1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-pm.h>
20#include <linux/blk-integrity.h>
21#include <linux/highmem.h>
22#include <linux/mm.h>
23#include <linux/pagemap.h>
24#include <linux/kernel_stat.h>
25#include <linux/string.h>
26#include <linux/init.h>
27#include <linux/completion.h>
28#include <linux/slab.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/task_io_accounting_ops.h>
32#include <linux/fault-inject.h>
33#include <linux/list_sort.h>
34#include <linux/delay.h>
35#include <linux/ratelimit.h>
36#include <linux/pm_runtime.h>
37#include <linux/t10-pi.h>
38#include <linux/debugfs.h>
39#include <linux/bpf.h>
40#include <linux/part_stat.h>
41#include <linux/sched/sysctl.h>
42#include <linux/blk-crypto.h>
43
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
46
47#include "blk.h"
48#include "blk-mq-sched.h"
49#include "blk-pm.h"
50#include "blk-cgroup.h"
51#include "blk-throttle.h"
52#include "blk-ioprio.h"
53
54struct dentry *blk_debugfs_root;
55
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62
63static DEFINE_IDA(blk_queue_ida);
64
65/*
66 * For queue allocation
67 */
68static struct kmem_cache *blk_requestq_cachep;
69
70/*
71 * Controlling structure to kblockd
72 */
73static struct workqueue_struct *kblockd_workqueue;
74
75/**
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
78 * @q: request queue
79 */
80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81{
82 set_bit(nr: flag, addr: &q->queue_flags);
83}
84EXPORT_SYMBOL(blk_queue_flag_set);
85
86/**
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
89 * @q: request queue
90 */
91void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92{
93 clear_bit(nr: flag, addr: &q->queue_flags);
94}
95EXPORT_SYMBOL(blk_queue_flag_clear);
96
97/**
98 * blk_queue_flag_test_and_set - atomically test and set a queue flag
99 * @flag: flag to be set
100 * @q: request queue
101 *
102 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
103 * the flag was already set.
104 */
105bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106{
107 return test_and_set_bit(nr: flag, addr: &q->queue_flags);
108}
109EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110
111#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
112static const char *const blk_op_name[] = {
113 REQ_OP_NAME(READ),
114 REQ_OP_NAME(WRITE),
115 REQ_OP_NAME(FLUSH),
116 REQ_OP_NAME(DISCARD),
117 REQ_OP_NAME(SECURE_ERASE),
118 REQ_OP_NAME(ZONE_RESET),
119 REQ_OP_NAME(ZONE_RESET_ALL),
120 REQ_OP_NAME(ZONE_OPEN),
121 REQ_OP_NAME(ZONE_CLOSE),
122 REQ_OP_NAME(ZONE_FINISH),
123 REQ_OP_NAME(ZONE_APPEND),
124 REQ_OP_NAME(WRITE_ZEROES),
125 REQ_OP_NAME(DRV_IN),
126 REQ_OP_NAME(DRV_OUT),
127};
128#undef REQ_OP_NAME
129
130/**
131 * blk_op_str - Return string XXX in the REQ_OP_XXX.
132 * @op: REQ_OP_XXX.
133 *
134 * Description: Centralize block layer function to convert REQ_OP_XXX into
135 * string format. Useful in the debugging and tracing bio or request. For
136 * invalid REQ_OP_XXX it returns string "UNKNOWN".
137 */
138inline const char *blk_op_str(enum req_op op)
139{
140 const char *op_str = "UNKNOWN";
141
142 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
143 op_str = blk_op_name[op];
144
145 return op_str;
146}
147EXPORT_SYMBOL_GPL(blk_op_str);
148
149static const struct {
150 int errno;
151 const char *name;
152} blk_errors[] = {
153 [BLK_STS_OK] = { .errno: 0, .name: "" },
154 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
155 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
156 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
157 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
158 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
159 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
160 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
161 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
162 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
163 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
164 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
165 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
166
167 /* device mapper special case, should not leak out: */
168 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
169
170 /* zone device specific errors */
171 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
172 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
173
174 /* Command duration limit device-side timeout */
175 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
176
177 /* everything else not covered above: */
178 [BLK_STS_IOERR] = { -EIO, "I/O" },
179};
180
181blk_status_t errno_to_blk_status(int errno)
182{
183 int i;
184
185 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
186 if (blk_errors[i].errno == errno)
187 return (__force blk_status_t)i;
188 }
189
190 return BLK_STS_IOERR;
191}
192EXPORT_SYMBOL_GPL(errno_to_blk_status);
193
194int blk_status_to_errno(blk_status_t status)
195{
196 int idx = (__force int)status;
197
198 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
199 return -EIO;
200 return blk_errors[idx].errno;
201}
202EXPORT_SYMBOL_GPL(blk_status_to_errno);
203
204const char *blk_status_to_str(blk_status_t status)
205{
206 int idx = (__force int)status;
207
208 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
209 return "<null>";
210 return blk_errors[idx].name;
211}
212EXPORT_SYMBOL_GPL(blk_status_to_str);
213
214/**
215 * blk_sync_queue - cancel any pending callbacks on a queue
216 * @q: the queue
217 *
218 * Description:
219 * The block layer may perform asynchronous callback activity
220 * on a queue, such as calling the unplug function after a timeout.
221 * A block device may call blk_sync_queue to ensure that any
222 * such activity is cancelled, thus allowing it to release resources
223 * that the callbacks might use. The caller must already have made sure
224 * that its ->submit_bio will not re-add plugging prior to calling
225 * this function.
226 *
227 * This function does not cancel any asynchronous activity arising
228 * out of elevator or throttling code. That would require elevator_exit()
229 * and blkcg_exit_queue() to be called with queue lock initialized.
230 *
231 */
232void blk_sync_queue(struct request_queue *q)
233{
234 del_timer_sync(timer: &q->timeout);
235 cancel_work_sync(work: &q->timeout_work);
236}
237EXPORT_SYMBOL(blk_sync_queue);
238
239/**
240 * blk_set_pm_only - increment pm_only counter
241 * @q: request queue pointer
242 */
243void blk_set_pm_only(struct request_queue *q)
244{
245 atomic_inc(v: &q->pm_only);
246}
247EXPORT_SYMBOL_GPL(blk_set_pm_only);
248
249void blk_clear_pm_only(struct request_queue *q)
250{
251 int pm_only;
252
253 pm_only = atomic_dec_return(v: &q->pm_only);
254 WARN_ON_ONCE(pm_only < 0);
255 if (pm_only == 0)
256 wake_up_all(&q->mq_freeze_wq);
257}
258EXPORT_SYMBOL_GPL(blk_clear_pm_only);
259
260static void blk_free_queue_rcu(struct rcu_head *rcu_head)
261{
262 struct request_queue *q = container_of(rcu_head,
263 struct request_queue, rcu_head);
264
265 percpu_ref_exit(ref: &q->q_usage_counter);
266 kmem_cache_free(s: blk_requestq_cachep, objp: q);
267}
268
269static void blk_free_queue(struct request_queue *q)
270{
271 blk_free_queue_stats(q->stats);
272 if (queue_is_mq(q))
273 blk_mq_release(q);
274
275 ida_free(&blk_queue_ida, id: q->id);
276 call_rcu(head: &q->rcu_head, func: blk_free_queue_rcu);
277}
278
279/**
280 * blk_put_queue - decrement the request_queue refcount
281 * @q: the request_queue structure to decrement the refcount for
282 *
283 * Decrements the refcount of the request_queue and free it when the refcount
284 * reaches 0.
285 */
286void blk_put_queue(struct request_queue *q)
287{
288 if (refcount_dec_and_test(r: &q->refs))
289 blk_free_queue(q);
290}
291EXPORT_SYMBOL(blk_put_queue);
292
293void blk_queue_start_drain(struct request_queue *q)
294{
295 /*
296 * When queue DYING flag is set, we need to block new req
297 * entering queue, so we call blk_freeze_queue_start() to
298 * prevent I/O from crossing blk_queue_enter().
299 */
300 blk_freeze_queue_start(q);
301 if (queue_is_mq(q))
302 blk_mq_wake_waiters(q);
303 /* Make blk_queue_enter() reexamine the DYING flag. */
304 wake_up_all(&q->mq_freeze_wq);
305}
306
307/**
308 * blk_queue_enter() - try to increase q->q_usage_counter
309 * @q: request queue pointer
310 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
311 */
312int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
313{
314 const bool pm = flags & BLK_MQ_REQ_PM;
315
316 while (!blk_try_enter_queue(q, pm)) {
317 if (flags & BLK_MQ_REQ_NOWAIT)
318 return -EAGAIN;
319
320 /*
321 * read pair of barrier in blk_freeze_queue_start(), we need to
322 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
323 * reading .mq_freeze_depth or queue dying flag, otherwise the
324 * following wait may never return if the two reads are
325 * reordered.
326 */
327 smp_rmb();
328 wait_event(q->mq_freeze_wq,
329 (!q->mq_freeze_depth &&
330 blk_pm_resume_queue(pm, q)) ||
331 blk_queue_dying(q));
332 if (blk_queue_dying(q))
333 return -ENODEV;
334 }
335
336 return 0;
337}
338
339int __bio_queue_enter(struct request_queue *q, struct bio *bio)
340{
341 while (!blk_try_enter_queue(q, pm: false)) {
342 struct gendisk *disk = bio->bi_bdev->bd_disk;
343
344 if (bio->bi_opf & REQ_NOWAIT) {
345 if (test_bit(GD_DEAD, &disk->state))
346 goto dead;
347 bio_wouldblock_error(bio);
348 return -EAGAIN;
349 }
350
351 /*
352 * read pair of barrier in blk_freeze_queue_start(), we need to
353 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
354 * reading .mq_freeze_depth or queue dying flag, otherwise the
355 * following wait may never return if the two reads are
356 * reordered.
357 */
358 smp_rmb();
359 wait_event(q->mq_freeze_wq,
360 (!q->mq_freeze_depth &&
361 blk_pm_resume_queue(false, q)) ||
362 test_bit(GD_DEAD, &disk->state));
363 if (test_bit(GD_DEAD, &disk->state))
364 goto dead;
365 }
366
367 return 0;
368dead:
369 bio_io_error(bio);
370 return -ENODEV;
371}
372
373void blk_queue_exit(struct request_queue *q)
374{
375 percpu_ref_put(ref: &q->q_usage_counter);
376}
377
378static void blk_queue_usage_counter_release(struct percpu_ref *ref)
379{
380 struct request_queue *q =
381 container_of(ref, struct request_queue, q_usage_counter);
382
383 wake_up_all(&q->mq_freeze_wq);
384}
385
386static void blk_rq_timed_out_timer(struct timer_list *t)
387{
388 struct request_queue *q = from_timer(q, t, timeout);
389
390 kblockd_schedule_work(work: &q->timeout_work);
391}
392
393static void blk_timeout_work(struct work_struct *work)
394{
395}
396
397struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
398{
399 struct request_queue *q;
400 int error;
401
402 q = kmem_cache_alloc_node(s: blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
403 node: node_id);
404 if (!q)
405 return ERR_PTR(error: -ENOMEM);
406
407 q->last_merge = NULL;
408
409 q->id = ida_alloc(ida: &blk_queue_ida, GFP_KERNEL);
410 if (q->id < 0) {
411 error = q->id;
412 goto fail_q;
413 }
414
415 q->stats = blk_alloc_queue_stats();
416 if (!q->stats) {
417 error = -ENOMEM;
418 goto fail_id;
419 }
420
421 error = blk_set_default_limits(lim);
422 if (error)
423 goto fail_stats;
424 q->limits = *lim;
425
426 q->node = node_id;
427
428 atomic_set(v: &q->nr_active_requests_shared_tags, i: 0);
429
430 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
431 INIT_WORK(&q->timeout_work, blk_timeout_work);
432 INIT_LIST_HEAD(list: &q->icq_list);
433
434 refcount_set(r: &q->refs, n: 1);
435 mutex_init(&q->debugfs_mutex);
436 mutex_init(&q->sysfs_lock);
437 mutex_init(&q->sysfs_dir_lock);
438 mutex_init(&q->limits_lock);
439 mutex_init(&q->rq_qos_mutex);
440 spin_lock_init(&q->queue_lock);
441
442 init_waitqueue_head(&q->mq_freeze_wq);
443 mutex_init(&q->mq_freeze_lock);
444
445 blkg_init_queue(q);
446
447 /*
448 * Init percpu_ref in atomic mode so that it's faster to shutdown.
449 * See blk_register_queue() for details.
450 */
451 error = percpu_ref_init(ref: &q->q_usage_counter,
452 release: blk_queue_usage_counter_release,
453 flags: PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
454 if (error)
455 goto fail_stats;
456
457 q->nr_requests = BLKDEV_DEFAULT_RQ;
458
459 return q;
460
461fail_stats:
462 blk_free_queue_stats(q->stats);
463fail_id:
464 ida_free(&blk_queue_ida, id: q->id);
465fail_q:
466 kmem_cache_free(s: blk_requestq_cachep, objp: q);
467 return ERR_PTR(error);
468}
469
470/**
471 * blk_get_queue - increment the request_queue refcount
472 * @q: the request_queue structure to increment the refcount for
473 *
474 * Increment the refcount of the request_queue kobject.
475 *
476 * Context: Any context.
477 */
478bool blk_get_queue(struct request_queue *q)
479{
480 if (unlikely(blk_queue_dying(q)))
481 return false;
482 refcount_inc(r: &q->refs);
483 return true;
484}
485EXPORT_SYMBOL(blk_get_queue);
486
487#ifdef CONFIG_FAIL_MAKE_REQUEST
488
489static DECLARE_FAULT_ATTR(fail_make_request);
490
491static int __init setup_fail_make_request(char *str)
492{
493 return setup_fault_attr(attr: &fail_make_request, str);
494}
495__setup("fail_make_request=", setup_fail_make_request);
496
497bool should_fail_request(struct block_device *part, unsigned int bytes)
498{
499 return part->bd_make_it_fail && should_fail(attr: &fail_make_request, size: bytes);
500}
501
502static int __init fail_make_request_debugfs(void)
503{
504 struct dentry *dir = fault_create_debugfs_attr(name: "fail_make_request",
505 NULL, attr: &fail_make_request);
506
507 return PTR_ERR_OR_ZERO(ptr: dir);
508}
509
510late_initcall(fail_make_request_debugfs);
511#endif /* CONFIG_FAIL_MAKE_REQUEST */
512
513static inline void bio_check_ro(struct bio *bio)
514{
515 if (op_is_write(op: bio_op(bio)) && bdev_read_only(bdev: bio->bi_bdev)) {
516 if (op_is_flush(op: bio->bi_opf) && !bio_sectors(bio))
517 return;
518
519 if (bio->bi_bdev->bd_ro_warned)
520 return;
521
522 bio->bi_bdev->bd_ro_warned = true;
523 /*
524 * Use ioctl to set underlying disk of raid/dm to read-only
525 * will trigger this.
526 */
527 pr_warn("Trying to write to read-only block-device %pg\n",
528 bio->bi_bdev);
529 }
530}
531
532static noinline int should_fail_bio(struct bio *bio)
533{
534 if (should_fail_request(bdev_whole(bio->bi_bdev), bytes: bio->bi_iter.bi_size))
535 return -EIO;
536 return 0;
537}
538ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
539
540/*
541 * Check whether this bio extends beyond the end of the device or partition.
542 * This may well happen - the kernel calls bread() without checking the size of
543 * the device, e.g., when mounting a file system.
544 */
545static inline int bio_check_eod(struct bio *bio)
546{
547 sector_t maxsector = bdev_nr_sectors(bdev: bio->bi_bdev);
548 unsigned int nr_sectors = bio_sectors(bio);
549
550 if (nr_sectors &&
551 (nr_sectors > maxsector ||
552 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
553 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
554 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
555 current->comm, bio->bi_bdev, bio->bi_opf,
556 bio->bi_iter.bi_sector, nr_sectors, maxsector);
557 return -EIO;
558 }
559 return 0;
560}
561
562/*
563 * Remap block n of partition p to block n+start(p) of the disk.
564 */
565static int blk_partition_remap(struct bio *bio)
566{
567 struct block_device *p = bio->bi_bdev;
568
569 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
570 return -EIO;
571 if (bio_sectors(bio)) {
572 bio->bi_iter.bi_sector += p->bd_start_sect;
573 trace_block_bio_remap(bio, dev: p->bd_dev,
574 from: bio->bi_iter.bi_sector -
575 p->bd_start_sect);
576 }
577 bio_set_flag(bio, bit: BIO_REMAPPED);
578 return 0;
579}
580
581/*
582 * Check write append to a zoned block device.
583 */
584static inline blk_status_t blk_check_zone_append(struct request_queue *q,
585 struct bio *bio)
586{
587 int nr_sectors = bio_sectors(bio);
588
589 /* Only applicable to zoned block devices */
590 if (!bdev_is_zoned(bdev: bio->bi_bdev))
591 return BLK_STS_NOTSUPP;
592
593 /* The bio sector must point to the start of a sequential zone */
594 if (!bdev_is_zone_start(bdev: bio->bi_bdev, sector: bio->bi_iter.bi_sector) ||
595 !bio_zone_is_seq(bio))
596 return BLK_STS_IOERR;
597
598 /*
599 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
600 * split and could result in non-contiguous sectors being written in
601 * different zones.
602 */
603 if (nr_sectors > q->limits.chunk_sectors)
604 return BLK_STS_IOERR;
605
606 /* Make sure the BIO is small enough and will not get split */
607 if (nr_sectors > q->limits.max_zone_append_sectors)
608 return BLK_STS_IOERR;
609
610 bio->bi_opf |= REQ_NOMERGE;
611
612 return BLK_STS_OK;
613}
614
615static void __submit_bio(struct bio *bio)
616{
617 if (unlikely(!blk_crypto_bio_prep(&bio)))
618 return;
619
620 if (!bio->bi_bdev->bd_has_submit_bio) {
621 blk_mq_submit_bio(bio);
622 } else if (likely(bio_queue_enter(bio) == 0)) {
623 struct gendisk *disk = bio->bi_bdev->bd_disk;
624
625 disk->fops->submit_bio(bio);
626 blk_queue_exit(q: disk->queue);
627 }
628}
629
630/*
631 * The loop in this function may be a bit non-obvious, and so deserves some
632 * explanation:
633 *
634 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
635 * that), so we have a list with a single bio.
636 * - We pretend that we have just taken it off a longer list, so we assign
637 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
638 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
639 * bios through a recursive call to submit_bio_noacct. If it did, we find a
640 * non-NULL value in bio_list and re-enter the loop from the top.
641 * - In this case we really did just take the bio of the top of the list (no
642 * pretending) and so remove it from bio_list, and call into ->submit_bio()
643 * again.
644 *
645 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
646 * bio_list_on_stack[1] contains bios that were submitted before the current
647 * ->submit_bio, but that haven't been processed yet.
648 */
649static void __submit_bio_noacct(struct bio *bio)
650{
651 struct bio_list bio_list_on_stack[2];
652
653 BUG_ON(bio->bi_next);
654
655 bio_list_init(bl: &bio_list_on_stack[0]);
656 current->bio_list = bio_list_on_stack;
657
658 do {
659 struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev);
660 struct bio_list lower, same;
661
662 /*
663 * Create a fresh bio_list for all subordinate requests.
664 */
665 bio_list_on_stack[1] = bio_list_on_stack[0];
666 bio_list_init(bl: &bio_list_on_stack[0]);
667
668 __submit_bio(bio);
669
670 /*
671 * Sort new bios into those for a lower level and those for the
672 * same level.
673 */
674 bio_list_init(bl: &lower);
675 bio_list_init(bl: &same);
676 while ((bio = bio_list_pop(bl: &bio_list_on_stack[0])) != NULL)
677 if (q == bdev_get_queue(bdev: bio->bi_bdev))
678 bio_list_add(bl: &same, bio);
679 else
680 bio_list_add(bl: &lower, bio);
681
682 /*
683 * Now assemble so we handle the lowest level first.
684 */
685 bio_list_merge(bl: &bio_list_on_stack[0], bl2: &lower);
686 bio_list_merge(bl: &bio_list_on_stack[0], bl2: &same);
687 bio_list_merge(bl: &bio_list_on_stack[0], bl2: &bio_list_on_stack[1]);
688 } while ((bio = bio_list_pop(bl: &bio_list_on_stack[0])));
689
690 current->bio_list = NULL;
691}
692
693static void __submit_bio_noacct_mq(struct bio *bio)
694{
695 struct bio_list bio_list[2] = { };
696
697 current->bio_list = bio_list;
698
699 do {
700 __submit_bio(bio);
701 } while ((bio = bio_list_pop(bl: &bio_list[0])));
702
703 current->bio_list = NULL;
704}
705
706void submit_bio_noacct_nocheck(struct bio *bio)
707{
708 blk_cgroup_bio_start(bio);
709 blkcg_bio_issue_init(bio);
710
711 if (!bio_flagged(bio, bit: BIO_TRACE_COMPLETION)) {
712 trace_block_bio_queue(bio);
713 /*
714 * Now that enqueuing has been traced, we need to trace
715 * completion as well.
716 */
717 bio_set_flag(bio, bit: BIO_TRACE_COMPLETION);
718 }
719
720 /*
721 * We only want one ->submit_bio to be active at a time, else stack
722 * usage with stacked devices could be a problem. Use current->bio_list
723 * to collect a list of requests submited by a ->submit_bio method while
724 * it is active, and then process them after it returned.
725 */
726 if (current->bio_list)
727 bio_list_add(bl: &current->bio_list[0], bio);
728 else if (!bio->bi_bdev->bd_has_submit_bio)
729 __submit_bio_noacct_mq(bio);
730 else
731 __submit_bio_noacct(bio);
732}
733
734/**
735 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
736 * @bio: The bio describing the location in memory and on the device.
737 *
738 * This is a version of submit_bio() that shall only be used for I/O that is
739 * resubmitted to lower level drivers by stacking block drivers. All file
740 * systems and other upper level users of the block layer should use
741 * submit_bio() instead.
742 */
743void submit_bio_noacct(struct bio *bio)
744{
745 struct block_device *bdev = bio->bi_bdev;
746 struct request_queue *q = bdev_get_queue(bdev);
747 blk_status_t status = BLK_STS_IOERR;
748
749 might_sleep();
750
751 /*
752 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
753 * if queue does not support NOWAIT.
754 */
755 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
756 goto not_supported;
757
758 if (should_fail_bio(bio))
759 goto end_io;
760 bio_check_ro(bio);
761 if (!bio_flagged(bio, bit: BIO_REMAPPED)) {
762 if (unlikely(bio_check_eod(bio)))
763 goto end_io;
764 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
765 goto end_io;
766 }
767
768 /*
769 * Filter flush bio's early so that bio based drivers without flush
770 * support don't have to worry about them.
771 */
772 if (op_is_flush(op: bio->bi_opf)) {
773 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
774 bio_op(bio) != REQ_OP_ZONE_APPEND))
775 goto end_io;
776 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
777 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
778 if (!bio_sectors(bio)) {
779 status = BLK_STS_OK;
780 goto end_io;
781 }
782 }
783 }
784
785 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
786 bio_clear_polled(bio);
787
788 switch (bio_op(bio)) {
789 case REQ_OP_READ:
790 case REQ_OP_WRITE:
791 break;
792 case REQ_OP_FLUSH:
793 /*
794 * REQ_OP_FLUSH can't be submitted through bios, it is only
795 * synthetized in struct request by the flush state machine.
796 */
797 goto not_supported;
798 case REQ_OP_DISCARD:
799 if (!bdev_max_discard_sectors(bdev))
800 goto not_supported;
801 break;
802 case REQ_OP_SECURE_ERASE:
803 if (!bdev_max_secure_erase_sectors(bdev))
804 goto not_supported;
805 break;
806 case REQ_OP_ZONE_APPEND:
807 status = blk_check_zone_append(q, bio);
808 if (status != BLK_STS_OK)
809 goto end_io;
810 break;
811 case REQ_OP_WRITE_ZEROES:
812 if (!q->limits.max_write_zeroes_sectors)
813 goto not_supported;
814 break;
815 case REQ_OP_ZONE_RESET:
816 case REQ_OP_ZONE_OPEN:
817 case REQ_OP_ZONE_CLOSE:
818 case REQ_OP_ZONE_FINISH:
819 if (!bdev_is_zoned(bdev: bio->bi_bdev))
820 goto not_supported;
821 break;
822 case REQ_OP_ZONE_RESET_ALL:
823 if (!bdev_is_zoned(bdev: bio->bi_bdev) || !blk_queue_zone_resetall(q))
824 goto not_supported;
825 break;
826 case REQ_OP_DRV_IN:
827 case REQ_OP_DRV_OUT:
828 /*
829 * Driver private operations are only used with passthrough
830 * requests.
831 */
832 fallthrough;
833 default:
834 goto not_supported;
835 }
836
837 if (blk_throtl_bio(bio))
838 return;
839 submit_bio_noacct_nocheck(bio);
840 return;
841
842not_supported:
843 status = BLK_STS_NOTSUPP;
844end_io:
845 bio->bi_status = status;
846 bio_endio(bio);
847}
848EXPORT_SYMBOL(submit_bio_noacct);
849
850static void bio_set_ioprio(struct bio *bio)
851{
852 /* Nobody set ioprio so far? Initialize it based on task's nice value */
853 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
854 bio->bi_ioprio = get_current_ioprio();
855 blkcg_set_ioprio(bio);
856}
857
858/**
859 * submit_bio - submit a bio to the block device layer for I/O
860 * @bio: The &struct bio which describes the I/O
861 *
862 * submit_bio() is used to submit I/O requests to block devices. It is passed a
863 * fully set up &struct bio that describes the I/O that needs to be done. The
864 * bio will be send to the device described by the bi_bdev field.
865 *
866 * The success/failure status of the request, along with notification of
867 * completion, is delivered asynchronously through the ->bi_end_io() callback
868 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
869 * been called.
870 */
871void submit_bio(struct bio *bio)
872{
873 if (bio_op(bio) == REQ_OP_READ) {
874 task_io_account_read(bytes: bio->bi_iter.bi_size);
875 count_vm_events(item: PGPGIN, bio_sectors(bio));
876 } else if (bio_op(bio) == REQ_OP_WRITE) {
877 count_vm_events(item: PGPGOUT, bio_sectors(bio));
878 }
879
880 bio_set_ioprio(bio);
881 submit_bio_noacct(bio);
882}
883EXPORT_SYMBOL(submit_bio);
884
885/**
886 * bio_poll - poll for BIO completions
887 * @bio: bio to poll for
888 * @iob: batches of IO
889 * @flags: BLK_POLL_* flags that control the behavior
890 *
891 * Poll for completions on queue associated with the bio. Returns number of
892 * completed entries found.
893 *
894 * Note: the caller must either be the context that submitted @bio, or
895 * be in a RCU critical section to prevent freeing of @bio.
896 */
897int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
898{
899 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
900 struct block_device *bdev;
901 struct request_queue *q;
902 int ret = 0;
903
904 bdev = READ_ONCE(bio->bi_bdev);
905 if (!bdev)
906 return 0;
907
908 q = bdev_get_queue(bdev);
909 if (cookie == BLK_QC_T_NONE ||
910 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
911 return 0;
912
913 /*
914 * As the requests that require a zone lock are not plugged in the
915 * first place, directly accessing the plug instead of using
916 * blk_mq_plug() should not have any consequences during flushing for
917 * zoned devices.
918 */
919 blk_flush_plug(current->plug, async: false);
920
921 /*
922 * We need to be able to enter a frozen queue, similar to how
923 * timeouts also need to do that. If that is blocked, then we can
924 * have pending IO when a queue freeze is started, and then the
925 * wait for the freeze to finish will wait for polled requests to
926 * timeout as the poller is preventer from entering the queue and
927 * completing them. As long as we prevent new IO from being queued,
928 * that should be all that matters.
929 */
930 if (!percpu_ref_tryget(ref: &q->q_usage_counter))
931 return 0;
932 if (queue_is_mq(q)) {
933 ret = blk_mq_poll(q, cookie, iob, flags);
934 } else {
935 struct gendisk *disk = q->disk;
936
937 if (disk && disk->fops->poll_bio)
938 ret = disk->fops->poll_bio(bio, iob, flags);
939 }
940 blk_queue_exit(q);
941 return ret;
942}
943EXPORT_SYMBOL_GPL(bio_poll);
944
945/*
946 * Helper to implement file_operations.iopoll. Requires the bio to be stored
947 * in iocb->private, and cleared before freeing the bio.
948 */
949int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
950 unsigned int flags)
951{
952 struct bio *bio;
953 int ret = 0;
954
955 /*
956 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
957 * point to a freshly allocated bio at this point. If that happens
958 * we have a few cases to consider:
959 *
960 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
961 * simply nothing in this case
962 * 2) the bio points to a not poll enabled device. bio_poll will catch
963 * this and return 0
964 * 3) the bio points to a poll capable device, including but not
965 * limited to the one that the original bio pointed to. In this
966 * case we will call into the actual poll method and poll for I/O,
967 * even if we don't need to, but it won't cause harm either.
968 *
969 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
970 * is still allocated. Because partitions hold a reference to the whole
971 * device bdev and thus disk, the disk is also still valid. Grabbing
972 * a reference to the queue in bio_poll() ensures the hctxs and requests
973 * are still valid as well.
974 */
975 rcu_read_lock();
976 bio = READ_ONCE(kiocb->private);
977 if (bio)
978 ret = bio_poll(bio, iob, flags);
979 rcu_read_unlock();
980
981 return ret;
982}
983EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
984
985void update_io_ticks(struct block_device *part, unsigned long now, bool end)
986{
987 unsigned long stamp;
988again:
989 stamp = READ_ONCE(part->bd_stamp);
990 if (unlikely(time_after(now, stamp))) {
991 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
992 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
993 }
994 if (part->bd_partno) {
995 part = bdev_whole(part);
996 goto again;
997 }
998}
999
1000unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1001 unsigned long start_time)
1002{
1003 part_stat_lock();
1004 update_io_ticks(part: bdev, now: start_time, end: false);
1005 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1006 part_stat_unlock();
1007
1008 return start_time;
1009}
1010EXPORT_SYMBOL(bdev_start_io_acct);
1011
1012/**
1013 * bio_start_io_acct - start I/O accounting for bio based drivers
1014 * @bio: bio to start account for
1015 *
1016 * Returns the start time that should be passed back to bio_end_io_acct().
1017 */
1018unsigned long bio_start_io_acct(struct bio *bio)
1019{
1020 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1021}
1022EXPORT_SYMBOL_GPL(bio_start_io_acct);
1023
1024void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1025 unsigned int sectors, unsigned long start_time)
1026{
1027 const int sgrp = op_stat_group(op);
1028 unsigned long now = READ_ONCE(jiffies);
1029 unsigned long duration = now - start_time;
1030
1031 part_stat_lock();
1032 update_io_ticks(part: bdev, now, end: true);
1033 part_stat_inc(bdev, ios[sgrp]);
1034 part_stat_add(bdev, sectors[sgrp], sectors);
1035 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1036 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1037 part_stat_unlock();
1038}
1039EXPORT_SYMBOL(bdev_end_io_acct);
1040
1041void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1042 struct block_device *orig_bdev)
1043{
1044 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1045}
1046EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1047
1048/**
1049 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1050 * @q : the queue of the device being checked
1051 *
1052 * Description:
1053 * Check if underlying low-level drivers of a device are busy.
1054 * If the drivers want to export their busy state, they must set own
1055 * exporting function using blk_queue_lld_busy() first.
1056 *
1057 * Basically, this function is used only by request stacking drivers
1058 * to stop dispatching requests to underlying devices when underlying
1059 * devices are busy. This behavior helps more I/O merging on the queue
1060 * of the request stacking driver and prevents I/O throughput regression
1061 * on burst I/O load.
1062 *
1063 * Return:
1064 * 0 - Not busy (The request stacking driver should dispatch request)
1065 * 1 - Busy (The request stacking driver should stop dispatching request)
1066 */
1067int blk_lld_busy(struct request_queue *q)
1068{
1069 if (queue_is_mq(q) && q->mq_ops->busy)
1070 return q->mq_ops->busy(q);
1071
1072 return 0;
1073}
1074EXPORT_SYMBOL_GPL(blk_lld_busy);
1075
1076int kblockd_schedule_work(struct work_struct *work)
1077{
1078 return queue_work(wq: kblockd_workqueue, work);
1079}
1080EXPORT_SYMBOL(kblockd_schedule_work);
1081
1082int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1083 unsigned long delay)
1084{
1085 return mod_delayed_work_on(cpu, wq: kblockd_workqueue, dwork, delay);
1086}
1087EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1088
1089void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1090{
1091 struct task_struct *tsk = current;
1092
1093 /*
1094 * If this is a nested plug, don't actually assign it.
1095 */
1096 if (tsk->plug)
1097 return;
1098
1099 plug->cur_ktime = 0;
1100 plug->mq_list = NULL;
1101 plug->cached_rq = NULL;
1102 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1103 plug->rq_count = 0;
1104 plug->multiple_queues = false;
1105 plug->has_elevator = false;
1106 INIT_LIST_HEAD(list: &plug->cb_list);
1107
1108 /*
1109 * Store ordering should not be needed here, since a potential
1110 * preempt will imply a full memory barrier
1111 */
1112 tsk->plug = plug;
1113}
1114
1115/**
1116 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1117 * @plug: The &struct blk_plug that needs to be initialized
1118 *
1119 * Description:
1120 * blk_start_plug() indicates to the block layer an intent by the caller
1121 * to submit multiple I/O requests in a batch. The block layer may use
1122 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1123 * is called. However, the block layer may choose to submit requests
1124 * before a call to blk_finish_plug() if the number of queued I/Os
1125 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1126 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1127 * the task schedules (see below).
1128 *
1129 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1130 * pending I/O should the task end up blocking between blk_start_plug() and
1131 * blk_finish_plug(). This is important from a performance perspective, but
1132 * also ensures that we don't deadlock. For instance, if the task is blocking
1133 * for a memory allocation, memory reclaim could end up wanting to free a
1134 * page belonging to that request that is currently residing in our private
1135 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1136 * this kind of deadlock.
1137 */
1138void blk_start_plug(struct blk_plug *plug)
1139{
1140 blk_start_plug_nr_ios(plug, nr_ios: 1);
1141}
1142EXPORT_SYMBOL(blk_start_plug);
1143
1144static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1145{
1146 LIST_HEAD(callbacks);
1147
1148 while (!list_empty(head: &plug->cb_list)) {
1149 list_splice_init(list: &plug->cb_list, head: &callbacks);
1150
1151 while (!list_empty(head: &callbacks)) {
1152 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1153 struct blk_plug_cb,
1154 list);
1155 list_del(entry: &cb->list);
1156 cb->callback(cb, from_schedule);
1157 }
1158 }
1159}
1160
1161struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1162 int size)
1163{
1164 struct blk_plug *plug = current->plug;
1165 struct blk_plug_cb *cb;
1166
1167 if (!plug)
1168 return NULL;
1169
1170 list_for_each_entry(cb, &plug->cb_list, list)
1171 if (cb->callback == unplug && cb->data == data)
1172 return cb;
1173
1174 /* Not currently on the callback list */
1175 BUG_ON(size < sizeof(*cb));
1176 cb = kzalloc(size, GFP_ATOMIC);
1177 if (cb) {
1178 cb->data = data;
1179 cb->callback = unplug;
1180 list_add(new: &cb->list, head: &plug->cb_list);
1181 }
1182 return cb;
1183}
1184EXPORT_SYMBOL(blk_check_plugged);
1185
1186void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1187{
1188 if (!list_empty(head: &plug->cb_list))
1189 flush_plug_callbacks(plug, from_schedule);
1190 blk_mq_flush_plug_list(plug, from_schedule);
1191 /*
1192 * Unconditionally flush out cached requests, even if the unplug
1193 * event came from schedule. Since we know hold references to the
1194 * queue for cached requests, we don't want a blocked task holding
1195 * up a queue freeze/quiesce event.
1196 */
1197 if (unlikely(!rq_list_empty(plug->cached_rq)))
1198 blk_mq_free_plug_rqs(plug);
1199
1200 plug->cur_ktime = 0;
1201 current->flags &= ~PF_BLOCK_TS;
1202}
1203
1204/**
1205 * blk_finish_plug - mark the end of a batch of submitted I/O
1206 * @plug: The &struct blk_plug passed to blk_start_plug()
1207 *
1208 * Description:
1209 * Indicate that a batch of I/O submissions is complete. This function
1210 * must be paired with an initial call to blk_start_plug(). The intent
1211 * is to allow the block layer to optimize I/O submission. See the
1212 * documentation for blk_start_plug() for more information.
1213 */
1214void blk_finish_plug(struct blk_plug *plug)
1215{
1216 if (plug == current->plug) {
1217 __blk_flush_plug(plug, from_schedule: false);
1218 current->plug = NULL;
1219 }
1220}
1221EXPORT_SYMBOL(blk_finish_plug);
1222
1223void blk_io_schedule(void)
1224{
1225 /* Prevent hang_check timer from firing at us during very long I/O */
1226 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1227
1228 if (timeout)
1229 io_schedule_timeout(timeout);
1230 else
1231 io_schedule();
1232}
1233EXPORT_SYMBOL_GPL(blk_io_schedule);
1234
1235int __init blk_dev_init(void)
1236{
1237 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1238 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1239 sizeof_field(struct request, cmd_flags));
1240 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1241 sizeof_field(struct bio, bi_opf));
1242
1243 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1244 kblockd_workqueue = alloc_workqueue(fmt: "kblockd",
1245 flags: WQ_MEM_RECLAIM | WQ_HIGHPRI, max_active: 0);
1246 if (!kblockd_workqueue)
1247 panic(fmt: "Failed to create kblockd\n");
1248
1249 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1250
1251 blk_debugfs_root = debugfs_create_dir(name: "block", NULL);
1252
1253 return 0;
1254}
1255

source code of linux/block/blk-core.c