialize runtime-PM-related fields for @q and start auto suspend for
3752 * @dev. Drivers that want to take advantage of request-based runtime PM
3753 * should call this function after @dev has been initialized, and its
3754 * request queue @q has been allocated, and runtime PM for it can not happen
3755 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3756 * cases, driver should call this function before any I/O has taken place.
3757 *
3758 * This function takes care of setting up using auto suspend for the device,
3759 * the autosuspend delay is set to -1 to make runtime suspend impossible
3760 * until an updated value is either set by user or by driver. Drivers do
3761 * not need to touch other autosuspend settings.
3762 *
3763 * The block layer runtime PM is request based, so only works for drivers
3764 * that use request as their IO unit instead of those directly use bio's.
3765 */
3766void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3767{
765e40b6
CH
3768 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3769 if (q->mq_ops)
3770 return;
3771
6c954667
LM
3772 q->dev = dev;
3773 q->rpm_status = RPM_ACTIVE;
3774 pm_runtime_set_autosuspend_delay(q->dev, -1);
3775 pm_runtime_use_autosuspend(q->dev);
3776}
3777EXPORT_SYMBOL(blk_pm_runtime_init);
3778
3779/**
3780 * blk_pre_runtime_suspend - Pre runtime suspend check
3781 * @q: the queue of the device
3782 *
3783 * Description:
3784 * This function will check if runtime suspend is allowed for the device
3785 * by examining if there are any requests pending in the queue. If there
3786 * are requests pending, the device can not be runtime suspended; otherwise,
3787 * the queue's status will be updated to SUSPENDING and the driver can
3788 * proceed to suspend the device.
3789 *
3790 * For the not allowed case, we mark last busy for the device so that
3791 * runtime PM core will try to autosuspend it some time later.
3792 *
3793 * This function should be called near the start of the device's
3794 * runtime_suspend callback.
3795 *
3796 * Return:
3797 * 0 - OK to runtime suspend the device
3798 * -EBUSY - Device should not be runtime suspended
3799 */
3800int blk_pre_runtime_suspend(struct request_queue *q)
3801{
3802 int ret = 0;
3803
4fd41a85
KX
3804 if (!q->dev)
3805 return ret;
3806
6c954667
LM
3807 spin_lock_irq(q->queue_lock);
3808 if (q->nr_pending) {
3809 ret = -EBUSY;
3810 pm_runtime_mark_last_busy(q->dev);
3811 } else {
3812 q->rpm_status = RPM_SUSPENDING;
3813 }
3814 spin_unlock_irq(q->queue_lock);
3815 return ret;
3816}
3817EXPORT_SYMBOL(blk_pre_runtime_suspend);
3818
3819/**
3820 * blk_post_runtime_suspend - Post runtime suspend processing
3821 * @q: the queue of the device
3822 * @err: return value of the device's runtime_suspend function
3823 *
3824 * Description:
3825 * Update the queue's runtime status according to the return value of the
3826 * device's runtime suspend function and mark last busy for the device so
3827 * that PM core will try to auto suspend the device at a later time.
3828 *
3829 * This function should be called near the end of the device's
3830 * runtime_suspend callback.
3831 */
3832void blk_post_runtime_suspend(struct request_queue *q, int err)
3833{
4fd41a85
KX
3834 if (!q->dev)
3835 return;
3836
6c954667
LM
3837 spin_lock_irq(q->queue_lock);
3838 if (!err) {
3839 q->rpm_status = RPM_SUSPENDED;
3840 } else {
3841 q->rpm_status = RPM_ACTIVE;
3842 pm_runtime_mark_last_busy(q->dev);
3843 }
3844 spin_unlock_irq(q->queue_lock);
3845}
3846EXPORT_SYMBOL(blk_post_runtime_suspend);
3847
3848/**
3849 * blk_pre_runtime_resume - Pre runtime resume processing
3850 * @q: the queue of the device
3851 *
3852 * Description:
3853 * Update the queue's runtime status to RESUMING in preparation for the
3854 * runtime resume of the device.
3855 *
3856 * This function should be called near the start of the device's
3857 * runtime_resume callback.
3858 */
3859void blk_pre_runtime_resume(struct request_queue *q)
3860{
4fd41a85
KX
3861 if (!q->dev)
3862 return;
3863
6c954667
LM
3864 spin_lock_irq(q->queue_lock);
3865 q->rpm_status = RPM_RESUMING;
3866 spin_unlock_irq(q->queue_lock);
3867}
3868EXPORT_SYMBOL(blk_pre_runtime_resume);
3869
3870/**
3871 * blk_post_runtime_resume - Post runtime resume processing
3872 * @q: the queue of the device
3873 * @err: return value of the device's runtime_resume function
3874 *
3875 * Description:
3876 * Update the queue's runtime status according to the return value of the
3877 * device's runtime_resume function. If it is successfully resumed, process
3878 * the requests that are queued into the device's queue when it is resuming
3879 * and then mark last busy and initiate autosuspend for it.
3880 *
3881 * This function should be called near the end of the device's
3882 * runtime_resume callback.
3883 */
3884void blk_post_runtime_resume(struct request_queue *q, int err)
3885{
4fd41a85
KX
3886 if (!q->dev)
3887 return;
3888
6c954667
LM
3889 spin_lock_irq(q->queue_lock);
3890 if (!err) {
3891 q->rpm_status = RPM_ACTIVE;
3892 __blk_run_queue(q);
3893 pm_runtime_mark_last_busy(q->dev);
c60855cd 3894 pm_request_autosuspend(q->dev);
6c954667
LM
3895 } else {
3896 q->rpm_status = RPM_SUSPENDED;
3897 }
3898 spin_unlock_irq(q->queue_lock);
3899}
3900EXPORT_SYMBOL(blk_post_runtime_resume);
d07ab6d1
MW
3901
3902/**
3903 * blk_set_runtime_active - Force runtime status of the queue to be active
3904 * @q: the queue of the device
3905 *
3906 * If the device is left runtime suspended during system suspend the resume
3907 * hook typically resumes the device and corrects runtime status
3908 * accordingly. However, that does not affect the queue runtime PM status
3909 * which is still "suspended". This prevents processing requests from the
3910 * queue.
3911 *
3912 * This function can be used in driver's resume hook to correct queue
3913 * runtime PM status and re-enable peeking requests from the queue. It
3914 * should be called before first request is added to the queue.
3915 */
3916void blk_set_runtime_active(struct request_queue *q)
3917{
3918 spin_lock_irq(q->queue_lock);
3919 q->rpm_status = RPM_ACTIVE;
3920 pm_runtime_mark_last_busy(q->dev);
3921 pm_request_autosuspend(q->dev);
3922 spin_unlock_irq(q->queue_lock);
3923}
3924EXPORT_SYMBOL(blk_set_runtime_active);
6c954667
LM
3925#endif
3926
1da177e4
LT
3927int __init blk_dev_init(void)
3928{
ef295ecf
CH
3929 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3930 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
0762b23d 3931 FIELD_SIZEOF(struct request, cmd_flags));
ef295ecf
CH
3932 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3933 FIELD_SIZEOF(struct bio, bi_opf));
9eb55b03 3934
89b90be2
TH
3935 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3936 kblockd_workqueue = alloc_workqueue("kblockd",
28747fcd 3937 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1da177e4
LT
3938 if (!kblockd_workqueue)
3939 panic("Failed to create kblockd\n");
3940
3941 request_cachep = kmem_cache_create("blkdev_requests",
20c2df83 3942 sizeof(struct request), 0, SLAB_PANIC, NULL);
1da177e4 3943
c2789bd4 3944 blk_requestq_cachep = kmem_cache_create("request_queue",
165125e1 3945 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1da177e4 3946
18fbda91
OS
3947#ifdef CONFIG_DEBUG_FS
3948 blk_debugfs_root = debugfs_create_dir("block", NULL);
3949#endif
3950
d38ecf93 3951 return 0;
1da177e4 3952}