]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - block/blk-core.c
block: Make q_usage_counter also track legacy requests
[mirror_ubuntu-jammy-kernel.git] / block / blk-core.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6728cb0e
JA
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
1da177e4
LT
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
1da177e4
LT
14#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
320ae51f 19#include <linux/blk-mq.h>
1da177e4
LT
20#include <linux/highmem.h>
21#include <linux/mm.h>
22#include <linux/kernel_stat.h>
23#include <linux/string.h>
24#include <linux/init.h>
1da177e4
LT
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
faccbd4b 29#include <linux/task_io_accounting_ops.h>
c17bb495 30#include <linux/fault-inject.h>
73c10101 31#include <linux/list_sort.h>
e3c78ca5 32#include <linux/delay.h>
aaf7c680 33#include <linux/ratelimit.h>
6c954667 34#include <linux/pm_runtime.h>
eea8f41c 35#include <linux/blk-cgroup.h>
18fbda91 36#include <linux/debugfs.h>
55782138
LZ
37
38#define CREATE_TRACE_POINTS
39#include <trace/events/block.h>
1da177e4 40
8324aa91 41#include "blk.h"
43a5e4e2 42#include "blk-mq.h"
bd166ef1 43#include "blk-mq-sched.h"
87760e5e 44#include "blk-wbt.h"
8324aa91 45
18fbda91
OS
46#ifdef CONFIG_DEBUG_FS
47struct dentry *blk_debugfs_root;
48#endif
49
d07335e5 50EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
b0da3f0d 51EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
0a82a8d1 52EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
3291fa57 53EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
cbae8d45 54EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
0bfc2455 55
a73f730d
TH
56DEFINE_IDA(blk_queue_ida);
57
1da177e4
LT
58/*
59 * For the allocated request tables
60 */
d674d414 61struct kmem_cache *request_cachep;
1da177e4
LT
62
63/*
64 * For queue allocation
65 */
6728cb0e 66struct kmem_cache *blk_requestq_cachep;
1da177e4 67
1da177e4
LT
68/*
69 * Controlling structure to kblockd
70 */
ff856bad 71static struct workqueue_struct *kblockd_workqueue;
1da177e4 72
d40f75a0
TH
73static void blk_clear_congested(struct request_list *rl, int sync)
74{
d40f75a0
TH
75#ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl->blkg->wb_congested, sync);
77#else
482cf79c
TH
78 /*
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
81 */
82 if (rl == &rl->q->root_rl)
dc3b17cc 83 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
d40f75a0
TH
84#endif
85}
86
87static void blk_set_congested(struct request_list *rl, int sync)
88{
d40f75a0
TH
89#ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl->blkg->wb_congested, sync);
91#else
482cf79c
TH
92 /* see blk_clear_congested() */
93 if (rl == &rl->q->root_rl)
dc3b17cc 94 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
d40f75a0
TH
95#endif
96}
97
8324aa91 98void blk_queue_congestion_threshold(struct request_queue *q)
1da177e4
LT
99{
100 int nr;
101
102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
103 if (nr > q->nr_requests)
104 nr = q->nr_requests;
105 q->nr_congestion_on = nr;
106
107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
108 if (nr < 1)
109 nr = 1;
110 q->nr_congestion_off = nr;
111}
112
2a4aa30c 113void blk_rq_init(struct request_queue *q, struct request *rq)
1da177e4 114{
1afb20f3
FT
115 memset(rq, 0, sizeof(*rq));
116
1da177e4 117 INIT_LIST_HEAD(&rq->queuelist);
242f9dcb 118 INIT_LIST_HEAD(&rq->timeout_list);
c7c22e4d 119 rq->cpu = -1;
63a71386 120 rq->q = q;
a2dec7b3 121 rq->__sector = (sector_t) -1;
2e662b65
JA
122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
63a71386 124 rq->tag = -1;
bd166ef1 125 rq->internal_tag = -1;
b243ddcb 126 rq->start_time = jiffies;
9195291e 127 set_start_time_ns(rq);
09e099d4 128 rq->part = NULL;
1da177e4 129}
2a4aa30c 130EXPORT_SYMBOL(blk_rq_init);
1da177e4 131
2a842aca
CH
132static const struct {
133 int errno;
134 const char *name;
135} blk_errors[] = {
136 [BLK_STS_OK] = { 0, "" },
137 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
138 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
139 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
140 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
141 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
142 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
143 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
144 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
145 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
03a07c92 146 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
2a842aca 147
4e4cbee9
CH
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
150
2a842aca
CH
151 /* everything else not covered above: */
152 [BLK_STS_IOERR] = { -EIO, "I/O" },
153};
154
155blk_status_t errno_to_blk_status(int errno)
156{
157 int i;
158
159 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
160 if (blk_errors[i].errno == errno)
161 return (__force blk_status_t)i;
162 }
163
164 return BLK_STS_IOERR;
165}
166EXPORT_SYMBOL_GPL(errno_to_blk_status);
167
168int blk_status_to_errno(blk_status_t status)
169{
170 int idx = (__force int)status;
171
34bd9c1c 172 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
2a842aca
CH
173 return -EIO;
174 return blk_errors[idx].errno;
175}
176EXPORT_SYMBOL_GPL(blk_status_to_errno);
177
178static void print_req_error(struct request *req, blk_status_t status)
179{
180 int idx = (__force int)status;
181
34bd9c1c 182 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
2a842aca
CH
183 return;
184
185 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
186 __func__, blk_errors[idx].name, req->rq_disk ?
187 req->rq_disk->disk_name : "?",
188 (unsigned long long)blk_rq_pos(req));
189}
190
5bb23a68 191static void req_bio_endio(struct request *rq, struct bio *bio,
2a842aca 192 unsigned int nbytes, blk_status_t error)
1da177e4 193{
78d8e58a 194 if (error)
4e4cbee9 195 bio->bi_status = error;
797e7dbb 196
e8064021 197 if (unlikely(rq->rq_flags & RQF_QUIET))
b7c44ed9 198 bio_set_flag(bio, BIO_QUIET);
08bafc03 199
f79ea416 200 bio_advance(bio, nbytes);
7ba1ba12 201
143a87f4 202 /* don't actually finish bio if it's part of flush sequence */
e8064021 203 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
4246a0b6 204 bio_endio(bio);
1da177e4 205}
1da177e4 206
1da177e4
LT
207void blk_dump_rq_flags(struct request *rq, char *msg)
208{
aebf526b
CH
209 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
210 rq->rq_disk ? rq->rq_disk->disk_name : "?",
5953316d 211 (unsigned long long) rq->cmd_flags);
1da177e4 212
83096ebf
TH
213 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
214 (unsigned long long)blk_rq_pos(rq),
215 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
b4f42e28
JA
216 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
217 rq->bio, rq->biotail, blk_rq_bytes(rq));
1da177e4 218}
1da177e4
LT
219EXPORT_SYMBOL(blk_dump_rq_flags);
220
3cca6dc1 221static void blk_delay_work(struct work_struct *work)
1da177e4 222{
3cca6dc1 223 struct request_queue *q;
1da177e4 224
3cca6dc1
JA
225 q = container_of(work, struct request_queue, delay_work.work);
226 spin_lock_irq(q->queue_lock);
24ecfbe2 227 __blk_run_queue(q);
3cca6dc1 228 spin_unlock_irq(q->queue_lock);
1da177e4 229}
1da177e4
LT
230
231/**
3cca6dc1
JA
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
1da177e4
LT
235 *
236 * Description:
3cca6dc1
JA
237 * Sometimes queueing needs to be postponed for a little while, to allow
238 * resources to come back. This function will make sure that queueing is
2fff8a92 239 * restarted around the specified time.
3cca6dc1
JA
240 */
241void blk_delay_queue(struct request_queue *q, unsigned long msecs)
2ad8b1ef 242{
2fff8a92 243 lockdep_assert_held(q->queue_lock);
332ebbf7 244 WARN_ON_ONCE(q->mq_ops);
2fff8a92 245
70460571
BVA
246 if (likely(!blk_queue_dead(q)))
247 queue_delayed_work(kblockd_workqueue, &q->delay_work,
248 msecs_to_jiffies(msecs));
2ad8b1ef 249}
3cca6dc1 250EXPORT_SYMBOL(blk_delay_queue);
2ad8b1ef 251
21491412
JA
252/**
253 * blk_start_queue_async - asynchronously restart a previously stopped queue
254 * @q: The &struct request_queue in question
255 *
256 * Description:
257 * blk_start_queue_async() will clear the stop flag on the queue, and
258 * ensure that the request_fn for the queue is run from an async
259 * context.
260 **/
261void blk_start_queue_async(struct request_queue *q)
262{
2fff8a92 263 lockdep_assert_held(q->queue_lock);
332ebbf7 264 WARN_ON_ONCE(q->mq_ops);
2fff8a92 265
21491412
JA
266 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
267 blk_run_queue_async(q);
268}
269EXPORT_SYMBOL(blk_start_queue_async);
270
1da177e4
LT
271/**
272 * blk_start_queue - restart a previously stopped queue
165125e1 273 * @q: The &struct request_queue in question
1da177e4
LT
274 *
275 * Description:
276 * blk_start_queue() will clear the stop flag on the queue, and call
277 * the request_fn for the queue if it was in a stopped state when
2fff8a92 278 * entered. Also see blk_stop_queue().
1da177e4 279 **/
165125e1 280void blk_start_queue(struct request_queue *q)
1da177e4 281{
2fff8a92 282 lockdep_assert_held(q->queue_lock);
4ddd56b0 283 WARN_ON(!in_interrupt() && !irqs_disabled());
332ebbf7 284 WARN_ON_ONCE(q->mq_ops);
a038e253 285
75ad23bc 286 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
24ecfbe2 287 __blk_run_queue(q);
1da177e4 288}
1da177e4
LT
289EXPORT_SYMBOL(blk_start_queue);
290
291/**
292 * blk_stop_queue - stop a queue
165125e1 293 * @q: The &struct request_queue in question
1da177e4
LT
294 *
295 * Description:
296 * The Linux block layer assumes that a block driver will consume all
297 * entries on the request queue when the request_fn strategy is called.
298 * Often this will not happen, because of hardware limitations (queue
299 * depth settings). If a device driver gets a 'queue full' response,
300 * or if it simply chooses not to queue more I/O at one point, it can
301 * call this function to prevent the request_fn from being called until
302 * the driver has signalled it's ready to go again. This happens by calling
2fff8a92 303 * blk_start_queue() to restart queue operations.
1da177e4 304 **/
165125e1 305void blk_stop_queue(struct request_queue *q)
1da177e4 306{
2fff8a92 307 lockdep_assert_held(q->queue_lock);
332ebbf7 308 WARN_ON_ONCE(q->mq_ops);
2fff8a92 309
136b5721 310 cancel_delayed_work(&q->delay_work);
75ad23bc 311 queue_flag_set(QUEUE_FLAG_STOPPED, q);
1da177e4
LT
312}
313EXPORT_SYMBOL(blk_stop_queue);
314
315/**
316 * blk_sync_queue - cancel any pending callbacks on a queue
317 * @q: the queue
318 *
319 * Description:
320 * The block layer may perform asynchronous callback activity
321 * on a queue, such as calling the unplug function after a timeout.
322 * A block device may call blk_sync_queue to ensure that any
323 * such activity is cancelled, thus allowing it to release resources
59c51591 324 * that the callbacks might use. The caller must already have made sure
1da177e4
LT
325 * that its ->make_request_fn will not re-add plugging prior to calling
326 * this function.
327 *
da527770 328 * This function does not cancel any asynchronous activity arising
da3dae54 329 * out of elevator or throttling code. That would require elevator_exit()
5efd6113 330 * and blkcg_exit_queue() to be called with queue lock initialized.
da527770 331 *
1da177e4
LT
332 */
333void blk_sync_queue(struct request_queue *q)
334{
70ed28b9 335 del_timer_sync(&q->timeout);
4e9b6f20 336 cancel_work_sync(&q->timeout_work);
f04c1fe7
ML
337
338 if (q->mq_ops) {
339 struct blk_mq_hw_ctx *hctx;
340 int i;
341
aba7afc5 342 cancel_delayed_work_sync(&q->requeue_work);
21c6e939 343 queue_for_each_hw_ctx(q, hctx, i)
9f993737 344 cancel_delayed_work_sync(&hctx->run_work);
f04c1fe7
ML
345 } else {
346 cancel_delayed_work_sync(&q->delay_work);
347 }
1da177e4
LT
348}
349EXPORT_SYMBOL(blk_sync_queue);
350
c246e80d
BVA
351/**
352 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
353 * @q: The queue to run
354 *
355 * Description:
356 * Invoke request handling on a queue if there are any pending requests.
357 * May be used to restart request handling after a request has completed.
358 * This variant runs the queue whether or not the queue has been
359 * stopped. Must be called with the queue lock held and interrupts
360 * disabled. See also @blk_run_queue.
361 */
362inline void __blk_run_queue_uncond(struct request_queue *q)
363{
2fff8a92 364 lockdep_assert_held(q->queue_lock);
332ebbf7 365 WARN_ON_ONCE(q->mq_ops);
2fff8a92 366
c246e80d
BVA
367 if (unlikely(blk_queue_dead(q)))
368 return;
369
24faf6f6
BVA
370 /*
371 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
372 * the queue lock internally. As a result multiple threads may be
373 * running such a request function concurrently. Keep track of the
374 * number of active request_fn invocations such that blk_drain_queue()
375 * can wait until all these request_fn calls have finished.
376 */
377 q->request_fn_active++;
c246e80d 378 q->request_fn(q);
24faf6f6 379 q->request_fn_active--;
c246e80d 380}
a7928c15 381EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
c246e80d 382
1da177e4 383/**
80a4b58e 384 * __blk_run_queue - run a single device queue
1da177e4 385 * @q: The queue to run
80a4b58e
JA
386 *
387 * Description:
2fff8a92 388 * See @blk_run_queue.
1da177e4 389 */
24ecfbe2 390void __blk_run_queue(struct request_queue *q)
1da177e4 391{
2fff8a92 392 lockdep_assert_held(q->queue_lock);
332ebbf7 393 WARN_ON_ONCE(q->mq_ops);
2fff8a92 394
a538cd03
TH
395 if (unlikely(blk_queue_stopped(q)))
396 return;
397
c246e80d 398 __blk_run_queue_uncond(q);
75ad23bc
NP
399}
400EXPORT_SYMBOL(__blk_run_queue);
dac07ec1 401
24ecfbe2
CH
402/**
403 * blk_run_queue_async - run a single device queue in workqueue context
404 * @q: The queue to run
405 *
406 * Description:
407 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
2fff8a92
BVA
408 * of us.
409 *
410 * Note:
411 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
412 * has canceled q->delay_work, callers must hold the queue lock to avoid
413 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
24ecfbe2
CH
414 */
415void blk_run_queue_async(struct request_queue *q)
416{
2fff8a92 417 lockdep_assert_held(q->queue_lock);
332ebbf7 418 WARN_ON_ONCE(q->mq_ops);
2fff8a92 419
70460571 420 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
e7c2f967 421 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
24ecfbe2 422}
c21e6beb 423EXPORT_SYMBOL(blk_run_queue_async);
24ecfbe2 424
75ad23bc
NP
425/**
426 * blk_run_queue - run a single device queue
427 * @q: The queue to run
80a4b58e
JA
428 *
429 * Description:
430 * Invoke request handling on this queue, if it has pending work to do.
a7f55792 431 * May be used to restart queueing when a request has completed.
75ad23bc
NP
432 */
433void blk_run_queue(struct request_queue *q)
434{
435 unsigned long flags;
436
332ebbf7
BVA
437 WARN_ON_ONCE(q->mq_ops);
438
75ad23bc 439 spin_lock_irqsave(q->queue_lock, flags);
24ecfbe2 440 __blk_run_queue(q);
1da177e4
LT
441 spin_unlock_irqrestore(q->queue_lock, flags);
442}
443EXPORT_SYMBOL(blk_run_queue);
444
165125e1 445void blk_put_queue(struct request_queue *q)
483f4afc
AV
446{
447 kobject_put(&q->kobj);
448}
d86e0e83 449EXPORT_SYMBOL(blk_put_queue);
483f4afc 450
e3c78ca5 451/**
807592a4 452 * __blk_drain_queue - drain requests from request_queue
e3c78ca5 453 * @q: queue to drain
c9a929dd 454 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
e3c78ca5 455 *
c9a929dd
TH
456 * Drain requests from @q. If @drain_all is set, all requests are drained.
457 * If not, only ELVPRIV requests are drained. The caller is responsible
458 * for ensuring that no new requests which need to be drained are queued.
e3c78ca5 459 */
807592a4
BVA
460static void __blk_drain_queue(struct request_queue *q, bool drain_all)
461 __releases(q->queue_lock)
462 __acquires(q->queue_lock)
e3c78ca5 463{
458f27a9
AH
464 int i;
465
807592a4 466 lockdep_assert_held(q->queue_lock);
332ebbf7 467 WARN_ON_ONCE(q->mq_ops);
807592a4 468
e3c78ca5 469 while (true) {
481a7d64 470 bool drain = false;
e3c78ca5 471
b855b04a
TH
472 /*
473 * The caller might be trying to drain @q before its
474 * elevator is initialized.
475 */
476 if (q->elevator)
477 elv_drain_elevator(q);
478
5efd6113 479 blkcg_drain_queue(q);
e3c78ca5 480
4eabc941
TH
481 /*
482 * This function might be called on a queue which failed
b855b04a
TH
483 * driver init after queue creation or is not yet fully
484 * active yet. Some drivers (e.g. fd and loop) get unhappy
485 * in such cases. Kick queue iff dispatch queue has
486 * something on it and @q has request_fn set.
4eabc941 487 */
b855b04a 488 if (!list_empty(&q->queue_head) && q->request_fn)
4eabc941 489 __blk_run_queue(q);
c9a929dd 490
8a5ecdd4 491 drain |= q->nr_rqs_elvpriv;
24faf6f6 492 drain |= q->request_fn_active;
481a7d64
TH
493
494 /*
495 * Unfortunately, requests are queued at and tracked from
496 * multiple places and there's no single counter which can
497 * be drained. Check all the queues and counters.
498 */
499 if (drain_all) {
e97c293c 500 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
481a7d64
TH
501 drain |= !list_empty(&q->queue_head);
502 for (i = 0; i < 2; i++) {
8a5ecdd4 503 drain |= q->nr_rqs[i];
481a7d64 504 drain |= q->in_flight[i];
7c94e1c1
ML
505 if (fq)
506 drain |= !list_empty(&fq->flush_queue[i]);
481a7d64
TH
507 }
508 }
e3c78ca5 509
481a7d64 510 if (!drain)
e3c78ca5 511 break;
807592a4
BVA
512
513 spin_unlock_irq(q->queue_lock);
514
e3c78ca5 515 msleep(10);
807592a4
BVA
516
517 spin_lock_irq(q->queue_lock);
e3c78ca5 518 }
458f27a9
AH
519
520 /*
521 * With queue marked dead, any woken up waiter will fail the
522 * allocation path, so the wakeup chaining is lost and we're
523 * left with hung waiters. We need to wake up those waiters.
524 */
525 if (q->request_fn) {
a051661c
TH
526 struct request_list *rl;
527
a051661c
TH
528 blk_queue_for_each_rl(rl, q)
529 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
530 wake_up_all(&rl->wait[i]);
458f27a9 531 }
e3c78ca5
TH
532}
533
d732580b
TH
534/**
535 * blk_queue_bypass_start - enter queue bypass mode
536 * @q: queue of interest
537 *
538 * In bypass mode, only the dispatch FIFO queue of @q is used. This
539 * function makes @q enter bypass mode and drains all requests which were
6ecf23af 540 * throttled or issued before. On return, it's guaranteed that no request
80fd9979
TH
541 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
542 * inside queue or RCU read lock.
d732580b
TH
543 */
544void blk_queue_bypass_start(struct request_queue *q)
545{
332ebbf7
BVA
546 WARN_ON_ONCE(q->mq_ops);
547
d732580b 548 spin_lock_irq(q->queue_lock);
776687bc 549 q->bypass_depth++;
d732580b
TH
550 queue_flag_set(QUEUE_FLAG_BYPASS, q);
551 spin_unlock_irq(q->queue_lock);
552
776687bc
TH
553 /*
554 * Queues start drained. Skip actual draining till init is
555 * complete. This avoids lenghty delays during queue init which
556 * can happen many times during boot.
557 */
558 if (blk_queue_init_done(q)) {
807592a4
BVA
559 spin_lock_irq(q->queue_lock);
560 __blk_drain_queue(q, false);
561 spin_unlock_irq(q->queue_lock);
562
b82d4b19
TH
563 /* ensure blk_queue_bypass() is %true inside RCU read lock */
564 synchronize_rcu();
565 }
d732580b
TH
566}
567EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
568
569/**
570 * blk_queue_bypass_end - leave queue bypass mode
571 * @q: queue of interest
572 *
573 * Leave bypass mode and restore the normal queueing behavior.
332ebbf7
BVA
574 *
575 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
576 * this function is called for both blk-sq and blk-mq queues.
d732580b
TH
577 */
578void blk_queue_bypass_end(struct request_queue *q)
579{
580 spin_lock_irq(q->queue_lock);
581 if (!--q->bypass_depth)
582 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
583 WARN_ON_ONCE(q->bypass_depth < 0);
584 spin_unlock_irq(q->queue_lock);
585}
586EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
587
aed3ea94
JA
588void blk_set_queue_dying(struct request_queue *q)
589{
1b856086
BVA
590 spin_lock_irq(q->queue_lock);
591 queue_flag_set(QUEUE_FLAG_DYING, q);
592 spin_unlock_irq(q->queue_lock);
aed3ea94 593
d3cfb2a0
ML
594 /*
595 * When queue DYING flag is set, we need to block new req
596 * entering queue, so we call blk_freeze_queue_start() to
597 * prevent I/O from crossing blk_queue_enter().
598 */
599 blk_freeze_queue_start(q);
600
aed3ea94
JA
601 if (q->mq_ops)
602 blk_mq_wake_waiters(q);
603 else {
604 struct request_list *rl;
605
bbfc3c5d 606 spin_lock_irq(q->queue_lock);
aed3ea94
JA
607 blk_queue_for_each_rl(rl, q) {
608 if (rl->rq_pool) {
609 wake_up(&rl->wait[BLK_RW_SYNC]);
610 wake_up(&rl->wait[BLK_RW_ASYNC]);
611 }
612 }
bbfc3c5d 613 spin_unlock_irq(q->queue_lock);
aed3ea94 614 }
055f6e18
ML
615
616 /* Make blk_queue_enter() reexamine the DYING flag. */
617 wake_up_all(&q->mq_freeze_wq);
aed3ea94
JA
618}
619EXPORT_SYMBOL_GPL(blk_set_queue_dying);
620
c9a929dd
TH
621/**
622 * blk_cleanup_queue - shutdown a request queue
623 * @q: request queue to shutdown
624 *
c246e80d
BVA
625 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
626 * put it. All future requests will be failed immediately with -ENODEV.
c94a96ac 627 */
6728cb0e 628void blk_cleanup_queue(struct request_queue *q)
483f4afc 629{
c9a929dd 630 spinlock_t *lock = q->queue_lock;
e3335de9 631
3f3299d5 632 /* mark @q DYING, no new request or merges will be allowed afterwards */
483f4afc 633 mutex_lock(&q->sysfs_lock);
aed3ea94 634 blk_set_queue_dying(q);
c9a929dd 635 spin_lock_irq(lock);
6ecf23af 636
80fd9979 637 /*
3f3299d5 638 * A dying queue is permanently in bypass mode till released. Note
80fd9979
TH
639 * that, unlike blk_queue_bypass_start(), we aren't performing
640 * synchronize_rcu() after entering bypass mode to avoid the delay
641 * as some drivers create and destroy a lot of queues while
642 * probing. This is still safe because blk_release_queue() will be
643 * called only after the queue refcnt drops to zero and nothing,
644 * RCU or not, would be traversing the queue by then.
645 */
6ecf23af
TH
646 q->bypass_depth++;
647 queue_flag_set(QUEUE_FLAG_BYPASS, q);
648
c9a929dd
TH
649 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
650 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
3f3299d5 651 queue_flag_set(QUEUE_FLAG_DYING, q);
c9a929dd
TH
652 spin_unlock_irq(lock);
653 mutex_unlock(&q->sysfs_lock);
654
c246e80d
BVA
655 /*
656 * Drain all requests queued before DYING marking. Set DEAD flag to
657 * prevent that q->request_fn() gets invoked after draining finished.
658 */
3ef28e83 659 blk_freeze_queue(q);
9c1051aa
OS
660 spin_lock_irq(lock);
661 if (!q->mq_ops)
43a5e4e2 662 __blk_drain_queue(q, true);
c246e80d 663 queue_flag_set(QUEUE_FLAG_DEAD, q);
807592a4 664 spin_unlock_irq(lock);
c9a929dd 665
5a48fc14
DW
666 /* for synchronous bio-based driver finish in-flight integrity i/o */
667 blk_flush_integrity();
668
c9a929dd 669 /* @q won't process any more request, flush async actions */
dc3b17cc 670 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
c9a929dd
TH
671 blk_sync_queue(q);
672
45a9c9d9
BVA
673 if (q->mq_ops)
674 blk_mq_free_queue(q);
3ef28e83 675 percpu_ref_exit(&q->q_usage_counter);
45a9c9d9 676
5e5cfac0
AH
677 spin_lock_irq(lock);
678 if (q->queue_lock != &q->__queue_lock)
679 q->queue_lock = &q->__queue_lock;
680 spin_unlock_irq(lock);
681
c9a929dd 682 /* @q is and will stay empty, shutdown and put */
483f4afc
AV
683 blk_put_queue(q);
684}
1da177e4
LT
685EXPORT_SYMBOL(blk_cleanup_queue);
686
271508db 687/* Allocate memory local to the request queue */
6d247d7f 688static void *alloc_request_simple(gfp_t gfp_mask, void *data)
271508db 689{
6d247d7f
CH
690 struct request_queue *q = data;
691
692 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
271508db
DR
693}
694
6d247d7f 695static void free_request_simple(void *element, void *data)
271508db
DR
696{
697 kmem_cache_free(request_cachep, element);
698}
699
6d247d7f
CH
700static void *alloc_request_size(gfp_t gfp_mask, void *data)
701{
702 struct request_queue *q = data;
703 struct request *rq;
704
705 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
706 q->node);
707 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
708 kfree(rq);
709 rq = NULL;
710 }
711 return rq;
712}
713
714static void free_request_size(void *element, void *data)
715{
716 struct request_queue *q = data;
717
718 if (q->exit_rq_fn)
719 q->exit_rq_fn(q, element);
720 kfree(element);
721}
722
5b788ce3
TH
723int blk_init_rl(struct request_list *rl, struct request_queue *q,
724 gfp_t gfp_mask)
1da177e4 725{
85acb3ba 726 if (unlikely(rl->rq_pool) || q->mq_ops)
1abec4fd
MS
727 return 0;
728
5b788ce3 729 rl->q = q;
1faa16d2
JA
730 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
731 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
1faa16d2
JA
732 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
733 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
1da177e4 734
6d247d7f
CH
735 if (q->cmd_size) {
736 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
737 alloc_request_size, free_request_size,
738 q, gfp_mask, q->node);
739 } else {
740 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
741 alloc_request_simple, free_request_simple,
742 q, gfp_mask, q->node);
743 }
1da177e4
LT
744 if (!rl->rq_pool)
745 return -ENOMEM;
746
b425e504
BVA
747 if (rl != &q->root_rl)
748 WARN_ON_ONCE(!blk_get_queue(q));
749
1da177e4
LT
750 return 0;
751}
752
b425e504 753void blk_exit_rl(struct request_queue *q, struct request_list *rl)
5b788ce3 754{
b425e504 755 if (rl->rq_pool) {
5b788ce3 756 mempool_destroy(rl->rq_pool);
b425e504
BVA
757 if (rl != &q->root_rl)
758 blk_put_queue(q);
759 }
5b788ce3
TH
760}
761
165125e1 762struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
1da177e4 763{
c304a51b 764 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
1946089a
CL
765}
766EXPORT_SYMBOL(blk_alloc_queue);
1da177e4 767
6f3b0e8b 768int blk_queue_enter(struct request_queue *q, bool nowait)
3ef28e83
DW
769{
770 while (true) {
771 int ret;
772
773 if (percpu_ref_tryget_live(&q->q_usage_counter))
774 return 0;
775
6f3b0e8b 776 if (nowait)
3ef28e83
DW
777 return -EBUSY;
778
5ed61d3f 779 /*
1671d522 780 * read pair of barrier in blk_freeze_queue_start(),
5ed61d3f 781 * we need to order reading __PERCPU_REF_DEAD flag of
d3cfb2a0
ML
782 * .q_usage_counter and reading .mq_freeze_depth or
783 * queue dying flag, otherwise the following wait may
784 * never return if the two reads are reordered.
5ed61d3f
ML
785 */
786 smp_rmb();
787
3ef28e83
DW
788 ret = wait_event_interruptible(q->mq_freeze_wq,
789 !atomic_read(&q->mq_freeze_depth) ||
790 blk_queue_dying(q));
791 if (blk_queue_dying(q))
792 return -ENODEV;
793 if (ret)
794 return ret;
795 }
796}
797
798void blk_queue_exit(struct request_queue *q)
799{
800 percpu_ref_put(&q->q_usage_counter);
801}
802
803static void blk_queue_usage_counter_release(struct percpu_ref *ref)
804{
805 struct request_queue *q =
806 container_of(ref, struct request_queue, q_usage_counter);
807
808 wake_up_all(&q->mq_freeze_wq);
809}
810
287922eb
CH
811static void blk_rq_timed_out_timer(unsigned long data)
812{
813 struct request_queue *q = (struct request_queue *)data;
814
815 kblockd_schedule_work(&q->timeout_work);
816}
817
165125e1 818struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
1946089a 819{
165125e1 820 struct request_queue *q;
1946089a 821
8324aa91 822 q = kmem_cache_alloc_node(blk_requestq_cachep,
94f6030c 823 gfp_mask | __GFP_ZERO, node_id);
1da177e4
LT
824 if (!q)
825 return NULL;
826
00380a40 827 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
a73f730d 828 if (q->id < 0)
3d2936f4 829 goto fail_q;
a73f730d 830
93b27e72 831 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
54efd50b
KO
832 if (!q->bio_split)
833 goto fail_id;
834
d03f6cdc
JK
835 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
836 if (!q->backing_dev_info)
837 goto fail_split;
838
a83b576c
JA
839 q->stats = blk_alloc_queue_stats();
840 if (!q->stats)
841 goto fail_stats;
842
dc3b17cc 843 q->backing_dev_info->ra_pages =
09cbfeaf 844 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
dc3b17cc
JK
845 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
846 q->backing_dev_info->name = "block";
5151412d 847 q->node = node_id;
0989a025 848
dc3b17cc 849 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
31373d09 850 laptop_mode_timer_fn, (unsigned long) q);
242f9dcb 851 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
4e9b6f20 852 INIT_WORK(&q->timeout_work, NULL);
b855b04a 853 INIT_LIST_HEAD(&q->queue_head);
242f9dcb 854 INIT_LIST_HEAD(&q->timeout_list);
a612fddf 855 INIT_LIST_HEAD(&q->icq_list);
4eef3049 856#ifdef CONFIG_BLK_CGROUP
e8989fae 857 INIT_LIST_HEAD(&q->blkg_list);
4eef3049 858#endif
3cca6dc1 859 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
483f4afc 860
8324aa91 861 kobject_init(&q->kobj, &blk_queue_ktype);
1da177e4 862
5acb3cc2
WL
863#ifdef CONFIG_BLK_DEV_IO_TRACE
864 mutex_init(&q->blk_trace_mutex);
865#endif
483f4afc 866 mutex_init(&q->sysfs_lock);
e7e72bf6 867 spin_lock_init(&q->__queue_lock);
483f4afc 868
c94a96ac
VG
869 /*
870 * By default initialize queue_lock to internal lock and driver can
871 * override it later if need be.
872 */
873 q->queue_lock = &q->__queue_lock;
874
b82d4b19
TH
875 /*
876 * A queue starts its life with bypass turned on to avoid
877 * unnecessary bypass on/off overhead and nasty surprises during
749fefe6
TH
878 * init. The initial bypass will be finished when the queue is
879 * registered by blk_register_queue().
b82d4b19
TH
880 */
881 q->bypass_depth = 1;
882 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
883
320ae51f
JA
884 init_waitqueue_head(&q->mq_freeze_wq);
885
3ef28e83
DW
886 /*
887 * Init percpu_ref in atomic mode so that it's faster to shutdown.
888 * See blk_register_queue() for details.
889 */
890 if (percpu_ref_init(&q->q_usage_counter,
891 blk_queue_usage_counter_release,
892 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
fff4996b 893 goto fail_bdi;
f51b802c 894
3ef28e83
DW
895 if (blkcg_init_queue(q))
896 goto fail_ref;
897
1da177e4 898 return q;
a73f730d 899
3ef28e83
DW
900fail_ref:
901 percpu_ref_exit(&q->q_usage_counter);
fff4996b 902fail_bdi:
a83b576c
JA
903 blk_free_queue_stats(q->stats);
904fail_stats:
d03f6cdc 905 bdi_put(q->backing_dev_info);
54efd50b
KO
906fail_split:
907 bioset_free(q->bio_split);
a73f730d
TH
908fail_id:
909 ida_simple_remove(&blk_queue_ida, q->id);
910fail_q:
911 kmem_cache_free(blk_requestq_cachep, q);
912 return NULL;
1da177e4 913}
1946089a 914EXPORT_SYMBOL(blk_alloc_queue_node);
1da177e4
LT
915
916/**
917 * blk_init_queue - prepare a request queue for use with a block device
918 * @rfn: The function to be called to process requests that have been
919 * placed on the queue.
920 * @lock: Request queue spin lock
921 *
922 * Description:
923 * If a block device wishes to use the standard request handling procedures,
924 * which sorts requests and coalesces adjacent requests, then it must
925 * call blk_init_queue(). The function @rfn will be called when there
926 * are requests on the queue that need to be processed. If the device
927 * supports plugging, then @rfn may not be called immediately when requests
928 * are available on the queue, but may be called at some time later instead.
929 * Plugged queues are generally unplugged when a buffer belonging to one
930 * of the requests on the queue is needed, or due to memory pressure.
931 *
932 * @rfn is not required, or even expected, to remove all requests off the
933 * queue, but only as many as it can handle at a time. If it does leave
934 * requests on the queue, it is responsible for arranging that the requests
935 * get dealt with eventually.
936 *
937 * The queue spin lock must be held while manipulating the requests on the
a038e253
PBG
938 * request queue; this lock will be taken also from interrupt context, so irq
939 * disabling is needed for it.
1da177e4 940 *
710027a4 941 * Function returns a pointer to the initialized request queue, or %NULL if
1da177e4
LT
942 * it didn't succeed.
943 *
944 * Note:
945 * blk_init_queue() must be paired with a blk_cleanup_queue() call
946 * when the block device is deactivated (such as at module unload).
947 **/
1946089a 948
165125e1 949struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1da177e4 950{
c304a51b 951 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1946089a
CL
952}
953EXPORT_SYMBOL(blk_init_queue);
954
165125e1 955struct request_queue *
1946089a
CL
956blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
957{
5ea708d1 958 struct request_queue *q;
1da177e4 959
5ea708d1
CH
960 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
961 if (!q)
c86d1b8a
MS
962 return NULL;
963
5ea708d1
CH
964 q->request_fn = rfn;
965 if (lock)
966 q->queue_lock = lock;
967 if (blk_init_allocated_queue(q) < 0) {
968 blk_cleanup_queue(q);
969 return NULL;
970 }
18741986 971
7982e90c 972 return q;
01effb0d
MS
973}
974EXPORT_SYMBOL(blk_init_queue_node);
975
dece1635 976static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
336b7e1f 977
1da177e4 978
5ea708d1
CH
979int blk_init_allocated_queue(struct request_queue *q)
980{
332ebbf7
BVA
981 WARN_ON_ONCE(q->mq_ops);
982
6d247d7f 983 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
ba483388 984 if (!q->fq)
5ea708d1 985 return -ENOMEM;
7982e90c 986
6d247d7f
CH
987 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
988 goto out_free_flush_queue;
7982e90c 989
a051661c 990 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
6d247d7f 991 goto out_exit_flush_rq;
1da177e4 992
287922eb 993 INIT_WORK(&q->timeout_work, blk_timeout_work);
60ea8226 994 q->queue_flags |= QUEUE_FLAG_DEFAULT;
c94a96ac 995
f3b144aa
JA
996 /*
997 * This also sets hw/phys segments, boundary and size
998 */
c20e8de2 999 blk_queue_make_request(q, blk_queue_bio);
1da177e4 1000
44ec9542
AS
1001 q->sg_reserved_size = INT_MAX;
1002
eb1c160b
TS
1003 /* Protect q->elevator from elevator_change */
1004 mutex_lock(&q->sysfs_lock);
1005
b82d4b19 1006 /* init elevator */
eb1c160b
TS
1007 if (elevator_init(q, NULL)) {
1008 mutex_unlock(&q->sysfs_lock);
6d247d7f 1009 goto out_exit_flush_rq;
eb1c160b
TS
1010 }
1011
1012 mutex_unlock(&q->sysfs_lock);
5ea708d1 1013 return 0;
eb1c160b 1014
6d247d7f
CH
1015out_exit_flush_rq:
1016 if (q->exit_rq_fn)
1017 q->exit_rq_fn(q, q->fq->flush_rq);
1018out_free_flush_queue:
ba483388 1019 blk_free_flush_queue(q->fq);
5ea708d1 1020 return -ENOMEM;
1da177e4 1021}
5151412d 1022EXPORT_SYMBOL(blk_init_allocated_queue);
1da177e4 1023
09ac46c4 1024bool blk_get_queue(struct request_queue *q)
1da177e4 1025{
3f3299d5 1026 if (likely(!blk_queue_dying(q))) {
09ac46c4
TH
1027 __blk_get_queue(q);
1028 return true;
1da177e4
LT
1029 }
1030
09ac46c4 1031 return false;
1da177e4 1032}
d86e0e83 1033EXPORT_SYMBOL(blk_get_queue);
1da177e4 1034
5b788ce3 1035static inline void blk_free_request(struct request_list *rl, struct request *rq)
1da177e4 1036{
e8064021 1037 if (rq->rq_flags & RQF_ELVPRIV) {
5b788ce3 1038 elv_put_request(rl->q, rq);
f1f8cc94 1039 if (rq->elv.icq)
11a3122f 1040 put_io_context(rq->elv.icq->ioc);
f1f8cc94
TH
1041 }
1042
5b788ce3 1043 mempool_free(rq, rl->rq_pool);
1da177e4
LT
1044}
1045
1da177e4
LT
1046/*
1047 * ioc_batching returns true if the ioc is a valid batching request and
1048 * should be given priority access to a request.
1049 */
165125e1 1050static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1051{
1052 if (!ioc)
1053 return 0;
1054
1055 /*
1056 * Make sure the process is able to allocate at least 1 request
1057 * even if the batch times out, otherwise we could theoretically
1058 * lose wakeups.
1059 */
1060 return ioc->nr_batch_requests == q->nr_batching ||
1061 (ioc->nr_batch_requests > 0
1062 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1063}
1064
1065/*
1066 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1067 * will cause the process to be a "batcher" on all queues in the system. This
1068 * is the behaviour we want though - once it gets a wakeup it should be given
1069 * a nice run.
1070 */
165125e1 1071static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1da177e4
LT
1072{
1073 if (!ioc || ioc_batching(q, ioc))
1074 return;
1075
1076 ioc->nr_batch_requests = q->nr_batching;
1077 ioc->last_waited = jiffies;
1078}
1079
5b788ce3 1080static void __freed_request(struct request_list *rl, int sync)
1da177e4 1081{
5b788ce3 1082 struct request_queue *q = rl->q;
1da177e4 1083
d40f75a0
TH
1084 if (rl->count[sync] < queue_congestion_off_threshold(q))
1085 blk_clear_congested(rl, sync);
1da177e4 1086
1faa16d2
JA
1087 if (rl->count[sync] + 1 <= q->nr_requests) {
1088 if (waitqueue_active(&rl->wait[sync]))
1089 wake_up(&rl->wait[sync]);
1da177e4 1090
5b788ce3 1091 blk_clear_rl_full(rl, sync);
1da177e4
LT
1092 }
1093}
1094
1095/*
1096 * A request has just been released. Account for it, update the full and
1097 * congestion status, wake up any waiters. Called under q->queue_lock.
1098 */
e8064021
CH
1099static void freed_request(struct request_list *rl, bool sync,
1100 req_flags_t rq_flags)
1da177e4 1101{
5b788ce3 1102 struct request_queue *q = rl->q;
1da177e4 1103
8a5ecdd4 1104 q->nr_rqs[sync]--;
1faa16d2 1105 rl->count[sync]--;
e8064021 1106 if (rq_flags & RQF_ELVPRIV)
8a5ecdd4 1107 q->nr_rqs_elvpriv--;
1da177e4 1108
5b788ce3 1109 __freed_request(rl, sync);
1da177e4 1110
1faa16d2 1111 if (unlikely(rl->starved[sync ^ 1]))
5b788ce3 1112 __freed_request(rl, sync ^ 1);
1da177e4
LT
1113}
1114
e3a2b3f9
JA
1115int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1116{
1117 struct request_list *rl;
d40f75a0 1118 int on_thresh, off_thresh;
e3a2b3f9 1119
332ebbf7
BVA
1120 WARN_ON_ONCE(q->mq_ops);
1121
e3a2b3f9
JA
1122 spin_lock_irq(q->queue_lock);
1123 q->nr_requests = nr;
1124 blk_queue_congestion_threshold(q);
d40f75a0
TH
1125 on_thresh = queue_congestion_on_threshold(q);
1126 off_thresh = queue_congestion_off_threshold(q);
e3a2b3f9 1127
d40f75a0
TH
1128 blk_queue_for_each_rl(rl, q) {
1129 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1130 blk_set_congested(rl, BLK_RW_SYNC);
1131 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1132 blk_clear_congested(rl, BLK_RW_SYNC);
e3a2b3f9 1133
d40f75a0
TH
1134 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1135 blk_set_congested(rl, BLK_RW_ASYNC);
1136 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1137 blk_clear_congested(rl, BLK_RW_ASYNC);
e3a2b3f9 1138
e3a2b3f9
JA
1139 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1140 blk_set_rl_full(rl, BLK_RW_SYNC);
1141 } else {
1142 blk_clear_rl_full(rl, BLK_RW_SYNC);
1143 wake_up(&rl->wait[BLK_RW_SYNC]);
1144 }
1145
1146 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1147 blk_set_rl_full(rl, BLK_RW_ASYNC);
1148 } else {
1149 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1150 wake_up(&rl->wait[BLK_RW_ASYNC]);
1151 }
1152 }
1153
1154 spin_unlock_irq(q->queue_lock);
1155 return 0;
1156}
1157
da8303c6 1158/**
a06e05e6 1159 * __get_request - get a free request
5b788ce3 1160 * @rl: request list to allocate from
ef295ecf 1161 * @op: operation and flags
da8303c6
TH
1162 * @bio: bio to allocate request for (can be %NULL)
1163 * @gfp_mask: allocation mask
1164 *
1165 * Get a free request from @q. This function may fail under memory
1166 * pressure or if @q is dead.
1167 *
da3dae54 1168 * Must be called with @q->queue_lock held and,
a492f075
JL
1169 * Returns ERR_PTR on failure, with @q->queue_lock held.
1170 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1171 */
ef295ecf
CH
1172static struct request *__get_request(struct request_list *rl, unsigned int op,
1173 struct bio *bio, gfp_t gfp_mask)
1da177e4 1174{
5b788ce3 1175 struct request_queue *q = rl->q;
b679281a 1176 struct request *rq;
7f4b35d1
TH
1177 struct elevator_type *et = q->elevator->type;
1178 struct io_context *ioc = rq_ioc(bio);
f1f8cc94 1179 struct io_cq *icq = NULL;
ef295ecf 1180 const bool is_sync = op_is_sync(op);
75eb6c37 1181 int may_queue;
e8064021 1182 req_flags_t rq_flags = RQF_ALLOCED;
88ee5ef1 1183
2fff8a92
BVA
1184 lockdep_assert_held(q->queue_lock);
1185
3f3299d5 1186 if (unlikely(blk_queue_dying(q)))
a492f075 1187 return ERR_PTR(-ENODEV);
da8303c6 1188
ef295ecf 1189 may_queue = elv_may_queue(q, op);
88ee5ef1
JA
1190 if (may_queue == ELV_MQUEUE_NO)
1191 goto rq_starved;
1192
1faa16d2
JA
1193 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1194 if (rl->count[is_sync]+1 >= q->nr_requests) {
88ee5ef1
JA
1195 /*
1196 * The queue will fill after this allocation, so set
1197 * it as full, and mark this process as "batching".
1198 * This process will be allowed to complete a batch of
1199 * requests, others will be blocked.
1200 */
5b788ce3 1201 if (!blk_rl_full(rl, is_sync)) {
88ee5ef1 1202 ioc_set_batching(q, ioc);
5b788ce3 1203 blk_set_rl_full(rl, is_sync);
88ee5ef1
JA
1204 } else {
1205 if (may_queue != ELV_MQUEUE_MUST
1206 && !ioc_batching(q, ioc)) {
1207 /*
1208 * The queue is full and the allocating
1209 * process is not a "batcher", and not
1210 * exempted by the IO scheduler
1211 */
a492f075 1212 return ERR_PTR(-ENOMEM);
88ee5ef1
JA
1213 }
1214 }
1da177e4 1215 }
d40f75a0 1216 blk_set_congested(rl, is_sync);
1da177e4
LT
1217 }
1218
082cf69e
JA
1219 /*
1220 * Only allow batching queuers to allocate up to 50% over the defined
1221 * limit of requests, otherwise we could have thousands of requests
1222 * allocated with any setting of ->nr_requests
1223 */
1faa16d2 1224 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
a492f075 1225 return ERR_PTR(-ENOMEM);
fd782a4a 1226
8a5ecdd4 1227 q->nr_rqs[is_sync]++;
1faa16d2
JA
1228 rl->count[is_sync]++;
1229 rl->starved[is_sync] = 0;
cb98fc8b 1230
f1f8cc94
TH
1231 /*
1232 * Decide whether the new request will be managed by elevator. If
e8064021 1233 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
f1f8cc94
TH
1234 * prevent the current elevator from being destroyed until the new
1235 * request is freed. This guarantees icq's won't be destroyed and
1236 * makes creating new ones safe.
1237 *
e6f7f93d
CH
1238 * Flush requests do not use the elevator so skip initialization.
1239 * This allows a request to share the flush and elevator data.
1240 *
f1f8cc94
TH
1241 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1242 * it will be created after releasing queue_lock.
1243 */
e6f7f93d 1244 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
e8064021 1245 rq_flags |= RQF_ELVPRIV;
8a5ecdd4 1246 q->nr_rqs_elvpriv++;
f1f8cc94
TH
1247 if (et->icq_cache && ioc)
1248 icq = ioc_lookup_icq(ioc, q);
9d5a4e94 1249 }
cb98fc8b 1250
f253b86b 1251 if (blk_queue_io_stat(q))
e8064021 1252 rq_flags |= RQF_IO_STAT;
1da177e4
LT
1253 spin_unlock_irq(q->queue_lock);
1254
29e2b09a 1255 /* allocate and init request */
5b788ce3 1256 rq = mempool_alloc(rl->rq_pool, gfp_mask);
29e2b09a 1257 if (!rq)
b679281a 1258 goto fail_alloc;
1da177e4 1259
29e2b09a 1260 blk_rq_init(q, rq);
a051661c 1261 blk_rq_set_rl(rq, rl);
ef295ecf 1262 rq->cmd_flags = op;
e8064021 1263 rq->rq_flags = rq_flags;
29e2b09a 1264
aaf7c680 1265 /* init elvpriv */
e8064021 1266 if (rq_flags & RQF_ELVPRIV) {
aaf7c680 1267 if (unlikely(et->icq_cache && !icq)) {
7f4b35d1
TH
1268 if (ioc)
1269 icq = ioc_create_icq(ioc, q, gfp_mask);
aaf7c680
TH
1270 if (!icq)
1271 goto fail_elvpriv;
29e2b09a 1272 }
aaf7c680
TH
1273
1274 rq->elv.icq = icq;
1275 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1276 goto fail_elvpriv;
1277
1278 /* @rq->elv.icq holds io_context until @rq is freed */
29e2b09a
TH
1279 if (icq)
1280 get_io_context(icq->ioc);
1281 }
aaf7c680 1282out:
88ee5ef1
JA
1283 /*
1284 * ioc may be NULL here, and ioc_batching will be false. That's
1285 * OK, if the queue is under the request limit then requests need
1286 * not count toward the nr_batch_requests limit. There will always
1287 * be some limit enforced by BLK_BATCH_TIME.
1288 */
1da177e4
LT
1289 if (ioc_batching(q, ioc))
1290 ioc->nr_batch_requests--;
6728cb0e 1291
e6a40b09 1292 trace_block_getrq(q, bio, op);
1da177e4 1293 return rq;
b679281a 1294
aaf7c680
TH
1295fail_elvpriv:
1296 /*
1297 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1298 * and may fail indefinitely under memory pressure and thus
1299 * shouldn't stall IO. Treat this request as !elvpriv. This will
1300 * disturb iosched and blkcg but weird is bettern than dead.
1301 */
7b2b10e0 1302 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
dc3b17cc 1303 __func__, dev_name(q->backing_dev_info->dev));
aaf7c680 1304
e8064021 1305 rq->rq_flags &= ~RQF_ELVPRIV;
aaf7c680
TH
1306 rq->elv.icq = NULL;
1307
1308 spin_lock_irq(q->queue_lock);
8a5ecdd4 1309 q->nr_rqs_elvpriv--;
aaf7c680
TH
1310 spin_unlock_irq(q->queue_lock);
1311 goto out;
1312
b679281a
TH
1313fail_alloc:
1314 /*
1315 * Allocation failed presumably due to memory. Undo anything we
1316 * might have messed up.
1317 *
1318 * Allocating task should really be put onto the front of the wait
1319 * queue, but this is pretty rare.
1320 */
1321 spin_lock_irq(q->queue_lock);
e8064021 1322 freed_request(rl, is_sync, rq_flags);
b679281a
TH
1323
1324 /*
1325 * in the very unlikely event that allocation failed and no
1326 * requests for this direction was pending, mark us starved so that
1327 * freeing of a request in the other direction will notice
1328 * us. another possible fix would be to split the rq mempool into
1329 * READ and WRITE
1330 */
1331rq_starved:
1332 if (unlikely(rl->count[is_sync] == 0))
1333 rl->starved[is_sync] = 1;
a492f075 1334 return ERR_PTR(-ENOMEM);
1da177e4
LT
1335}
1336
da8303c6 1337/**
a06e05e6 1338 * get_request - get a free request
da8303c6 1339 * @q: request_queue to allocate request from
ef295ecf 1340 * @op: operation and flags
da8303c6 1341 * @bio: bio to allocate request for (can be %NULL)
a06e05e6 1342 * @gfp_mask: allocation mask
da8303c6 1343 *
d0164adc
MG
1344 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1345 * this function keeps retrying under memory pressure and fails iff @q is dead.
d6344532 1346 *
da3dae54 1347 * Must be called with @q->queue_lock held and,
a492f075
JL
1348 * Returns ERR_PTR on failure, with @q->queue_lock held.
1349 * Returns request pointer on success, with @q->queue_lock *not held*.
1da177e4 1350 */
ef295ecf
CH
1351static struct request *get_request(struct request_queue *q, unsigned int op,
1352 struct bio *bio, gfp_t gfp_mask)
1da177e4 1353{
ef295ecf 1354 const bool is_sync = op_is_sync(op);
a06e05e6 1355 DEFINE_WAIT(wait);
a051661c 1356 struct request_list *rl;
1da177e4 1357 struct request *rq;
a051661c 1358
2fff8a92 1359 lockdep_assert_held(q->queue_lock);
332ebbf7 1360 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1361
a051661c 1362 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
a06e05e6 1363retry:
ef295ecf 1364 rq = __get_request(rl, op, bio, gfp_mask);
a492f075 1365 if (!IS_ERR(rq))
a06e05e6 1366 return rq;
1da177e4 1367
03a07c92
GR
1368 if (op & REQ_NOWAIT) {
1369 blk_put_rl(rl);
1370 return ERR_PTR(-EAGAIN);
1371 }
1372
d0164adc 1373 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
a051661c 1374 blk_put_rl(rl);
a492f075 1375 return rq;
a051661c 1376 }
1da177e4 1377
a06e05e6
TH
1378 /* wait on @rl and retry */
1379 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1380 TASK_UNINTERRUPTIBLE);
1da177e4 1381
e6a40b09 1382 trace_block_sleeprq(q, bio, op);
1da177e4 1383
a06e05e6
TH
1384 spin_unlock_irq(q->queue_lock);
1385 io_schedule();
d6344532 1386
a06e05e6
TH
1387 /*
1388 * After sleeping, we become a "batching" process and will be able
1389 * to allocate at least one request, and up to a big batch of them
1390 * for a small period time. See ioc_batching, ioc_set_batching
1391 */
a06e05e6 1392 ioc_set_batching(q, current->io_context);
05caf8db 1393
a06e05e6
TH
1394 spin_lock_irq(q->queue_lock);
1395 finish_wait(&rl->wait[is_sync], &wait);
1da177e4 1396
a06e05e6 1397 goto retry;
1da177e4
LT
1398}
1399
cd6ce148
BVA
1400static struct request *blk_old_get_request(struct request_queue *q,
1401 unsigned int op, gfp_t gfp_mask)
1da177e4
LT
1402{
1403 struct request *rq;
055f6e18 1404 int ret = 0;
1da177e4 1405
332ebbf7
BVA
1406 WARN_ON_ONCE(q->mq_ops);
1407
7f4b35d1
TH
1408 /* create ioc upfront */
1409 create_io_context(gfp_mask, q->node);
1410
055f6e18
ML
1411 ret = blk_queue_enter(q, !(gfp_mask & __GFP_DIRECT_RECLAIM) ||
1412 (op & REQ_NOWAIT));
1413 if (ret)
1414 return ERR_PTR(ret);
d6344532 1415 spin_lock_irq(q->queue_lock);
cd6ce148 1416 rq = get_request(q, op, NULL, gfp_mask);
0c4de0f3 1417 if (IS_ERR(rq)) {
da8303c6 1418 spin_unlock_irq(q->queue_lock);
055f6e18 1419 blk_queue_exit(q);
0c4de0f3
CH
1420 return rq;
1421 }
1da177e4 1422
0c4de0f3
CH
1423 /* q->queue_lock is unlocked at this point */
1424 rq->__data_len = 0;
1425 rq->__sector = (sector_t) -1;
1426 rq->bio = rq->biotail = NULL;
1da177e4
LT
1427 return rq;
1428}
320ae51f 1429
cd6ce148
BVA
1430struct request *blk_get_request(struct request_queue *q, unsigned int op,
1431 gfp_t gfp_mask)
320ae51f 1432{
d280bab3
BVA
1433 struct request *req;
1434
1435 if (q->mq_ops) {
1436 req = blk_mq_alloc_request(q, op,
6f3b0e8b
CH
1437 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1438 0 : BLK_MQ_REQ_NOWAIT);
d280bab3
BVA
1439 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1440 q->mq_ops->initialize_rq_fn(req);
1441 } else {
1442 req = blk_old_get_request(q, op, gfp_mask);
1443 if (!IS_ERR(req) && q->initialize_rq_fn)
1444 q->initialize_rq_fn(req);
1445 }
1446
1447 return req;
320ae51f 1448}
1da177e4
LT
1449EXPORT_SYMBOL(blk_get_request);
1450
1451/**
1452 * blk_requeue_request - put a request back on queue
1453 * @q: request queue where request should be inserted
1454 * @rq: request to be inserted
1455 *
1456 * Description:
1457 * Drivers often keep queueing requests until the hardware cannot accept
1458 * more, when that condition happens we need to put the request back
1459 * on the queue. Must be called with queue lock held.
1460 */
165125e1 1461void blk_requeue_request(struct request_queue *q, struct request *rq)
1da177e4 1462{
2fff8a92 1463 lockdep_assert_held(q->queue_lock);
332ebbf7 1464 WARN_ON_ONCE(q->mq_ops);
2fff8a92 1465
242f9dcb
JA
1466 blk_delete_timer(rq);
1467 blk_clear_rq_complete(rq);
5f3ea37c 1468 trace_block_rq_requeue(q, rq);
87760e5e 1469 wbt_requeue(q->rq_wb, &rq->issue_stat);
2056a782 1470
e8064021 1471 if (rq->rq_flags & RQF_QUEUED)
1da177e4
LT
1472 blk_queue_end_tag(q, rq);
1473
ba396a6c
JB
1474 BUG_ON(blk_queued_rq(rq));
1475
1da177e4
LT
1476 elv_requeue_request(q, rq);
1477}
1da177e4
LT
1478EXPORT_SYMBOL(blk_requeue_request);
1479
73c10101
JA
1480static void add_acct_request(struct request_queue *q, struct request *rq,
1481 int where)
1482{
320ae51f 1483 blk_account_io_start(rq, true);
7eaceacc 1484 __elv_add_request(q, rq, where);
73c10101
JA
1485}
1486
d62e26b3 1487static void part_round_stats_single(struct request_queue *q, int cpu,
b8d62b3a
JA
1488 struct hd_struct *part, unsigned long now,
1489 unsigned int inflight)
074a7aca 1490{
b8d62b3a 1491 if (inflight) {
074a7aca 1492 __part_stat_add(cpu, part, time_in_queue,
b8d62b3a 1493 inflight * (now - part->stamp));
074a7aca
TH
1494 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1495 }
1496 part->stamp = now;
1497}
1498
1499/**
496aa8a9 1500 * part_round_stats() - Round off the performance stats on a struct disk_stats.
d62e26b3 1501 * @q: target block queue
496aa8a9
RD
1502 * @cpu: cpu number for stats access
1503 * @part: target partition
1da177e4
LT
1504 *
1505 * The average IO queue length and utilisation statistics are maintained
1506 * by observing the current state of the queue length and the amount of
1507 * time it has been in this state for.
1508 *
1509 * Normally, that accounting is done on IO completion, but that can result
1510 * in more than a second's worth of IO being accounted for within any one
1511 * second, leading to >100% utilisation. To deal with that, we call this
1512 * function to do a round-off before returning the results when reading
1513 * /proc/diskstats. This accounts immediately for all queue usage up to
1514 * the current jiffies and restarts the counters again.
1515 */
d62e26b3 1516void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
6f2576af 1517{
b8d62b3a 1518 struct hd_struct *part2 = NULL;
6f2576af 1519 unsigned long now = jiffies;
b8d62b3a
JA
1520 unsigned int inflight[2];
1521 int stats = 0;
1522
1523 if (part->stamp != now)
1524 stats |= 1;
1525
1526 if (part->partno) {
1527 part2 = &part_to_disk(part)->part0;
1528 if (part2->stamp != now)
1529 stats |= 2;
1530 }
1531
1532 if (!stats)
1533 return;
1534
1535 part_in_flight(q, part, inflight);
6f2576af 1536
b8d62b3a
JA
1537 if (stats & 2)
1538 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1539 if (stats & 1)
1540 part_round_stats_single(q, cpu, part, now, inflight[0]);
6f2576af 1541}
074a7aca 1542EXPORT_SYMBOL_GPL(part_round_stats);
6f2576af 1543
47fafbc7 1544#ifdef CONFIG_PM
c8158819
LM
1545static void blk_pm_put_request(struct request *rq)
1546{
e8064021 1547 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
c8158819
LM
1548 pm_runtime_mark_last_busy(rq->q->dev);
1549}
1550#else
1551static inline void blk_pm_put_request(struct request *rq) {}
1552#endif
1553
165125e1 1554void __blk_put_request(struct request_queue *q, struct request *req)
1da177e4 1555{
e8064021
CH
1556 req_flags_t rq_flags = req->rq_flags;
1557
1da177e4
LT
1558 if (unlikely(!q))
1559 return;
1da177e4 1560
6f5ba581
CH
1561 if (q->mq_ops) {
1562 blk_mq_free_request(req);
1563 return;
1564 }
1565
2fff8a92
BVA
1566 lockdep_assert_held(q->queue_lock);
1567
c8158819
LM
1568 blk_pm_put_request(req);
1569
8922e16c
TH
1570 elv_completed_request(q, req);
1571
1cd96c24
BH
1572 /* this is a bio leak */
1573 WARN_ON(req->bio != NULL);
1574
87760e5e
JA
1575 wbt_done(q->rq_wb, &req->issue_stat);
1576
1da177e4
LT
1577 /*
1578 * Request may not have originated from ll_rw_blk. if not,
1579 * it didn't come out of our reserved rq pools
1580 */
e8064021 1581 if (rq_flags & RQF_ALLOCED) {
a051661c 1582 struct request_list *rl = blk_rq_rl(req);
ef295ecf 1583 bool sync = op_is_sync(req->cmd_flags);
1da177e4 1584
1da177e4 1585 BUG_ON(!list_empty(&req->queuelist));
360f92c2 1586 BUG_ON(ELV_ON_HASH(req));
1da177e4 1587
a051661c 1588 blk_free_request(rl, req);
e8064021 1589 freed_request(rl, sync, rq_flags);
a051661c 1590 blk_put_rl(rl);
055f6e18 1591 blk_queue_exit(q);
1da177e4
LT
1592 }
1593}
6e39b69e
MC
1594EXPORT_SYMBOL_GPL(__blk_put_request);
1595
1da177e4
LT
1596void blk_put_request(struct request *req)
1597{
165125e1 1598 struct request_queue *q = req->q;
8922e16c 1599
320ae51f
JA
1600 if (q->mq_ops)
1601 blk_mq_free_request(req);
1602 else {
1603 unsigned long flags;
1604
1605 spin_lock_irqsave(q->queue_lock, flags);
1606 __blk_put_request(q, req);
1607 spin_unlock_irqrestore(q->queue_lock, flags);
1608 }
1da177e4 1609}
1da177e4
LT
1610EXPORT_SYMBOL(blk_put_request);
1611
320ae51f
JA
1612bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1613 struct bio *bio)
73c10101 1614{
1eff9d32 1615 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1616
73c10101
JA
1617 if (!ll_back_merge_fn(q, req, bio))
1618 return false;
1619
8c1cf6bb 1620 trace_block_bio_backmerge(q, req, bio);
73c10101
JA
1621
1622 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1623 blk_rq_set_mixed_merge(req);
1624
1625 req->biotail->bi_next = bio;
1626 req->biotail = bio;
4f024f37 1627 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1628 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1629
320ae51f 1630 blk_account_io_start(req, false);
73c10101
JA
1631 return true;
1632}
1633
320ae51f
JA
1634bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1635 struct bio *bio)
73c10101 1636{
1eff9d32 1637 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
73c10101 1638
73c10101
JA
1639 if (!ll_front_merge_fn(q, req, bio))
1640 return false;
1641
8c1cf6bb 1642 trace_block_bio_frontmerge(q, req, bio);
73c10101
JA
1643
1644 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1645 blk_rq_set_mixed_merge(req);
1646
73c10101
JA
1647 bio->bi_next = req->bio;
1648 req->bio = bio;
1649
4f024f37
KO
1650 req->__sector = bio->bi_iter.bi_sector;
1651 req->__data_len += bio->bi_iter.bi_size;
73c10101
JA
1652 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1653
320ae51f 1654 blk_account_io_start(req, false);
73c10101
JA
1655 return true;
1656}
1657
1e739730
CH
1658bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1659 struct bio *bio)
1660{
1661 unsigned short segments = blk_rq_nr_discard_segments(req);
1662
1663 if (segments >= queue_max_discard_segments(q))
1664 goto no_merge;
1665 if (blk_rq_sectors(req) + bio_sectors(bio) >
1666 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1667 goto no_merge;
1668
1669 req->biotail->bi_next = bio;
1670 req->biotail = bio;
1671 req->__data_len += bio->bi_iter.bi_size;
1672 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1673 req->nr_phys_segments = segments + 1;
1674
1675 blk_account_io_start(req, false);
1676 return true;
1677no_merge:
1678 req_set_nomerge(q, req);
1679 return false;
1680}
1681
bd87b589 1682/**
320ae51f 1683 * blk_attempt_plug_merge - try to merge with %current's plugged list
bd87b589
TH
1684 * @q: request_queue new bio is being queued at
1685 * @bio: new bio being queued
1686 * @request_count: out parameter for number of traversed plugged requests
ccc2600b
RD
1687 * @same_queue_rq: pointer to &struct request that gets filled in when
1688 * another request associated with @q is found on the plug list
1689 * (optional, may be %NULL)
bd87b589
TH
1690 *
1691 * Determine whether @bio being queued on @q can be merged with a request
1692 * on %current's plugged list. Returns %true if merge was successful,
1693 * otherwise %false.
1694 *
07c2bd37
TH
1695 * Plugging coalesces IOs from the same issuer for the same purpose without
1696 * going through @q->queue_lock. As such it's more of an issuing mechanism
1697 * than scheduling, and the request, while may have elvpriv data, is not
1698 * added on the elevator at this point. In addition, we don't have
1699 * reliable access to the elevator outside queue lock. Only check basic
1700 * merging parameters without querying the elevator.
da41a589
RE
1701 *
1702 * Caller must ensure !blk_queue_nomerges(q) beforehand.
73c10101 1703 */
320ae51f 1704bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
5b3f341f
SL
1705 unsigned int *request_count,
1706 struct request **same_queue_rq)
73c10101
JA
1707{
1708 struct blk_plug *plug;
1709 struct request *rq;
92f399c7 1710 struct list_head *plug_list;
73c10101 1711
bd87b589 1712 plug = current->plug;
73c10101 1713 if (!plug)
34fe7c05 1714 return false;
56ebdaf2 1715 *request_count = 0;
73c10101 1716
92f399c7
SL
1717 if (q->mq_ops)
1718 plug_list = &plug->mq_list;
1719 else
1720 plug_list = &plug->list;
1721
1722 list_for_each_entry_reverse(rq, plug_list, queuelist) {
34fe7c05 1723 bool merged = false;
73c10101 1724
5b3f341f 1725 if (rq->q == q) {
1b2e19f1 1726 (*request_count)++;
5b3f341f
SL
1727 /*
1728 * Only blk-mq multiple hardware queues case checks the
1729 * rq in the same queue, there should be only one such
1730 * rq in a queue
1731 **/
1732 if (same_queue_rq)
1733 *same_queue_rq = rq;
1734 }
56ebdaf2 1735
07c2bd37 1736 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
73c10101
JA
1737 continue;
1738
34fe7c05
CH
1739 switch (blk_try_merge(rq, bio)) {
1740 case ELEVATOR_BACK_MERGE:
1741 merged = bio_attempt_back_merge(q, rq, bio);
1742 break;
1743 case ELEVATOR_FRONT_MERGE:
1744 merged = bio_attempt_front_merge(q, rq, bio);
1745 break;
1e739730
CH
1746 case ELEVATOR_DISCARD_MERGE:
1747 merged = bio_attempt_discard_merge(q, rq, bio);
1748 break;
34fe7c05
CH
1749 default:
1750 break;
73c10101 1751 }
34fe7c05
CH
1752
1753 if (merged)
1754 return true;
73c10101 1755 }
34fe7c05
CH
1756
1757 return false;
73c10101
JA
1758}
1759
0809e3ac
JM
1760unsigned int blk_plug_queued_count(struct request_queue *q)
1761{
1762 struct blk_plug *plug;
1763 struct request *rq;
1764 struct list_head *plug_list;
1765 unsigned int ret = 0;
1766
1767 plug = current->plug;
1768 if (!plug)
1769 goto out;
1770
1771 if (q->mq_ops)
1772 plug_list = &plug->mq_list;
1773 else
1774 plug_list = &plug->list;
1775
1776 list_for_each_entry(rq, plug_list, queuelist) {
1777 if (rq->q == q)
1778 ret++;
1779 }
1780out:
1781 return ret;
1782}
1783
da8d7f07 1784void blk_init_request_from_bio(struct request *req, struct bio *bio)
52d9e675 1785{
0be0dee6
BVA
1786 struct io_context *ioc = rq_ioc(bio);
1787
1eff9d32 1788 if (bio->bi_opf & REQ_RAHEAD)
a82afdfc 1789 req->cmd_flags |= REQ_FAILFAST_MASK;
b31dc66a 1790
4f024f37 1791 req->__sector = bio->bi_iter.bi_sector;
5dc8b362
AM
1792 if (ioprio_valid(bio_prio(bio)))
1793 req->ioprio = bio_prio(bio);
0be0dee6
BVA
1794 else if (ioc)
1795 req->ioprio = ioc->ioprio;
1796 else
1797 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
cb6934f8 1798 req->write_hint = bio->bi_write_hint;
bc1c56fd 1799 blk_rq_bio_prep(req->q, req, bio);
52d9e675 1800}
da8d7f07 1801EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
52d9e675 1802
dece1635 1803static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1da177e4 1804{
73c10101 1805 struct blk_plug *plug;
34fe7c05 1806 int where = ELEVATOR_INSERT_SORT;
e4d750c9 1807 struct request *req, *free;
56ebdaf2 1808 unsigned int request_count = 0;
87760e5e 1809 unsigned int wb_acct;
1da177e4 1810
1da177e4
LT
1811 /*
1812 * low level driver can indicate that it wants pages above a
1813 * certain limit bounced to low memory (ie for highmem, or even
1814 * ISA dma in theory)
1815 */
1816 blk_queue_bounce(q, &bio);
1817
af67c31f 1818 blk_queue_split(q, &bio);
23688bf4 1819
e23947bd 1820 if (!bio_integrity_prep(bio))
dece1635 1821 return BLK_QC_T_NONE;
ffecfd1a 1822
f73f44eb 1823 if (op_is_flush(bio->bi_opf)) {
73c10101 1824 spin_lock_irq(q->queue_lock);
ae1b1539 1825 where = ELEVATOR_INSERT_FLUSH;
28e7d184
TH
1826 goto get_rq;
1827 }
1828
73c10101
JA
1829 /*
1830 * Check if we can merge with the plugged list before grabbing
1831 * any locks.
1832 */
0809e3ac
JM
1833 if (!blk_queue_nomerges(q)) {
1834 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
dece1635 1835 return BLK_QC_T_NONE;
0809e3ac
JM
1836 } else
1837 request_count = blk_plug_queued_count(q);
1da177e4 1838
73c10101 1839 spin_lock_irq(q->queue_lock);
2056a782 1840
34fe7c05
CH
1841 switch (elv_merge(q, &req, bio)) {
1842 case ELEVATOR_BACK_MERGE:
1843 if (!bio_attempt_back_merge(q, req, bio))
1844 break;
1845 elv_bio_merged(q, req, bio);
1846 free = attempt_back_merge(q, req);
1847 if (free)
1848 __blk_put_request(q, free);
1849 else
1850 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1851 goto out_unlock;
1852 case ELEVATOR_FRONT_MERGE:
1853 if (!bio_attempt_front_merge(q, req, bio))
1854 break;
1855 elv_bio_merged(q, req, bio);
1856 free = attempt_front_merge(q, req);
1857 if (free)
1858 __blk_put_request(q, free);
1859 else
1860 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1861 goto out_unlock;
1862 default:
1863 break;
1da177e4
LT
1864 }
1865
450991bc 1866get_rq:
87760e5e
JA
1867 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1868
1da177e4 1869 /*
450991bc 1870 * Grab a free request. This is might sleep but can not fail.
d6344532 1871 * Returns with the queue unlocked.
450991bc 1872 */
055f6e18 1873 blk_queue_enter_live(q);
ef295ecf 1874 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
a492f075 1875 if (IS_ERR(req)) {
055f6e18 1876 blk_queue_exit(q);
87760e5e 1877 __wbt_done(q->rq_wb, wb_acct);
4e4cbee9
CH
1878 if (PTR_ERR(req) == -ENOMEM)
1879 bio->bi_status = BLK_STS_RESOURCE;
1880 else
1881 bio->bi_status = BLK_STS_IOERR;
4246a0b6 1882 bio_endio(bio);
da8303c6
TH
1883 goto out_unlock;
1884 }
d6344532 1885
87760e5e
JA
1886 wbt_track(&req->issue_stat, wb_acct);
1887
450991bc
NP
1888 /*
1889 * After dropping the lock and possibly sleeping here, our request
1890 * may now be mergeable after it had proven unmergeable (above).
1891 * We don't worry about that case for efficiency. It won't happen
1892 * often, and the elevators are able to handle it.
1da177e4 1893 */
da8d7f07 1894 blk_init_request_from_bio(req, bio);
1da177e4 1895
9562ad9a 1896 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
11ccf116 1897 req->cpu = raw_smp_processor_id();
73c10101
JA
1898
1899 plug = current->plug;
721a9602 1900 if (plug) {
dc6d36c9
JA
1901 /*
1902 * If this is the first request added after a plug, fire
7aef2e78 1903 * of a plug trace.
0a6219a9
ML
1904 *
1905 * @request_count may become stale because of schedule
1906 * out, so check plug list again.
dc6d36c9 1907 */
0a6219a9 1908 if (!request_count || list_empty(&plug->list))
dc6d36c9 1909 trace_block_plug(q);
3540d5e8 1910 else {
50d24c34
SL
1911 struct request *last = list_entry_rq(plug->list.prev);
1912 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1913 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
3540d5e8 1914 blk_flush_plug_list(plug, false);
019ceb7d
SL
1915 trace_block_plug(q);
1916 }
73c10101 1917 }
73c10101 1918 list_add_tail(&req->queuelist, &plug->list);
320ae51f 1919 blk_account_io_start(req, true);
73c10101
JA
1920 } else {
1921 spin_lock_irq(q->queue_lock);
1922 add_acct_request(q, req, where);
24ecfbe2 1923 __blk_run_queue(q);
73c10101
JA
1924out_unlock:
1925 spin_unlock_irq(q->queue_lock);
1926 }
dece1635
JA
1927
1928 return BLK_QC_T_NONE;
1da177e4
LT
1929}
1930
1da177e4
LT
1931static void handle_bad_sector(struct bio *bio)
1932{
1933 char b[BDEVNAME_SIZE];
1934
1935 printk(KERN_INFO "attempt to access beyond end of device\n");
6296b960 1936 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
74d46992 1937 bio_devname(bio, b), bio->bi_opf,
f73a1c7d 1938 (unsigned long long)bio_end_sector(bio),
74d46992 1939 (long long)get_capacity(bio->bi_disk));
1da177e4
LT
1940}
1941
c17bb495
AM
1942#ifdef CONFIG_FAIL_MAKE_REQUEST
1943
1944static DECLARE_FAULT_ATTR(fail_make_request);
1945
1946static int __init setup_fail_make_request(char *str)
1947{
1948 return setup_fault_attr(&fail_make_request, str);
1949}
1950__setup("fail_make_request=", setup_fail_make_request);
1951
b2c9cd37 1952static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
c17bb495 1953{
b2c9cd37 1954 return part->make_it_fail && should_fail(&fail_make_request, bytes);
c17bb495
AM
1955}
1956
1957static int __init fail_make_request_debugfs(void)
1958{
dd48c085
AM
1959 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1960 NULL, &fail_make_request);
1961
21f9fcd8 1962 return PTR_ERR_OR_ZERO(dir);
c17bb495
AM
1963}
1964
1965late_initcall(fail_make_request_debugfs);
1966
1967#else /* CONFIG_FAIL_MAKE_REQUEST */
1968
b2c9cd37
AM
1969static inline bool should_fail_request(struct hd_struct *part,
1970 unsigned int bytes)
c17bb495 1971{
b2c9cd37 1972 return false;
c17bb495
AM
1973}
1974
1975#endif /* CONFIG_FAIL_MAKE_REQUEST */
1976
74d46992
CH
1977/*
1978 * Remap block n of partition p to block n+start(p) of the disk.
1979 */
1980static inline int blk_partition_remap(struct bio *bio)
1981{
1982 struct hd_struct *p;
1983 int ret = 0;
1984
1985 /*
1986 * Zone reset does not include bi_size so bio_sectors() is always 0.
1987 * Include a test for the reset op code and perform the remap if needed.
1988 */
1989 if (!bio->bi_partno ||
1990 (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET))
1991 return 0;
1992
1993 rcu_read_lock();
1994 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
1995 if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) {
1996 bio->bi_iter.bi_sector += p->start_sect;
1997 bio->bi_partno = 0;
1998 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
1999 bio->bi_iter.bi_sector - p->start_sect);
2000 } else {
2001 printk("%s: fail for partition %d\n", __func__, bio->bi_partno);
2002 ret = -EIO;
2003 }
2004 rcu_read_unlock();
2005
2006 return ret;
2007}
2008
c07e2b41
JA
2009/*
2010 * Check whether this bio extends beyond the end of the device.
2011 */
2012static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
2013{
2014 sector_t maxsector;
2015
2016 if (!nr_sectors)
2017 return 0;
2018
2019 /* Test device or partition size, when known. */
74d46992 2020 maxsector = get_capacity(bio->bi_disk);
c07e2b41 2021 if (maxsector) {
4f024f37 2022 sector_t sector = bio->bi_iter.bi_sector;
c07e2b41
JA
2023
2024 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2025 /*
2026 * This may well happen - the kernel calls bread()
2027 * without checking the size of the device, e.g., when
2028 * mounting a device.
2029 */
2030 handle_bad_sector(bio);
2031 return 1;
2032 }
2033 }
2034
2035 return 0;
2036}
2037
27a84d54
CH
2038static noinline_for_stack bool
2039generic_make_request_checks(struct bio *bio)
1da177e4 2040{
165125e1 2041 struct request_queue *q;
5a7bbad2 2042 int nr_sectors = bio_sectors(bio);
4e4cbee9 2043 blk_status_t status = BLK_STS_IOERR;
5a7bbad2 2044 char b[BDEVNAME_SIZE];
1da177e4
LT
2045
2046 might_sleep();
1da177e4 2047
c07e2b41
JA
2048 if (bio_check_eod(bio, nr_sectors))
2049 goto end_io;
1da177e4 2050
74d46992 2051 q = bio->bi_disk->queue;
5a7bbad2
CH
2052 if (unlikely(!q)) {
2053 printk(KERN_ERR
2054 "generic_make_request: Trying to access "
2055 "nonexistent block-device %s (%Lu)\n",
74d46992 2056 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
5a7bbad2
CH
2057 goto end_io;
2058 }
c17bb495 2059
03a07c92
GR
2060 /*
2061 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2062 * if queue is not a request based queue.
2063 */
2064
2065 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2066 goto not_supported;
2067
74d46992 2068 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
5a7bbad2 2069 goto end_io;
2056a782 2070
74d46992
CH
2071 if (blk_partition_remap(bio))
2072 goto end_io;
2056a782 2073
5a7bbad2
CH
2074 if (bio_check_eod(bio, nr_sectors))
2075 goto end_io;
1e87901e 2076
5a7bbad2
CH
2077 /*
2078 * Filter flush bio's early so that make_request based
2079 * drivers without flush support don't have to worry
2080 * about them.
2081 */
f3a8ab7d 2082 if (op_is_flush(bio->bi_opf) &&
c888a8f9 2083 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1eff9d32 2084 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
5a7bbad2 2085 if (!nr_sectors) {
4e4cbee9 2086 status = BLK_STS_OK;
51fd77bd
JA
2087 goto end_io;
2088 }
5a7bbad2 2089 }
5ddfe969 2090
288dab8a
CH
2091 switch (bio_op(bio)) {
2092 case REQ_OP_DISCARD:
2093 if (!blk_queue_discard(q))
2094 goto not_supported;
2095 break;
2096 case REQ_OP_SECURE_ERASE:
2097 if (!blk_queue_secure_erase(q))
2098 goto not_supported;
2099 break;
2100 case REQ_OP_WRITE_SAME:
74d46992 2101 if (!q->limits.max_write_same_sectors)
288dab8a 2102 goto not_supported;
58886785 2103 break;
2d253440
ST
2104 case REQ_OP_ZONE_REPORT:
2105 case REQ_OP_ZONE_RESET:
74d46992 2106 if (!blk_queue_is_zoned(q))
2d253440 2107 goto not_supported;
288dab8a 2108 break;
a6f0788e 2109 case REQ_OP_WRITE_ZEROES:
74d46992 2110 if (!q->limits.max_write_zeroes_sectors)
a6f0788e
CK
2111 goto not_supported;
2112 break;
288dab8a
CH
2113 default:
2114 break;
5a7bbad2 2115 }
01edede4 2116
7f4b35d1
TH
2117 /*
2118 * Various block parts want %current->io_context and lazy ioc
2119 * allocation ends up trading a lot of pain for a small amount of
2120 * memory. Just allocate it upfront. This may fail and block
2121 * layer knows how to live with it.
2122 */
2123 create_io_context(GFP_ATOMIC, q->node);
2124
ae118896
TH
2125 if (!blkcg_bio_issue_check(q, bio))
2126 return false;
27a84d54 2127
fbbaf700
N
2128 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2129 trace_block_bio_queue(q, bio);
2130 /* Now that enqueuing has been traced, we need to trace
2131 * completion as well.
2132 */
2133 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2134 }
27a84d54 2135 return true;
a7384677 2136
288dab8a 2137not_supported:
4e4cbee9 2138 status = BLK_STS_NOTSUPP;
a7384677 2139end_io:
4e4cbee9 2140 bio->bi_status = status;
4246a0b6 2141 bio_endio(bio);
27a84d54 2142 return false;
1da177e4
LT
2143}
2144
27a84d54
CH
2145/**
2146 * generic_make_request - hand a buffer to its device driver for I/O
2147 * @bio: The bio describing the location in memory and on the device.
2148 *
2149 * generic_make_request() is used to make I/O requests of block
2150 * devices. It is passed a &struct bio, which describes the I/O that needs
2151 * to be done.
2152 *
2153 * generic_make_request() does not return any status. The
2154 * success/failure status of the request, along with notification of
2155 * completion, is delivered asynchronously through the bio->bi_end_io
2156 * function described (one day) else where.
2157 *
2158 * The caller of generic_make_request must make sure that bi_io_vec
2159 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2160 * set to describe the device address, and the
2161 * bi_end_io and optionally bi_private are set to describe how
2162 * completion notification should be signaled.
2163 *
2164 * generic_make_request and the drivers it calls may use bi_next if this
2165 * bio happens to be merged with someone else, and may resubmit the bio to
2166 * a lower device by calling into generic_make_request recursively, which
2167 * means the bio should NOT be touched after the call to ->make_request_fn.
d89d8796 2168 */
dece1635 2169blk_qc_t generic_make_request(struct bio *bio)
d89d8796 2170{
f5fe1b51
N
2171 /*
2172 * bio_list_on_stack[0] contains bios submitted by the current
2173 * make_request_fn.
2174 * bio_list_on_stack[1] contains bios that were submitted before
2175 * the current make_request_fn, but that haven't been processed
2176 * yet.
2177 */
2178 struct bio_list bio_list_on_stack[2];
dece1635 2179 blk_qc_t ret = BLK_QC_T_NONE;
bddd87c7 2180
27a84d54 2181 if (!generic_make_request_checks(bio))
dece1635 2182 goto out;
27a84d54
CH
2183
2184 /*
2185 * We only want one ->make_request_fn to be active at a time, else
2186 * stack usage with stacked devices could be a problem. So use
2187 * current->bio_list to keep a list of requests submited by a
2188 * make_request_fn function. current->bio_list is also used as a
2189 * flag to say if generic_make_request is currently active in this
2190 * task or not. If it is NULL, then no make_request is active. If
2191 * it is non-NULL, then a make_request is active, and new requests
2192 * should be added at the tail
2193 */
bddd87c7 2194 if (current->bio_list) {
f5fe1b51 2195 bio_list_add(&current->bio_list[0], bio);
dece1635 2196 goto out;
d89d8796 2197 }
27a84d54 2198
d89d8796
NB
2199 /* following loop may be a bit non-obvious, and so deserves some
2200 * explanation.
2201 * Before entering the loop, bio->bi_next is NULL (as all callers
2202 * ensure that) so we have a list with a single bio.
2203 * We pretend that we have just taken it off a longer list, so
bddd87c7
AM
2204 * we assign bio_list to a pointer to the bio_list_on_stack,
2205 * thus initialising the bio_list of new bios to be
27a84d54 2206 * added. ->make_request() may indeed add some more bios
d89d8796
NB
2207 * through a recursive call to generic_make_request. If it
2208 * did, we find a non-NULL value in bio_list and re-enter the loop
2209 * from the top. In this case we really did just take the bio
bddd87c7 2210 * of the top of the list (no pretending) and so remove it from
27a84d54 2211 * bio_list, and call into ->make_request() again.
d89d8796
NB
2212 */
2213 BUG_ON(bio->bi_next);
f5fe1b51
N
2214 bio_list_init(&bio_list_on_stack[0]);
2215 current->bio_list = bio_list_on_stack;
d89d8796 2216 do {
74d46992 2217 struct request_queue *q = bio->bi_disk->queue;
27a84d54 2218
03a07c92 2219 if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) {
79bd9959
N
2220 struct bio_list lower, same;
2221
2222 /* Create a fresh bio_list for all subordinate requests */
f5fe1b51
N
2223 bio_list_on_stack[1] = bio_list_on_stack[0];
2224 bio_list_init(&bio_list_on_stack[0]);
dece1635 2225 ret = q->make_request_fn(q, bio);
3ef28e83
DW
2226
2227 blk_queue_exit(q);
27a84d54 2228
79bd9959
N
2229 /* sort new bios into those for a lower level
2230 * and those for the same level
2231 */
2232 bio_list_init(&lower);
2233 bio_list_init(&same);
f5fe1b51 2234 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
74d46992 2235 if (q == bio->bi_disk->queue)
79bd9959
N
2236 bio_list_add(&same, bio);
2237 else
2238 bio_list_add(&lower, bio);
2239 /* now assemble so we handle the lowest level first */
f5fe1b51
N
2240 bio_list_merge(&bio_list_on_stack[0], &lower);
2241 bio_list_merge(&bio_list_on_stack[0], &same);
2242 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
3ef28e83 2243 } else {
03a07c92
GR
2244 if (unlikely(!blk_queue_dying(q) &&
2245 (bio->bi_opf & REQ_NOWAIT)))
2246 bio_wouldblock_error(bio);
2247 else
2248 bio_io_error(bio);
3ef28e83 2249 }
f5fe1b51 2250 bio = bio_list_pop(&bio_list_on_stack[0]);
d89d8796 2251 } while (bio);
bddd87c7 2252 current->bio_list = NULL; /* deactivate */
dece1635
JA
2253
2254out:
2255 return ret;
d89d8796 2256}
1da177e4
LT
2257EXPORT_SYMBOL(generic_make_request);
2258
f421e1d9
CH
2259/**
2260 * direct_make_request - hand a buffer directly to its device driver for I/O
2261 * @bio: The bio describing the location in memory and on the device.
2262 *
2263 * This function behaves like generic_make_request(), but does not protect
2264 * against recursion. Must only be used if the called driver is known
2265 * to not call generic_make_request (or direct_make_request) again from
2266 * its make_request function. (Calling direct_make_request again from
2267 * a workqueue is perfectly fine as that doesn't recurse).
2268 */
2269blk_qc_t direct_make_request(struct bio *bio)
2270{
2271 struct request_queue *q = bio->bi_disk->queue;
2272 bool nowait = bio->bi_opf & REQ_NOWAIT;
2273 blk_qc_t ret;
2274
2275 if (!generic_make_request_checks(bio))
2276 return BLK_QC_T_NONE;
2277
2278 if (unlikely(blk_queue_enter(q, nowait))) {
2279 if (nowait && !blk_queue_dying(q))
2280 bio->bi_status = BLK_STS_AGAIN;
2281 else
2282 bio->bi_status = BLK_STS_IOERR;
2283 bio_endio(bio);
2284 return BLK_QC_T_NONE;
2285 }
2286
2287 ret = q->make_request_fn(q, bio);
2288 blk_queue_exit(q);
2289 return ret;
2290}
2291EXPORT_SYMBOL_GPL(direct_make_request);
2292
1da177e4 2293/**
710027a4 2294 * submit_bio - submit a bio to the block device layer for I/O
1da177e4
LT
2295 * @bio: The &struct bio which describes the I/O
2296 *
2297 * submit_bio() is very similar in purpose to generic_make_request(), and
2298 * uses that function to do most of the work. Both are fairly rough
710027a4 2299 * interfaces; @bio must be presetup and ready for I/O.
1da177e4
LT
2300 *
2301 */
4e49ea4a 2302blk_qc_t submit_bio(struct bio *bio)
1da177e4 2303{
bf2de6f5
JA
2304 /*
2305 * If it's a regular read/write or a barrier with data attached,
2306 * go through the normal accounting stuff before submission.
2307 */
e2a60da7 2308 if (bio_has_data(bio)) {
4363ac7c
MP
2309 unsigned int count;
2310
95fe6c1a 2311 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
74d46992 2312 count = queue_logical_block_size(bio->bi_disk->queue);
4363ac7c
MP
2313 else
2314 count = bio_sectors(bio);
2315
a8ebb056 2316 if (op_is_write(bio_op(bio))) {
bf2de6f5
JA
2317 count_vm_events(PGPGOUT, count);
2318 } else {
4f024f37 2319 task_io_account_read(bio->bi_iter.bi_size);
bf2de6f5
JA
2320 count_vm_events(PGPGIN, count);
2321 }
2322
2323 if (unlikely(block_dump)) {
2324 char b[BDEVNAME_SIZE];
8dcbdc74 2325 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
ba25f9dc 2326 current->comm, task_pid_nr(current),
a8ebb056 2327 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
4f024f37 2328 (unsigned long long)bio->bi_iter.bi_sector,
74d46992 2329 bio_devname(bio, b), count);
bf2de6f5 2330 }
1da177e4
LT
2331 }
2332
dece1635 2333 return generic_make_request(bio);
1da177e4 2334}
1da177e4
LT
2335EXPORT_SYMBOL(submit_bio);
2336
ea435e1b
CH
2337bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2338{
2339 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2340 return false;
2341
2342 if (current->plug)
2343 blk_flush_plug_list(current->plug, false);
2344 return q->poll_fn(q, cookie);
2345}
2346EXPORT_SYMBOL_GPL(blk_poll);
2347
82124d60 2348/**
bf4e6b4e
HR
2349 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2350 * for new the queue limits
82124d60
KU
2351 * @q: the queue
2352 * @rq: the request being checked
2353 *
2354 * Description:
2355 * @rq may have been made based on weaker limitations of upper-level queues
2356 * in request stacking drivers, and it may violate the limitation of @q.
2357 * Since the block layer and the underlying device driver trust @rq
2358 * after it is inserted to @q, it should be checked against @q before
2359 * the insertion using this generic function.
2360 *
82124d60 2361 * Request stacking drivers like request-based dm may change the queue
bf4e6b4e
HR
2362 * limits when retrying requests on other queues. Those requests need
2363 * to be checked against the new queue limits again during dispatch.
82124d60 2364 */
bf4e6b4e
HR
2365static int blk_cloned_rq_check_limits(struct request_queue *q,
2366 struct request *rq)
82124d60 2367{
8fe0d473 2368 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
82124d60
KU
2369 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2370 return -EIO;
2371 }
2372
2373 /*
2374 * queue's settings related to segment counting like q->bounce_pfn
2375 * may differ from that of other stacking queues.
2376 * Recalculate it to check the request correctly on this queue's
2377 * limitation.
2378 */
2379 blk_recalc_rq_segments(rq);
8a78362c 2380 if (rq->nr_phys_segments > queue_max_segments(q)) {
82124d60
KU
2381 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2382 return -EIO;
2383 }
2384
2385 return 0;
2386}
82124d60
KU
2387
2388/**
2389 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2390 * @q: the queue to submit the request
2391 * @rq: the request being queued
2392 */
2a842aca 2393blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
82124d60
KU
2394{
2395 unsigned long flags;
4853abaa 2396 int where = ELEVATOR_INSERT_BACK;
82124d60 2397
bf4e6b4e 2398 if (blk_cloned_rq_check_limits(q, rq))
2a842aca 2399 return BLK_STS_IOERR;
82124d60 2400
b2c9cd37
AM
2401 if (rq->rq_disk &&
2402 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2a842aca 2403 return BLK_STS_IOERR;
82124d60 2404
7fb4898e
KB
2405 if (q->mq_ops) {
2406 if (blk_queue_io_stat(q))
2407 blk_account_io_start(rq, true);
157f377b
JA
2408 /*
2409 * Since we have a scheduler attached on the top device,
2410 * bypass a potential scheduler on the bottom device for
2411 * insert.
2412 */
b0850297 2413 blk_mq_request_bypass_insert(rq, true);
2a842aca 2414 return BLK_STS_OK;
7fb4898e
KB
2415 }
2416
82124d60 2417 spin_lock_irqsave(q->queue_lock, flags);
3f3299d5 2418 if (unlikely(blk_queue_dying(q))) {
8ba61435 2419 spin_unlock_irqrestore(q->queue_lock, flags);
2a842aca 2420 return BLK_STS_IOERR;
8ba61435 2421 }
82124d60
KU
2422
2423 /*
2424 * Submitting request must be dequeued before calling this function
2425 * because it will be linked to another request_queue
2426 */
2427 BUG_ON(blk_queued_rq(rq));
2428
f73f44eb 2429 if (op_is_flush(rq->cmd_flags))
4853abaa
JM
2430 where = ELEVATOR_INSERT_FLUSH;
2431
2432 add_acct_request(q, rq, where);
e67b77c7
JM
2433 if (where == ELEVATOR_INSERT_FLUSH)
2434 __blk_run_queue(q);
82124d60
KU
2435 spin_unlock_irqrestore(q->queue_lock, flags);
2436
2a842aca 2437 return BLK_STS_OK;
82124d60
KU
2438}
2439EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2440
80a761fd
TH
2441/**
2442 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2443 * @rq: request to examine
2444 *
2445 * Description:
2446 * A request could be merge of IOs which require different failure
2447 * handling. This function determines the number of bytes which
2448 * can be failed from the beginning of the request without
2449 * crossing into area which need to be retried further.
2450 *
2451 * Return:
2452 * The number of bytes to fail.
80a761fd
TH
2453 */
2454unsigned int blk_rq_err_bytes(const struct request *rq)
2455{
2456 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2457 unsigned int bytes = 0;
2458 struct bio *bio;
2459
e8064021 2460 if (!(rq->rq_flags & RQF_MIXED_MERGE))
80a761fd
TH
2461 return blk_rq_bytes(rq);
2462
2463 /*
2464 * Currently the only 'mixing' which can happen is between
2465 * different fastfail types. We can safely fail portions
2466 * which have all the failfast bits that the first one has -
2467 * the ones which are at least as eager to fail as the first
2468 * one.
2469 */
2470 for (bio = rq->bio; bio; bio = bio->bi_next) {
1eff9d32 2471 if ((bio->bi_opf & ff) != ff)
80a761fd 2472 break;
4f024f37 2473 bytes += bio->bi_iter.bi_size;
80a761fd
TH
2474 }
2475
2476 /* this could lead to infinite loop */
2477 BUG_ON(blk_rq_bytes(rq) && !bytes);
2478 return bytes;
2479}
2480EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2481
320ae51f 2482void blk_account_io_completion(struct request *req, unsigned int bytes)
bc58ba94 2483{
c2553b58 2484 if (blk_do_io_stat(req)) {
bc58ba94
JA
2485 const int rw = rq_data_dir(req);
2486 struct hd_struct *part;
2487 int cpu;
2488
2489 cpu = part_stat_lock();
09e099d4 2490 part = req->part;
bc58ba94
JA
2491 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2492 part_stat_unlock();
2493 }
2494}
2495
320ae51f 2496void blk_account_io_done(struct request *req)
bc58ba94 2497{
bc58ba94 2498 /*
dd4c133f
TH
2499 * Account IO completion. flush_rq isn't accounted as a
2500 * normal IO on queueing nor completion. Accounting the
2501 * containing request is enough.
bc58ba94 2502 */
e8064021 2503 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
bc58ba94
JA
2504 unsigned long duration = jiffies - req->start_time;
2505 const int rw = rq_data_dir(req);
2506 struct hd_struct *part;
2507 int cpu;
2508
2509 cpu = part_stat_lock();
09e099d4 2510 part = req->part;
bc58ba94
JA
2511
2512 part_stat_inc(cpu, part, ios[rw]);
2513 part_stat_add(cpu, part, ticks[rw], duration);
d62e26b3
JA
2514 part_round_stats(req->q, cpu, part);
2515 part_dec_in_flight(req->q, part, rw);
bc58ba94 2516
6c23a968 2517 hd_struct_put(part);
bc58ba94
JA
2518 part_stat_unlock();
2519 }
2520}
2521
47fafbc7 2522#ifdef CONFIG_PM
c8158819
LM
2523/*
2524 * Don't process normal requests when queue is suspended
2525 * or in the process of suspending/resuming
2526 */
e4f36b24 2527static bool blk_pm_allow_request(struct request *rq)
c8158819 2528{
e4f36b24
CH
2529 switch (rq->q->rpm_status) {
2530 case RPM_RESUMING:
2531 case RPM_SUSPENDING:
2532 return rq->rq_flags & RQF_PM;
2533 case RPM_SUSPENDED:
2534 return false;
2535 }
2536
2537 return true;
c8158819
LM
2538}
2539#else
e4f36b24 2540static bool blk_pm_allow_request(struct request *rq)
c8158819 2541{
e4f36b24 2542 return true;
c8158819
LM
2543}
2544#endif
2545
320ae51f
JA
2546void blk_account_io_start(struct request *rq, bool new_io)
2547{
2548 struct hd_struct *part;
2549 int rw = rq_data_dir(rq);
2550 int cpu;
2551
2552 if (!blk_do_io_stat(rq))
2553 return;
2554
2555 cpu = part_stat_lock();
2556
2557 if (!new_io) {
2558 part = rq->part;
2559 part_stat_inc(cpu, part, merges[rw]);
2560 } else {
2561 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2562 if (!hd_struct_try_get(part)) {
2563 /*
2564 * The partition is already being removed,
2565 * the request will be accounted on the disk only
2566 *
2567 * We take a reference on disk->part0 although that
2568 * partition will never be deleted, so we can treat
2569 * it as any other partition.
2570 */
2571 part = &rq->rq_disk->part0;
2572 hd_struct_get(part);
2573 }
d62e26b3
JA
2574 part_round_stats(rq->q, cpu, part);
2575 part_inc_in_flight(rq->q, part, rw);
320ae51f
JA
2576 rq->part = part;
2577 }
2578
2579 part_stat_unlock();
2580}
2581
9c988374
CH
2582static struct request *elv_next_request(struct request_queue *q)
2583{
2584 struct request *rq;
2585 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2586
2587 WARN_ON_ONCE(q->mq_ops);
2588
2589 while (1) {
e4f36b24
CH
2590 list_for_each_entry(rq, &q->queue_head, queuelist) {
2591 if (blk_pm_allow_request(rq))
2592 return rq;
2593
2594 if (rq->rq_flags & RQF_SOFTBARRIER)
2595 break;
9c988374
CH
2596 }
2597
2598 /*
2599 * Flush request is running and flush request isn't queueable
2600 * in the drive, we can hold the queue till flush request is
2601 * finished. Even we don't do this, driver can't dispatch next
2602 * requests and will requeue them. And this can improve
2603 * throughput too. For example, we have request flush1, write1,
2604 * flush 2. flush1 is dispatched, then queue is hold, write1
2605 * isn't inserted to queue. After flush1 is finished, flush2
2606 * will be dispatched. Since disk cache is already clean,
2607 * flush2 will be finished very soon, so looks like flush2 is
2608 * folded to flush1.
2609 * Since the queue is hold, a flag is set to indicate the queue
2610 * should be restarted later. Please see flush_end_io() for
2611 * details.
2612 */
2613 if (fq->flush_pending_idx != fq->flush_running_idx &&
2614 !queue_flush_queueable(q)) {
2615 fq->flush_queue_delayed = 1;
2616 return NULL;
2617 }
2618 if (unlikely(blk_queue_bypass(q)) ||
2619 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2620 return NULL;
2621 }
2622}
2623
3bcddeac 2624/**
9934c8c0
TH
2625 * blk_peek_request - peek at the top of a request queue
2626 * @q: request queue to peek at
2627 *
2628 * Description:
2629 * Return the request at the top of @q. The returned request
2630 * should be started using blk_start_request() before LLD starts
2631 * processing it.
2632 *
2633 * Return:
2634 * Pointer to the request at the top of @q if available. Null
2635 * otherwise.
9934c8c0
TH
2636 */
2637struct request *blk_peek_request(struct request_queue *q)
158dbda0
TH
2638{
2639 struct request *rq;
2640 int ret;
2641
2fff8a92 2642 lockdep_assert_held(q->queue_lock);
332ebbf7 2643 WARN_ON_ONCE(q->mq_ops);
2fff8a92 2644
9c988374 2645 while ((rq = elv_next_request(q)) != NULL) {
e8064021 2646 if (!(rq->rq_flags & RQF_STARTED)) {
158dbda0
TH
2647 /*
2648 * This is the first time the device driver
2649 * sees this request (possibly after
2650 * requeueing). Notify IO scheduler.
2651 */
e8064021 2652 if (rq->rq_flags & RQF_SORTED)
158dbda0
TH
2653 elv_activate_rq(q, rq);
2654
2655 /*
2656 * just mark as started even if we don't start
2657 * it, a request that has been delayed should
2658 * not be passed by new incoming requests
2659 */
e8064021 2660 rq->rq_flags |= RQF_STARTED;
158dbda0
TH
2661 trace_block_rq_issue(q, rq);
2662 }
2663
2664 if (!q->boundary_rq || q->boundary_rq == rq) {
2665 q->end_sector = rq_end_sector(rq);
2666 q->boundary_rq = NULL;
2667 }
2668
e8064021 2669 if (rq->rq_flags & RQF_DONTPREP)
158dbda0
TH
2670 break;
2671
2e46e8b2 2672 if (q->dma_drain_size && blk_rq_bytes(rq)) {
158dbda0
TH
2673 /*
2674 * make sure space for the drain appears we
2675 * know we can do this because max_hw_segments
2676 * has been adjusted to be one fewer than the
2677 * device can handle
2678 */
2679 rq->nr_phys_segments++;
2680 }
2681
2682 if (!q->prep_rq_fn)
2683 break;
2684
2685 ret = q->prep_rq_fn(q, rq);
2686 if (ret == BLKPREP_OK) {
2687 break;
2688 } else if (ret == BLKPREP_DEFER) {
2689 /*
2690 * the request may have been (partially) prepped.
2691 * we need to keep this request in the front to
e8064021 2692 * avoid resource deadlock. RQF_STARTED will
158dbda0
TH
2693 * prevent other fs requests from passing this one.
2694 */
2e46e8b2 2695 if (q->dma_drain_size && blk_rq_bytes(rq) &&
e8064021 2696 !(rq->rq_flags & RQF_DONTPREP)) {
158dbda0
TH
2697 /*
2698 * remove the space for the drain we added
2699 * so that we don't add it again
2700 */
2701 --rq->nr_phys_segments;
2702 }
2703
2704 rq = NULL;
2705 break;
0fb5b1fb 2706 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
e8064021 2707 rq->rq_flags |= RQF_QUIET;
c143dc90
JB
2708 /*
2709 * Mark this request as started so we don't trigger
2710 * any debug logic in the end I/O path.
2711 */
2712 blk_start_request(rq);
2a842aca
CH
2713 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2714 BLK_STS_TARGET : BLK_STS_IOERR);
158dbda0
TH
2715 } else {
2716 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2717 break;
2718 }
2719 }
2720
2721 return rq;
2722}
9934c8c0 2723EXPORT_SYMBOL(blk_peek_request);
158dbda0 2724
5034435c 2725static void blk_dequeue_request(struct request *rq)
158dbda0 2726{
9934c8c0
TH
2727 struct request_queue *q = rq->q;
2728
158dbda0
TH
2729 BUG_ON(list_empty(&rq->queuelist));
2730 BUG_ON(ELV_ON_HASH(rq));
2731
2732 list_del_init(&rq->queuelist);
2733
2734 /*
2735 * the time frame between a request being removed from the lists
2736 * and to it is freed is accounted as io that is in progress at
2737 * the driver side.
2738 */
9195291e 2739 if (blk_account_rq(rq)) {
0a7ae2ff 2740 q->in_flight[rq_is_sync(rq)]++;
9195291e
DS
2741 set_io_start_time_ns(rq);
2742 }
158dbda0
TH
2743}
2744
9934c8c0
TH
2745/**
2746 * blk_start_request - start request processing on the driver
2747 * @req: request to dequeue
2748 *
2749 * Description:
2750 * Dequeue @req and start timeout timer on it. This hands off the
2751 * request to the driver.
9934c8c0
TH
2752 */
2753void blk_start_request(struct request *req)
2754{
2fff8a92 2755 lockdep_assert_held(req->q->queue_lock);
332ebbf7 2756 WARN_ON_ONCE(req->q->mq_ops);
2fff8a92 2757
9934c8c0
TH
2758 blk_dequeue_request(req);
2759
cf43e6be 2760 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
88eeca49 2761 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
cf43e6be 2762 req->rq_flags |= RQF_STATS;
87760e5e 2763 wbt_issue(req->q->rq_wb, &req->issue_stat);
cf43e6be
JA
2764 }
2765
4912aa6c 2766 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
9934c8c0
TH
2767 blk_add_timer(req);
2768}
2769EXPORT_SYMBOL(blk_start_request);
2770
2771/**
2772 * blk_fetch_request - fetch a request from a request queue
2773 * @q: request queue to fetch a request from
2774 *
2775 * Description:
2776 * Return the request at the top of @q. The request is started on
2777 * return and LLD can start processing it immediately.
2778 *
2779 * Return:
2780 * Pointer to the request at the top of @q if available. Null
2781 * otherwise.
9934c8c0
TH
2782 */
2783struct request *blk_fetch_request(struct request_queue *q)
2784{
2785 struct request *rq;
2786
2fff8a92 2787 lockdep_assert_held(q->queue_lock);
332ebbf7 2788 WARN_ON_ONCE(q->mq_ops);
2fff8a92 2789
9934c8c0
TH
2790 rq = blk_peek_request(q);
2791 if (rq)
2792 blk_start_request(rq);
2793 return rq;
2794}
2795EXPORT_SYMBOL(blk_fetch_request);
2796
ef71de8b
CH
2797/*
2798 * Steal bios from a request and add them to a bio list.
2799 * The request must not have been partially completed before.
2800 */
2801void blk_steal_bios(struct bio_list *list, struct request *rq)
2802{
2803 if (rq->bio) {
2804 if (list->tail)
2805 list->tail->bi_next = rq->bio;
2806 else
2807 list->head = rq->bio;
2808 list->tail = rq->biotail;
2809
2810 rq->bio = NULL;
2811 rq->biotail = NULL;
2812 }
2813
2814 rq->__data_len = 0;
2815}
2816EXPORT_SYMBOL_GPL(blk_steal_bios);
2817
3bcddeac 2818/**
2e60e022 2819 * blk_update_request - Special helper function for request stacking drivers
8ebf9756 2820 * @req: the request being processed
2a842aca 2821 * @error: block status code
8ebf9756 2822 * @nr_bytes: number of bytes to complete @req
3bcddeac
KU
2823 *
2824 * Description:
8ebf9756
RD
2825 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2826 * the request structure even if @req doesn't have leftover.
2827 * If @req has leftover, sets it up for the next range of segments.
2e60e022
TH
2828 *
2829 * This special helper function is only for request stacking drivers
2830 * (e.g. request-based dm) so that they can handle partial completion.
2831 * Actual device drivers should use blk_end_request instead.
2832 *
2833 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2834 * %false return from this function.
3bcddeac
KU
2835 *
2836 * Return:
2e60e022
TH
2837 * %false - this request doesn't have any more data
2838 * %true - this request has more data
3bcddeac 2839 **/
2a842aca
CH
2840bool blk_update_request(struct request *req, blk_status_t error,
2841 unsigned int nr_bytes)
1da177e4 2842{
f79ea416 2843 int total_bytes;
1da177e4 2844
2a842aca 2845 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
4a0efdc9 2846
2e60e022
TH
2847 if (!req->bio)
2848 return false;
2849
2a842aca
CH
2850 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2851 !(req->rq_flags & RQF_QUIET)))
2852 print_req_error(req, error);
1da177e4 2853
bc58ba94 2854 blk_account_io_completion(req, nr_bytes);
d72d904a 2855
f79ea416
KO
2856 total_bytes = 0;
2857 while (req->bio) {
2858 struct bio *bio = req->bio;
4f024f37 2859 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1da177e4 2860
4f024f37 2861 if (bio_bytes == bio->bi_iter.bi_size)
1da177e4 2862 req->bio = bio->bi_next;
1da177e4 2863
fbbaf700
N
2864 /* Completion has already been traced */
2865 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
f79ea416 2866 req_bio_endio(req, bio, bio_bytes, error);
1da177e4 2867
f79ea416
KO
2868 total_bytes += bio_bytes;
2869 nr_bytes -= bio_bytes;
1da177e4 2870
f79ea416
KO
2871 if (!nr_bytes)
2872 break;
1da177e4
LT
2873 }
2874
2875 /*
2876 * completely done
2877 */
2e60e022
TH
2878 if (!req->bio) {
2879 /*
2880 * Reset counters so that the request stacking driver
2881 * can find how many bytes remain in the request
2882 * later.
2883 */
a2dec7b3 2884 req->__data_len = 0;
2e60e022
TH
2885 return false;
2886 }
1da177e4 2887
a2dec7b3 2888 req->__data_len -= total_bytes;
2e46e8b2
TH
2889
2890 /* update sector only for requests with clear definition of sector */
57292b58 2891 if (!blk_rq_is_passthrough(req))
a2dec7b3 2892 req->__sector += total_bytes >> 9;
2e46e8b2 2893
80a761fd 2894 /* mixed attributes always follow the first bio */
e8064021 2895 if (req->rq_flags & RQF_MIXED_MERGE) {
80a761fd 2896 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1eff9d32 2897 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
80a761fd
TH
2898 }
2899
ed6565e7
CH
2900 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2901 /*
2902 * If total number of sectors is less than the first segment
2903 * size, something has gone terribly wrong.
2904 */
2905 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2906 blk_dump_rq_flags(req, "request botched");
2907 req->__data_len = blk_rq_cur_bytes(req);
2908 }
2e46e8b2 2909
ed6565e7
CH
2910 /* recalculate the number of segments */
2911 blk_recalc_rq_segments(req);
2912 }
2e46e8b2 2913
2e60e022 2914 return true;
1da177e4 2915}
2e60e022 2916EXPORT_SYMBOL_GPL(blk_update_request);
1da177e4 2917
2a842aca 2918static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2e60e022
TH
2919 unsigned int nr_bytes,
2920 unsigned int bidi_bytes)
5efccd17 2921{
2e60e022
TH
2922 if (blk_update_request(rq, error, nr_bytes))
2923 return true;
5efccd17 2924
2e60e022
TH
2925 /* Bidi request must be completed as a whole */
2926 if (unlikely(blk_bidi_rq(rq)) &&
2927 blk_update_request(rq->next_rq, error, bidi_bytes))
2928 return true;
5efccd17 2929
e2e1a148
JA
2930 if (blk_queue_add_random(rq->q))
2931 add_disk_randomness(rq->rq_disk);
2e60e022
TH
2932
2933 return false;
1da177e4
LT
2934}
2935
28018c24
JB
2936/**
2937 * blk_unprep_request - unprepare a request
2938 * @req: the request
2939 *
2940 * This function makes a request ready for complete resubmission (or
2941 * completion). It happens only after all error handling is complete,
2942 * so represents the appropriate moment to deallocate any resources
2943 * that were allocated to the request in the prep_rq_fn. The queue
2944 * lock is held when calling this.
2945 */
2946void blk_unprep_request(struct request *req)
2947{
2948 struct request_queue *q = req->q;
2949
e8064021 2950 req->rq_flags &= ~RQF_DONTPREP;
28018c24
JB
2951 if (q->unprep_rq_fn)
2952 q->unprep_rq_fn(q, req);
2953}
2954EXPORT_SYMBOL_GPL(blk_unprep_request);
2955
2a842aca 2956void blk_finish_request(struct request *req, blk_status_t error)
1da177e4 2957{
cf43e6be
JA
2958 struct request_queue *q = req->q;
2959
2fff8a92 2960 lockdep_assert_held(req->q->queue_lock);
332ebbf7 2961 WARN_ON_ONCE(q->mq_ops);
2fff8a92 2962
cf43e6be 2963 if (req->rq_flags & RQF_STATS)
34dbad5d 2964 blk_stat_add(req);
cf43e6be 2965
e8064021 2966 if (req->rq_flags & RQF_QUEUED)
cf43e6be 2967 blk_queue_end_tag(q, req);
b8286239 2968
ba396a6c 2969 BUG_ON(blk_queued_rq(req));
1da177e4 2970
57292b58 2971 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
dc3b17cc 2972 laptop_io_completion(req->q->backing_dev_info);
1da177e4 2973
e78042e5
MA
2974 blk_delete_timer(req);
2975
e8064021 2976 if (req->rq_flags & RQF_DONTPREP)
28018c24
JB
2977 blk_unprep_request(req);
2978
bc58ba94 2979 blk_account_io_done(req);
b8286239 2980
87760e5e
JA
2981 if (req->end_io) {
2982 wbt_done(req->q->rq_wb, &req->issue_stat);
8ffdc655 2983 req->end_io(req, error);
87760e5e 2984 } else {
b8286239
KU
2985 if (blk_bidi_rq(req))
2986 __blk_put_request(req->next_rq->q, req->next_rq);
2987
cf43e6be 2988 __blk_put_request(q, req);
b8286239 2989 }
1da177e4 2990}
12120077 2991EXPORT_SYMBOL(blk_finish_request);
1da177e4 2992
3b11313a 2993/**
2e60e022
TH
2994 * blk_end_bidi_request - Complete a bidi request
2995 * @rq: the request to complete
2a842aca 2996 * @error: block status code
2e60e022
TH
2997 * @nr_bytes: number of bytes to complete @rq
2998 * @bidi_bytes: number of bytes to complete @rq->next_rq
a0cd1285
JA
2999 *
3000 * Description:
e3a04fe3 3001 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2e60e022
TH
3002 * Drivers that supports bidi can safely call this member for any
3003 * type of request, bidi or uni. In the later case @bidi_bytes is
3004 * just ignored.
336cdb40
KU
3005 *
3006 * Return:
2e60e022
TH
3007 * %false - we are done with this request
3008 * %true - still buffers pending for this request
a0cd1285 3009 **/
2a842aca 3010static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
32fab448
KU
3011 unsigned int nr_bytes, unsigned int bidi_bytes)
3012{
336cdb40 3013 struct request_queue *q = rq->q;
2e60e022 3014 unsigned long flags;
32fab448 3015
332ebbf7
BVA
3016 WARN_ON_ONCE(q->mq_ops);
3017
2e60e022
TH
3018 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3019 return true;
32fab448 3020
336cdb40 3021 spin_lock_irqsave(q->queue_lock, flags);
2e60e022 3022 blk_finish_request(rq, error);
336cdb40
KU
3023 spin_unlock_irqrestore(q->queue_lock, flags);
3024
2e60e022 3025 return false;
32fab448
KU
3026}
3027
336cdb40 3028/**
2e60e022
TH
3029 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3030 * @rq: the request to complete
2a842aca 3031 * @error: block status code
e3a04fe3
KU
3032 * @nr_bytes: number of bytes to complete @rq
3033 * @bidi_bytes: number of bytes to complete @rq->next_rq
336cdb40
KU
3034 *
3035 * Description:
2e60e022
TH
3036 * Identical to blk_end_bidi_request() except that queue lock is
3037 * assumed to be locked on entry and remains so on return.
336cdb40
KU
3038 *
3039 * Return:
2e60e022
TH
3040 * %false - we are done with this request
3041 * %true - still buffers pending for this request
336cdb40 3042 **/
2a842aca 3043static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
b1f74493 3044 unsigned int nr_bytes, unsigned int bidi_bytes)
336cdb40 3045{
2fff8a92 3046 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3047 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3048
2e60e022
TH
3049 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3050 return true;
336cdb40 3051
2e60e022 3052 blk_finish_request(rq, error);
336cdb40 3053
2e60e022 3054 return false;
336cdb40 3055}
e19a3ab0
KU
3056
3057/**
3058 * blk_end_request - Helper function for drivers to complete the request.
3059 * @rq: the request being processed
2a842aca 3060 * @error: block status code
e19a3ab0
KU
3061 * @nr_bytes: number of bytes to complete
3062 *
3063 * Description:
3064 * Ends I/O on a number of bytes attached to @rq.
3065 * If @rq has leftover, sets it up for the next range of segments.
3066 *
3067 * Return:
b1f74493
FT
3068 * %false - we are done with this request
3069 * %true - still buffers pending for this request
e19a3ab0 3070 **/
2a842aca
CH
3071bool blk_end_request(struct request *rq, blk_status_t error,
3072 unsigned int nr_bytes)
e19a3ab0 3073{
332ebbf7 3074 WARN_ON_ONCE(rq->q->mq_ops);
b1f74493 3075 return blk_end_bidi_request(rq, error, nr_bytes, 0);
e19a3ab0 3076}
56ad1740 3077EXPORT_SYMBOL(blk_end_request);
336cdb40
KU
3078
3079/**
b1f74493
FT
3080 * blk_end_request_all - Helper function for drives to finish the request.
3081 * @rq: the request to finish
2a842aca 3082 * @error: block status code
336cdb40
KU
3083 *
3084 * Description:
b1f74493
FT
3085 * Completely finish @rq.
3086 */
2a842aca 3087void blk_end_request_all(struct request *rq, blk_status_t error)
336cdb40 3088{
b1f74493
FT
3089 bool pending;
3090 unsigned int bidi_bytes = 0;
336cdb40 3091
b1f74493
FT
3092 if (unlikely(blk_bidi_rq(rq)))
3093 bidi_bytes = blk_rq_bytes(rq->next_rq);
336cdb40 3094
b1f74493
FT
3095 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3096 BUG_ON(pending);
3097}
56ad1740 3098EXPORT_SYMBOL(blk_end_request_all);
336cdb40 3099
e3a04fe3 3100/**
b1f74493
FT
3101 * __blk_end_request - Helper function for drivers to complete the request.
3102 * @rq: the request being processed
2a842aca 3103 * @error: block status code
b1f74493 3104 * @nr_bytes: number of bytes to complete
e3a04fe3
KU
3105 *
3106 * Description:
b1f74493 3107 * Must be called with queue lock held unlike blk_end_request().
e3a04fe3
KU
3108 *
3109 * Return:
b1f74493
FT
3110 * %false - we are done with this request
3111 * %true - still buffers pending for this request
e3a04fe3 3112 **/
2a842aca
CH
3113bool __blk_end_request(struct request *rq, blk_status_t error,
3114 unsigned int nr_bytes)
e3a04fe3 3115{
2fff8a92 3116 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3117 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3118
b1f74493 3119 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
e3a04fe3 3120}
56ad1740 3121EXPORT_SYMBOL(__blk_end_request);
e3a04fe3 3122
32fab448 3123/**
b1f74493
FT
3124 * __blk_end_request_all - Helper function for drives to finish the request.
3125 * @rq: the request to finish
2a842aca 3126 * @error: block status code
32fab448
KU
3127 *
3128 * Description:
b1f74493 3129 * Completely finish @rq. Must be called with queue lock held.
32fab448 3130 */
2a842aca 3131void __blk_end_request_all(struct request *rq, blk_status_t error)
32fab448 3132{
b1f74493
FT
3133 bool pending;
3134 unsigned int bidi_bytes = 0;
3135
2fff8a92 3136 lockdep_assert_held(rq->q->queue_lock);
332ebbf7 3137 WARN_ON_ONCE(rq->q->mq_ops);
2fff8a92 3138
b1f74493
FT
3139 if (unlikely(blk_bidi_rq(rq)))
3140 bidi_bytes = blk_rq_bytes(rq->next_rq);
3141
3142 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3143 BUG_ON(pending);
32fab448 3144}
56ad1740 3145EXPORT_SYMBOL(__blk_end_request_all);
32fab448 3146
e19a3ab0 3147/**
b1f74493
FT
3148 * __blk_end_request_cur - Helper function to finish the current request chunk.
3149 * @rq: the request to finish the current chunk for
2a842aca 3150 * @error: block status code
e19a3ab0
KU
3151 *
3152 * Description:
b1f74493
FT
3153 * Complete the current consecutively mapped chunk from @rq. Must
3154 * be called with queue lock held.
e19a3ab0
KU
3155 *
3156 * Return:
b1f74493
FT
3157 * %false - we are done with this request
3158 * %true - still buffers pending for this request
3159 */
2a842aca 3160bool __blk_end_request_cur(struct request *rq, blk_status_t error)
e19a3ab0 3161{
b1f74493 3162 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
e19a3ab0 3163}
56ad1740 3164EXPORT_SYMBOL(__blk_end_request_cur);
e19a3ab0 3165
86db1e29
JA
3166void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3167 struct bio *bio)
1da177e4 3168{
b4f42e28 3169 if (bio_has_data(bio))
fb2dce86 3170 rq->nr_phys_segments = bio_phys_segments(q, bio);
b4f42e28 3171
4f024f37 3172 rq->__data_len = bio->bi_iter.bi_size;
1da177e4 3173 rq->bio = rq->biotail = bio;
1da177e4 3174
74d46992
CH
3175 if (bio->bi_disk)
3176 rq->rq_disk = bio->bi_disk;
66846572 3177}
1da177e4 3178
2d4dc890
IL
3179#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3180/**
3181 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3182 * @rq: the request to be flushed
3183 *
3184 * Description:
3185 * Flush all pages in @rq.
3186 */
3187void rq_flush_dcache_pages(struct request *rq)
3188{
3189 struct req_iterator iter;
7988613b 3190 struct bio_vec bvec;
2d4dc890
IL
3191
3192 rq_for_each_segment(bvec, rq, iter)
7988613b 3193 flush_dcache_page(bvec.bv_page);
2d4dc890
IL
3194}
3195EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3196#endif
3197
ef9e3fac
KU
3198/**
3199 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3200 * @q : the queue of the device being checked
3201 *
3202 * Description:
3203 * Check if underlying low-level drivers of a device are busy.
3204 * If the drivers want to export their busy state, they must set own
3205 * exporting function using blk_queue_lld_busy() first.
3206 *
3207 * Basically, this function is used only by request stacking drivers
3208 * to stop dispatching requests to underlying devices when underlying
3209 * devices are busy. This behavior helps more I/O merging on the queue
3210 * of the request stacking driver and prevents I/O throughput regression
3211 * on burst I/O load.
3212 *
3213 * Return:
3214 * 0 - Not busy (The request stacking driver should dispatch request)
3215 * 1 - Busy (The request stacking driver should stop dispatching request)
3216 */
3217int blk_lld_busy(struct request_queue *q)
3218{
3219 if (q->lld_busy_fn)
3220 return q->lld_busy_fn(q);
3221
3222 return 0;
3223}
3224EXPORT_SYMBOL_GPL(blk_lld_busy);
3225
78d8e58a
MS
3226/**
3227 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3228 * @rq: the clone request to be cleaned up
3229 *
3230 * Description:
3231 * Free all bios in @rq for a cloned request.
3232 */
3233void blk_rq_unprep_clone(struct request *rq)
3234{
3235 struct bio *bio;
3236
3237 while ((bio = rq->bio) != NULL) {
3238 rq->bio = bio->bi_next;
3239
3240 bio_put(bio);
3241 }
3242}
3243EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3244
3245/*
3246 * Copy attributes of the original request to the clone request.
3247 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3248 */
3249static void __blk_rq_prep_clone(struct request *dst, struct request *src)
b0fd271d
KU
3250{
3251 dst->cpu = src->cpu;
b0fd271d
KU
3252 dst->__sector = blk_rq_pos(src);
3253 dst->__data_len = blk_rq_bytes(src);
3254 dst->nr_phys_segments = src->nr_phys_segments;
3255 dst->ioprio = src->ioprio;
3256 dst->extra_len = src->extra_len;
78d8e58a
MS
3257}
3258
3259/**
3260 * blk_rq_prep_clone - Helper function to setup clone request
3261 * @rq: the request to be setup
3262 * @rq_src: original request to be cloned
3263 * @bs: bio_set that bios for clone are allocated from
3264 * @gfp_mask: memory allocation mask for bio
3265 * @bio_ctr: setup function to be called for each clone bio.
3266 * Returns %0 for success, non %0 for failure.
3267 * @data: private data to be passed to @bio_ctr
3268 *
3269 * Description:
3270 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3271 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3272 * are not copied, and copying such parts is the caller's responsibility.
3273 * Also, pages which the original bios are pointing to are not copied
3274 * and the cloned bios just point same pages.
3275 * So cloned bios must be completed before original bios, which means
3276 * the caller must complete @rq before @rq_src.
3277 */
3278int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3279 struct bio_set *bs, gfp_t gfp_mask,
3280 int (*bio_ctr)(struct bio *, struct bio *, void *),
3281 void *data)
3282{
3283 struct bio *bio, *bio_src;
3284
3285 if (!bs)
3286 bs = fs_bio_set;
3287
3288 __rq_for_each_bio(bio_src, rq_src) {
3289 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3290 if (!bio)
3291 goto free_and_out;
3292
3293 if (bio_ctr && bio_ctr(bio, bio_src, data))
3294 goto free_and_out;
3295
3296 if (rq->bio) {
3297 rq->biotail->bi_next = bio;
3298 rq->biotail = bio;
3299 } else
3300 rq->bio = rq->biotail = bio;
3301 }
3302
3303 __blk_rq_prep_clone(rq, rq_src);
3304
3305 return 0;
3306
3307free_and_out:
3308 if (bio)
3309 bio_put(bio);
3310 blk_rq_unprep_clone(rq);
3311
3312 return -ENOMEM;
b0fd271d
KU
3313}
3314EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3315
59c3d45e 3316int kblockd_schedule_work(struct work_struct *work)
1da177e4
LT
3317{
3318 return queue_work(kblockd_workqueue, work);
3319}
1da177e4
LT
3320EXPORT_SYMBOL(kblockd_schedule_work);
3321
ee63cfa7
JA
3322int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3323{
3324 return queue_work_on(cpu, kblockd_workqueue, work);
3325}
3326EXPORT_SYMBOL(kblockd_schedule_work_on);
3327
818cd1cb
JA
3328int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3329 unsigned long delay)
3330{
3331 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3332}
3333EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3334
59c3d45e
JA
3335int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3336 unsigned long delay)
e43473b7
VG
3337{
3338 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3339}
3340EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3341
8ab14595
JA
3342int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3343 unsigned long delay)
3344{
3345 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3346}
3347EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3348
75df7136
SJ
3349/**
3350 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3351 * @plug: The &struct blk_plug that needs to be initialized
3352 *
3353 * Description:
3354 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3355 * pending I/O should the task end up blocking between blk_start_plug() and
3356 * blk_finish_plug(). This is important from a performance perspective, but
3357 * also ensures that we don't deadlock. For instance, if the task is blocking
3358 * for a memory allocation, memory reclaim could end up wanting to free a
3359 * page belonging to that request that is currently residing in our private
3360 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3361 * this kind of deadlock.
3362 */
73c10101
JA
3363void blk_start_plug(struct blk_plug *plug)
3364{
3365 struct task_struct *tsk = current;
3366
dd6cf3e1
SL
3367 /*
3368 * If this is a nested plug, don't actually assign it.
3369 */
3370 if (tsk->plug)
3371 return;
3372
73c10101 3373 INIT_LIST_HEAD(&plug->list);
320ae51f 3374 INIT_LIST_HEAD(&plug->mq_list);
048c9374 3375 INIT_LIST_HEAD(&plug->cb_list);
73c10101 3376 /*
dd6cf3e1
SL
3377 * Store ordering should not be needed here, since a potential
3378 * preempt will imply a full memory barrier
73c10101 3379 */
dd6cf3e1 3380 tsk->plug = plug;
73c10101
JA
3381}
3382EXPORT_SYMBOL(blk_start_plug);
3383
3384static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3385{
3386 struct request *rqa = container_of(a, struct request, queuelist);
3387 struct request *rqb = container_of(b, struct request, queuelist);
3388
975927b9
JM
3389 return !(rqa->q < rqb->q ||
3390 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
73c10101
JA
3391}
3392
49cac01e
JA
3393/*
3394 * If 'from_schedule' is true, then postpone the dispatch of requests
3395 * until a safe kblockd context. We due this to avoid accidental big
3396 * additional stack usage in driver dispatch, in places where the originally
3397 * plugger did not intend it.
3398 */
f6603783 3399static void queue_unplugged(struct request_queue *q, unsigned int depth,
49cac01e 3400 bool from_schedule)
99e22598 3401 __releases(q->queue_lock)
94b5eb28 3402{
2fff8a92
BVA
3403 lockdep_assert_held(q->queue_lock);
3404
49cac01e 3405 trace_block_unplug(q, depth, !from_schedule);
99e22598 3406
70460571 3407 if (from_schedule)
24ecfbe2 3408 blk_run_queue_async(q);
70460571 3409 else
24ecfbe2 3410 __blk_run_queue(q);
70460571 3411 spin_unlock(q->queue_lock);
94b5eb28
JA
3412}
3413
74018dc3 3414static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
048c9374
N
3415{
3416 LIST_HEAD(callbacks);
3417
2a7d5559
SL
3418 while (!list_empty(&plug->cb_list)) {
3419 list_splice_init(&plug->cb_list, &callbacks);
048c9374 3420
2a7d5559
SL
3421 while (!list_empty(&callbacks)) {
3422 struct blk_plug_cb *cb = list_first_entry(&callbacks,
048c9374
N
3423 struct blk_plug_cb,
3424 list);
2a7d5559 3425 list_del(&cb->list);
74018dc3 3426 cb->callback(cb, from_schedule);
2a7d5559 3427 }
048c9374
N
3428 }
3429}
3430
9cbb1750
N
3431struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3432 int size)
3433{
3434 struct blk_plug *plug = current->plug;
3435 struct blk_plug_cb *cb;
3436
3437 if (!plug)
3438 return NULL;
3439
3440 list_for_each_entry(cb, &plug->cb_list, list)
3441 if (cb->callback == unplug && cb->data == data)
3442 return cb;
3443
3444 /* Not currently on the callback list */
3445 BUG_ON(size < sizeof(*cb));
3446 cb = kzalloc(size, GFP_ATOMIC);
3447 if (cb) {
3448 cb->data = data;
3449 cb->callback = unplug;
3450 list_add(&cb->list, &plug->cb_list);
3451 }
3452 return cb;
3453}
3454EXPORT_SYMBOL(blk_check_plugged);
3455
49cac01e 3456void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
73c10101
JA
3457{
3458 struct request_queue *q;
3459 unsigned long flags;
3460 struct request *rq;
109b8129 3461 LIST_HEAD(list);
94b5eb28 3462 unsigned int depth;
73c10101 3463
74018dc3 3464 flush_plug_callbacks(plug, from_schedule);
320ae51f
JA
3465
3466 if (!list_empty(&plug->mq_list))
3467 blk_mq_flush_plug_list(plug, from_schedule);
3468
73c10101
JA
3469 if (list_empty(&plug->list))
3470 return;
3471
109b8129
N
3472 list_splice_init(&plug->list, &list);
3473
422765c2 3474 list_sort(NULL, &list, plug_rq_cmp);
73c10101
JA
3475
3476 q = NULL;
94b5eb28 3477 depth = 0;
18811272
JA
3478
3479 /*
3480 * Save and disable interrupts here, to avoid doing it for every
3481 * queue lock we have to take.
3482 */
73c10101 3483 local_irq_save(flags);
109b8129
N
3484 while (!list_empty(&list)) {
3485 rq = list_entry_rq(list.next);
73c10101 3486 list_del_init(&rq->queuelist);
73c10101
JA
3487 BUG_ON(!rq->q);
3488 if (rq->q != q) {
99e22598
JA
3489 /*
3490 * This drops the queue lock
3491 */
3492 if (q)
49cac01e 3493 queue_unplugged(q, depth, from_schedule);
73c10101 3494 q = rq->q;
94b5eb28 3495 depth = 0;
73c10101
JA
3496 spin_lock(q->queue_lock);
3497 }
8ba61435
TH
3498
3499 /*
3500 * Short-circuit if @q is dead
3501 */
3f3299d5 3502 if (unlikely(blk_queue_dying(q))) {
2a842aca 3503 __blk_end_request_all(rq, BLK_STS_IOERR);
8ba61435
TH
3504 continue;
3505 }
3506
73c10101
JA
3507 /*
3508 * rq is already accounted, so use raw insert
3509 */
f73f44eb 3510 if (op_is_flush(rq->cmd_flags))
401a18e9
JA
3511 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3512 else
3513 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
94b5eb28
JA
3514
3515 depth++;
73c10101
JA
3516 }
3517
99e22598
JA
3518 /*
3519 * This drops the queue lock
3520 */
3521 if (q)
49cac01e 3522 queue_unplugged(q, depth, from_schedule);
73c10101 3523
73c10101
JA
3524 local_irq_restore(flags);
3525}
73c10101
JA
3526
3527void blk_finish_plug(struct blk_plug *plug)
3528{
dd6cf3e1
SL
3529 if (plug != current->plug)
3530 return;
f6603783 3531 blk_flush_plug_list(plug, false);
73c10101 3532
dd6cf3e1 3533 current->plug = NULL;
73c10101 3534}
88b996cd 3535EXPORT_SYMBOL(blk_finish_plug);
73c10101 3536
47fafbc7 3537#ifdef CONFIG_PM
6c954667
LM
3538/**
3539 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3540 * @q: the queue of the device
3541 * @dev: the device the queue belongs to
3542 *
3543 * Description:
3544 * Initialize runtime-PM-related fields for @q and start auto suspend for
3545 * @dev. Drivers that want to take advantage of request-based runtime PM
3546 * should call this function after @dev has been initialized, and its
3547 * request queue @q has been allocated, and runtime PM for it can not happen
3548 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3549 * cases, driver should call this function before any I/O has taken place.
3550 *
3551 * This function takes care of setting up using auto suspend for the device,
3552 * the autosuspend delay is set to -1 to make runtime suspend impossible
3553 * until an updated value is either set by user or by driver. Drivers do
3554 * not need to touch other autosuspend settings.
3555 *
3556 * The block layer runtime PM is request based, so only works for drivers
3557 * that use request as their IO unit instead of those directly use bio's.
3558 */
3559void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3560{
765e40b6
CH
3561 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3562 if (q->mq_ops)
3563 return;
3564
6c954667
LM
3565 q->dev = dev;
3566 q->rpm_status = RPM_ACTIVE;
3567 pm_runtime_set_autosuspend_delay(q->dev, -1);
3568 pm_runtime_use_autosuspend(q->dev);
3569}
3570EXPORT_SYMBOL(blk_pm_runtime_init);
3571
3572/**
3573 * blk_pre_runtime_suspend - Pre runtime suspend check
3574 * @q: the queue of the device
3575 *
3576 * Description:
3577 * This function will check if runtime suspend is allowed for the device
3578 * by examining if there are any requests pending in the queue. If there
3579 * are requests pending, the device can not be runtime suspended; otherwise,
3580 * the queue's status will be updated to SUSPENDING and the driver can
3581 * proceed to suspend the device.
3582 *
3583 * For the not allowed case, we mark last busy for the device so that
3584 * runtime PM core will try to autosuspend it some time later.
3585 *
3586 * This function should be called near the start of the device's
3587 * runtime_suspend callback.
3588 *
3589 * Return:
3590 * 0 - OK to runtime suspend the device
3591 * -EBUSY - Device should not be runtime suspended
3592 */
3593int blk_pre_runtime_suspend(struct request_queue *q)
3594{
3595 int ret = 0;
3596
4fd41a85
KX
3597 if (!q->dev)
3598 return ret;
3599
6c954667
LM
3600 spin_lock_irq(q->queue_lock);
3601 if (q->nr_pending) {
3602 ret = -EBUSY;
3603 pm_runtime_mark_last_busy(q->dev);
3604 } else {
3605 q->rpm_status = RPM_SUSPENDING;
3606 }
3607 spin_unlock_irq(q->queue_lock);
3608 return ret;
3609}
3610EXPORT_SYMBOL(blk_pre_runtime_suspend);
3611
3612/**
3613 * blk_post_runtime_suspend - Post runtime suspend processing
3614 * @q: the queue of the device
3615 * @err: return value of the device's runtime_suspend function
3616 *
3617 * Description:
3618 * Update the queue's runtime status according to the return value of the
3619 * device's runtime suspend function and mark last busy for the device so
3620 * that PM core will try to auto suspend the device at a later time.
3621 *
3622 * This function should be called near the end of the device's
3623 * runtime_suspend callback.
3624 */
3625void blk_post_runtime_suspend(struct request_queue *q, int err)
3626{
4fd41a85
KX
3627 if (!q->dev)
3628 return;
3629
6c954667
LM
3630 spin_lock_irq(q->queue_lock);
3631 if (!err) {
3632 q->rpm_status = RPM_SUSPENDED;
3633 } else {
3634 q->rpm_status = RPM_ACTIVE;
3635 pm_runtime_mark_last_busy(q->dev);
3636 }
3637 spin_unlock_irq(q->queue_lock);
3638}
3639EXPORT_SYMBOL(blk_post_runtime_suspend);
3640
3641/**
3642 * blk_pre_runtime_resume - Pre runtime resume processing
3643 * @q: the queue of the device
3644 *
3645 * Description:
3646 * Update the queue's runtime status to RESUMING in preparation for the
3647 * runtime resume of the device.
3648 *
3649 * This function should be called near the start of the device's
3650 * runtime_resume callback.
3651 */
3652void blk_pre_runtime_resume(struct request_queue *q)
3653{
4fd41a85
KX
3654 if (!q->dev)
3655 return;
3656
6c954667
LM
3657 spin_lock_irq(q->queue_lock);
3658 q->rpm_status = RPM_RESUMING;
3659 spin_unlock_irq(q->queue_lock);
3660}
3661EXPORT_SYMBOL(blk_pre_runtime_resume);
3662
3663/**
3664 * blk_post_runtime_resume - Post runtime resume processing
3665 * @q: the queue of the device
3666 * @err: return value of the device's runtime_resume function
3667 *
3668 * Description:
3669 * Update the queue's runtime status according to the return value of the
3670 * device's runtime_resume function. If it is successfully resumed, process
3671 * the requests that are queued into the device's queue when it is resuming
3672 * and then mark last busy and initiate autosuspend for it.
3673 *
3674 * This function should be called near the end of the device's
3675 * runtime_resume callback.
3676 */
3677void blk_post_runtime_resume(struct request_queue *q, int err)
3678{
4fd41a85
KX
3679 if (!q->dev)
3680 return;
3681
6c954667
LM
3682 spin_lock_irq(q->queue_lock);
3683 if (!err) {
3684 q->rpm_status = RPM_ACTIVE;
3685 __blk_run_queue(q);
3686 pm_runtime_mark_last_busy(q->dev);
c60855cd 3687 pm_request_autosuspend(q->dev);
6c954667
LM
3688 } else {
3689 q->rpm_status = RPM_SUSPENDED;
3690 }
3691 spin_unlock_irq(q->queue_lock);
3692}
3693EXPORT_SYMBOL(blk_post_runtime_resume);
d07ab6d1
MW
3694
3695/**
3696 * blk_set_runtime_active - Force runtime status of the queue to be active
3697 * @q: the queue of the device
3698 *
3699 * If the device is left runtime suspended during system suspend the resume
3700 * hook typically resumes the device and corrects runtime status
3701 * accordingly. However, that does not affect the queue runtime PM status
3702 * which is still "suspended". This prevents processing requests from the
3703 * queue.
3704 *
3705 * This function can be used in driver's resume hook to correct queue
3706 * runtime PM status and re-enable peeking requests from the queue. It
3707 * should be called before first request is added to the queue.
3708 */
3709void blk_set_runtime_active(struct request_queue *q)
3710{
3711 spin_lock_irq(q->queue_lock);
3712 q->rpm_status = RPM_ACTIVE;
3713 pm_runtime_mark_last_busy(q->dev);
3714 pm_request_autosuspend(q->dev);
3715 spin_unlock_irq(q->queue_lock);
3716}
3717EXPORT_SYMBOL(blk_set_runtime_active);
6c954667
LM
3718#endif
3719
1da177e4
LT
3720int __init blk_dev_init(void)
3721{
ef295ecf
CH
3722 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3723 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
0762b23d 3724 FIELD_SIZEOF(struct request, cmd_flags));
ef295ecf
CH
3725 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3726 FIELD_SIZEOF(struct bio, bi_opf));
9eb55b03 3727
89b90be2
TH
3728 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3729 kblockd_workqueue = alloc_workqueue("kblockd",
28747fcd 3730 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1da177e4
LT
3731 if (!kblockd_workqueue)
3732 panic("Failed to create kblockd\n");
3733
3734 request_cachep = kmem_cache_create("blkdev_requests",
20c2df83 3735 sizeof(struct request), 0, SLAB_PANIC, NULL);
1da177e4 3736
c2789bd4 3737 blk_requestq_cachep = kmem_cache_create("request_queue",
165125e1 3738 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1da177e4 3739
18fbda91
OS
3740#ifdef CONFIG_DEBUG_FS
3741 blk_debugfs_root = debugfs_create_dir("block", NULL);
3742#endif
3743
d38ecf93 3744 return 0;
1da177e4 3745}