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