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#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
98static const char *const blk_op_name[] = {
99 REQ_OP_NAME(READ),
100 REQ_OP_NAME(WRITE),
101 REQ_OP_NAME(FLUSH),
102 REQ_OP_NAME(DISCARD),
103 REQ_OP_NAME(SECURE_ERASE),
104 REQ_OP_NAME(ZONE_RESET),
105 REQ_OP_NAME(ZONE_RESET_ALL),
106 REQ_OP_NAME(ZONE_OPEN),
107 REQ_OP_NAME(ZONE_CLOSE),
108 REQ_OP_NAME(ZONE_FINISH),
109 REQ_OP_NAME(ZONE_APPEND),
110 REQ_OP_NAME(WRITE_ZEROES),
111 REQ_OP_NAME(DRV_IN),
112 REQ_OP_NAME(DRV_OUT),
113};
114#undef REQ_OP_NAME
115
116/**
117 * blk_op_str - Return string XXX in the REQ_OP_XXX.
118 * @op: REQ_OP_XXX.
119 *
120 * Description: Centralize block layer function to convert REQ_OP_XXX into
121 * string format. Useful in the debugging and tracing bio or request. For
122 * invalid REQ_OP_XXX it returns string "UNKNOWN".
123 */
124inline const char *blk_op_str(enum req_op op)
125{
126 const char *op_str = "UNKNOWN";
127
128 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
129 op_str = blk_op_name[op];
130
131 return op_str;
132}
133EXPORT_SYMBOL_GPL(blk_op_str);
134
135static const struct {
136 int errno;
137 const char *name;
138} blk_errors[] = {
139 [BLK_STS_OK] = { 0, "" },
140 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
141 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
142 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
143 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
144 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
145 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
146 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
147 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
148 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
149 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
150 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
151 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
152
153 /* device mapper special case, should not leak out: */
154 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
155
156 /* zone device specific errors */
157 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
158 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
159
160 /* Command duration limit device-side timeout */
161 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
162
163 [BLK_STS_INVAL] = { -EINVAL, "invalid" },
164
165 /* everything else not covered above: */
166 [BLK_STS_IOERR] = { -EIO, "I/O" },
167};
168
169blk_status_t errno_to_blk_status(int errno)
170{
171 int i;
172
173 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
174 if (blk_errors[i].errno == errno)
175 return (__force blk_status_t)i;
176 }
177
178 return BLK_STS_IOERR;
179}
180EXPORT_SYMBOL_GPL(errno_to_blk_status);
181
182int blk_status_to_errno(blk_status_t status)
183{
184 int idx = (__force int)status;
185
186 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
187 return -EIO;
188 return blk_errors[idx].errno;
189}
190EXPORT_SYMBOL_GPL(blk_status_to_errno);
191
192const char *blk_status_to_str(blk_status_t status)
193{
194 int idx = (__force int)status;
195
196 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 return "<null>";
198 return blk_errors[idx].name;
199}
200EXPORT_SYMBOL_GPL(blk_status_to_str);
201
202/**
203 * blk_sync_queue - cancel any pending callbacks on a queue
204 * @q: the queue
205 *
206 * Description:
207 * The block layer may perform asynchronous callback activity
208 * on a queue, such as calling the unplug function after a timeout.
209 * A block device may call blk_sync_queue to ensure that any
210 * such activity is cancelled, thus allowing it to release resources
211 * that the callbacks might use. The caller must already have made sure
212 * that its ->submit_bio will not re-add plugging prior to calling
213 * this function.
214 *
215 * This function does not cancel any asynchronous activity arising
216 * out of elevator or throttling code. That would require elevator_exit()
217 * and blkcg_exit_queue() to be called with queue lock initialized.
218 *
219 */
220void blk_sync_queue(struct request_queue *q)
221{
222 del_timer_sync(&q->timeout);
223 cancel_work_sync(&q->timeout_work);
224}
225EXPORT_SYMBOL(blk_sync_queue);
226
227/**
228 * blk_set_pm_only - increment pm_only counter
229 * @q: request queue pointer
230 */
231void blk_set_pm_only(struct request_queue *q)
232{
233 atomic_inc(&q->pm_only);
234}
235EXPORT_SYMBOL_GPL(blk_set_pm_only);
236
237void blk_clear_pm_only(struct request_queue *q)
238{
239 int pm_only;
240
241 pm_only = atomic_dec_return(&q->pm_only);
242 WARN_ON_ONCE(pm_only < 0);
243 if (pm_only == 0)
244 wake_up_all(&q->mq_freeze_wq);
245}
246EXPORT_SYMBOL_GPL(blk_clear_pm_only);
247
248static void blk_free_queue_rcu(struct rcu_head *rcu_head)
249{
250 struct request_queue *q = container_of(rcu_head,
251 struct request_queue, rcu_head);
252
253 percpu_ref_exit(&q->q_usage_counter);
254 kmem_cache_free(blk_requestq_cachep, q);
255}
256
257static void blk_free_queue(struct request_queue *q)
258{
259 blk_free_queue_stats(q->stats);
260 if (queue_is_mq(q))
261 blk_mq_release(q);
262
263 ida_free(&blk_queue_ida, q->id);
264 lockdep_unregister_key(&q->io_lock_cls_key);
265 lockdep_unregister_key(&q->q_lock_cls_key);
266 call_rcu(&q->rcu_head, blk_free_queue_rcu);
267}
268
269/**
270 * blk_put_queue - decrement the request_queue refcount
271 * @q: the request_queue structure to decrement the refcount for
272 *
273 * Decrements the refcount of the request_queue and free it when the refcount
274 * reaches 0.
275 */
276void blk_put_queue(struct request_queue *q)
277{
278 if (refcount_dec_and_test(&q->refs))
279 blk_free_queue(q);
280}
281EXPORT_SYMBOL(blk_put_queue);
282
283bool blk_queue_start_drain(struct request_queue *q)
284{
285 /*
286 * When queue DYING flag is set, we need to block new req
287 * entering queue, so we call blk_freeze_queue_start() to
288 * prevent I/O from crossing blk_queue_enter().
289 */
290 bool freeze = __blk_freeze_queue_start(q, current);
291 if (queue_is_mq(q))
292 blk_mq_wake_waiters(q);
293 /* Make blk_queue_enter() reexamine the DYING flag. */
294 wake_up_all(&q->mq_freeze_wq);
295
296 return freeze;
297}
298
299/**
300 * blk_queue_enter() - try to increase q->q_usage_counter
301 * @q: request queue pointer
302 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
303 */
304int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
305{
306 const bool pm = flags & BLK_MQ_REQ_PM;
307
308 while (!blk_try_enter_queue(q, pm)) {
309 if (flags & BLK_MQ_REQ_NOWAIT)
310 return -EAGAIN;
311
312 /*
313 * read pair of barrier in blk_freeze_queue_start(), we need to
314 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
315 * reading .mq_freeze_depth or queue dying flag, otherwise the
316 * following wait may never return if the two reads are
317 * reordered.
318 */
319 smp_rmb();
320 wait_event(q->mq_freeze_wq,
321 (!q->mq_freeze_depth &&
322 blk_pm_resume_queue(pm, q)) ||
323 blk_queue_dying(q));
324 if (blk_queue_dying(q))
325 return -ENODEV;
326 }
327
328 rwsem_acquire_read(&q->q_lockdep_map, 0, 0, _RET_IP_);
329 rwsem_release(&q->q_lockdep_map, _RET_IP_);
330 return 0;
331}
332
333int __bio_queue_enter(struct request_queue *q, struct bio *bio)
334{
335 while (!blk_try_enter_queue(q, false)) {
336 struct gendisk *disk = bio->bi_bdev->bd_disk;
337
338 if (bio->bi_opf & REQ_NOWAIT) {
339 if (test_bit(GD_DEAD, &disk->state))
340 goto dead;
341 bio_wouldblock_error(bio);
342 return -EAGAIN;
343 }
344
345 /*
346 * read pair of barrier in blk_freeze_queue_start(), we need to
347 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
348 * reading .mq_freeze_depth or queue dying flag, otherwise the
349 * following wait may never return if the two reads are
350 * reordered.
351 */
352 smp_rmb();
353 wait_event(q->mq_freeze_wq,
354 (!q->mq_freeze_depth &&
355 blk_pm_resume_queue(false, q)) ||
356 test_bit(GD_DEAD, &disk->state));
357 if (test_bit(GD_DEAD, &disk->state))
358 goto dead;
359 }
360
361 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_);
362 rwsem_release(&q->io_lockdep_map, _RET_IP_);
363 return 0;
364dead:
365 bio_io_error(bio);
366 return -ENODEV;
367}
368
369void blk_queue_exit(struct request_queue *q)
370{
371 percpu_ref_put(&q->q_usage_counter);
372}
373
374static void blk_queue_usage_counter_release(struct percpu_ref *ref)
375{
376 struct request_queue *q =
377 container_of(ref, struct request_queue, q_usage_counter);
378
379 wake_up_all(&q->mq_freeze_wq);
380}
381
382static void blk_rq_timed_out_timer(struct timer_list *t)
383{
384 struct request_queue *q = from_timer(q, t, timeout);
385
386 kblockd_schedule_work(&q->timeout_work);
387}
388
389static void blk_timeout_work(struct work_struct *work)
390{
391}
392
393struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
394{
395 struct request_queue *q;
396 int error;
397
398 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
399 node_id);
400 if (!q)
401 return ERR_PTR(-ENOMEM);
402
403 q->last_merge = NULL;
404
405 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
406 if (q->id < 0) {
407 error = q->id;
408 goto fail_q;
409 }
410
411 q->stats = blk_alloc_queue_stats();
412 if (!q->stats) {
413 error = -ENOMEM;
414 goto fail_id;
415 }
416
417 error = blk_set_default_limits(lim);
418 if (error)
419 goto fail_stats;
420 q->limits = *lim;
421
422 q->node = node_id;
423
424 atomic_set(&q->nr_active_requests_shared_tags, 0);
425
426 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
427 INIT_WORK(&q->timeout_work, blk_timeout_work);
428 INIT_LIST_HEAD(&q->icq_list);
429
430 refcount_set(&q->refs, 1);
431 mutex_init(&q->debugfs_mutex);
432 mutex_init(&q->sysfs_lock);
433 mutex_init(&q->sysfs_dir_lock);
434 mutex_init(&q->limits_lock);
435 mutex_init(&q->rq_qos_mutex);
436 spin_lock_init(&q->queue_lock);
437
438 init_waitqueue_head(&q->mq_freeze_wq);
439 mutex_init(&q->mq_freeze_lock);
440
441 blkg_init_queue(q);
442
443 /*
444 * Init percpu_ref in atomic mode so that it's faster to shutdown.
445 * See blk_register_queue() for details.
446 */
447 error = percpu_ref_init(&q->q_usage_counter,
448 blk_queue_usage_counter_release,
449 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
450 if (error)
451 goto fail_stats;
452 lockdep_register_key(&q->io_lock_cls_key);
453 lockdep_register_key(&q->q_lock_cls_key);
454 lockdep_init_map(&q->io_lockdep_map, "&q->q_usage_counter(io)",
455 &q->io_lock_cls_key, 0);
456 lockdep_init_map(&q->q_lockdep_map, "&q->q_usage_counter(queue)",
457 &q->q_lock_cls_key, 0);
458
459 q->nr_requests = BLKDEV_DEFAULT_RQ;
460
461 return q;
462
463fail_stats:
464 blk_free_queue_stats(q->stats);
465fail_id:
466 ida_free(&blk_queue_ida, q->id);
467fail_q:
468 kmem_cache_free(blk_requestq_cachep, q);
469 return ERR_PTR(error);
470}
471
472/**
473 * blk_get_queue - increment the request_queue refcount
474 * @q: the request_queue structure to increment the refcount for
475 *
476 * Increment the refcount of the request_queue kobject.
477 *
478 * Context: Any context.
479 */
480bool blk_get_queue(struct request_queue *q)
481{
482 if (unlikely(blk_queue_dying(q)))
483 return false;
484 refcount_inc(&q->refs);
485 return true;
486}
487EXPORT_SYMBOL(blk_get_queue);
488
489#ifdef CONFIG_FAIL_MAKE_REQUEST
490
491static DECLARE_FAULT_ATTR(fail_make_request);
492
493static int __init setup_fail_make_request(char *str)
494{
495 return setup_fault_attr(&fail_make_request, str);
496}
497__setup("fail_make_request=", setup_fail_make_request);
498
499bool should_fail_request(struct block_device *part, unsigned int bytes)
500{
501 return bdev_test_flag(part, BD_MAKE_IT_FAIL) &&
502 should_fail(&fail_make_request, bytes);
503}
504
505static int __init fail_make_request_debugfs(void)
506{
507 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
508 NULL, &fail_make_request);
509
510 return PTR_ERR_OR_ZERO(dir);
511}
512
513late_initcall(fail_make_request_debugfs);
514#endif /* CONFIG_FAIL_MAKE_REQUEST */
515
516static inline void bio_check_ro(struct bio *bio)
517{
518 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
519 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
520 return;
521
522 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED))
523 return;
524
525 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED);
526
527 /*
528 * Use ioctl to set underlying disk of raid/dm to read-only
529 * will trigger this.
530 */
531 pr_warn("Trying to write to read-only block-device %pg\n",
532 bio->bi_bdev);
533 }
534}
535
536static noinline int should_fail_bio(struct bio *bio)
537{
538 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
539 return -EIO;
540 return 0;
541}
542ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
543
544/*
545 * Check whether this bio extends beyond the end of the device or partition.
546 * This may well happen - the kernel calls bread() without checking the size of
547 * the device, e.g., when mounting a file system.
548 */
549static inline int bio_check_eod(struct bio *bio)
550{
551 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
552 unsigned int nr_sectors = bio_sectors(bio);
553
554 if (nr_sectors &&
555 (nr_sectors > maxsector ||
556 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
557 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
558 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
559 current->comm, bio->bi_bdev, bio->bi_opf,
560 bio->bi_iter.bi_sector, nr_sectors, maxsector);
561 return -EIO;
562 }
563 return 0;
564}
565
566/*
567 * Remap block n of partition p to block n+start(p) of the disk.
568 */
569static int blk_partition_remap(struct bio *bio)
570{
571 struct block_device *p = bio->bi_bdev;
572
573 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
574 return -EIO;
575 if (bio_sectors(bio)) {
576 bio->bi_iter.bi_sector += p->bd_start_sect;
577 trace_block_bio_remap(bio, p->bd_dev,
578 bio->bi_iter.bi_sector -
579 p->bd_start_sect);
580 }
581 bio_set_flag(bio, BIO_REMAPPED);
582 return 0;
583}
584
585/*
586 * Check write append to a zoned block device.
587 */
588static inline blk_status_t blk_check_zone_append(struct request_queue *q,
589 struct bio *bio)
590{
591 int nr_sectors = bio_sectors(bio);
592
593 /* Only applicable to zoned block devices */
594 if (!bdev_is_zoned(bio->bi_bdev))
595 return BLK_STS_NOTSUPP;
596
597 /* The bio sector must point to the start of a sequential zone */
598 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector))
599 return BLK_STS_IOERR;
600
601 /*
602 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
603 * split and could result in non-contiguous sectors being written in
604 * different zones.
605 */
606 if (nr_sectors > q->limits.chunk_sectors)
607 return BLK_STS_IOERR;
608
609 /* Make sure the BIO is small enough and will not get split */
610 if (nr_sectors > q->limits.max_zone_append_sectors)
611 return BLK_STS_IOERR;
612
613 bio->bi_opf |= REQ_NOMERGE;
614
615 return BLK_STS_OK;
616}
617
618static void __submit_bio(struct bio *bio)
619{
620 /* If plug is not used, add new plug here to cache nsecs time. */
621 struct blk_plug plug;
622
623 if (unlikely(!blk_crypto_bio_prep(&bio)))
624 return;
625
626 blk_start_plug(&plug);
627
628 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) {
629 blk_mq_submit_bio(bio);
630 } else if (likely(bio_queue_enter(bio) == 0)) {
631 struct gendisk *disk = bio->bi_bdev->bd_disk;
632
633 if ((bio->bi_opf & REQ_POLLED) &&
634 !(disk->queue->limits.features & BLK_FEAT_POLL)) {
635 bio->bi_status = BLK_STS_NOTSUPP;
636 bio_endio(bio);
637 } else {
638 disk->fops->submit_bio(bio);
639 }
640 blk_queue_exit(disk->queue);
641 }
642
643 blk_finish_plug(&plug);
644}
645
646/*
647 * The loop in this function may be a bit non-obvious, and so deserves some
648 * explanation:
649 *
650 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
651 * that), so we have a list with a single bio.
652 * - We pretend that we have just taken it off a longer list, so we assign
653 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
654 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
655 * bios through a recursive call to submit_bio_noacct. If it did, we find a
656 * non-NULL value in bio_list and re-enter the loop from the top.
657 * - In this case we really did just take the bio of the top of the list (no
658 * pretending) and so remove it from bio_list, and call into ->submit_bio()
659 * again.
660 *
661 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
662 * bio_list_on_stack[1] contains bios that were submitted before the current
663 * ->submit_bio, but that haven't been processed yet.
664 */
665static void __submit_bio_noacct(struct bio *bio)
666{
667 struct bio_list bio_list_on_stack[2];
668
669 BUG_ON(bio->bi_next);
670
671 bio_list_init(&bio_list_on_stack[0]);
672 current->bio_list = bio_list_on_stack;
673
674 do {
675 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
676 struct bio_list lower, same;
677
678 /*
679 * Create a fresh bio_list for all subordinate requests.
680 */
681 bio_list_on_stack[1] = bio_list_on_stack[0];
682 bio_list_init(&bio_list_on_stack[0]);
683
684 __submit_bio(bio);
685
686 /*
687 * Sort new bios into those for a lower level and those for the
688 * same level.
689 */
690 bio_list_init(&lower);
691 bio_list_init(&same);
692 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
693 if (q == bdev_get_queue(bio->bi_bdev))
694 bio_list_add(&same, bio);
695 else
696 bio_list_add(&lower, bio);
697
698 /*
699 * Now assemble so we handle the lowest level first.
700 */
701 bio_list_merge(&bio_list_on_stack[0], &lower);
702 bio_list_merge(&bio_list_on_stack[0], &same);
703 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
704 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
705
706 current->bio_list = NULL;
707}
708
709static void __submit_bio_noacct_mq(struct bio *bio)
710{
711 struct bio_list bio_list[2] = { };
712
713 current->bio_list = bio_list;
714
715 do {
716 __submit_bio(bio);
717 } while ((bio = bio_list_pop(&bio_list[0])));
718
719 current->bio_list = NULL;
720}
721
722void submit_bio_noacct_nocheck(struct bio *bio)
723{
724 blk_cgroup_bio_start(bio);
725 blkcg_bio_issue_init(bio);
726
727 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
728 trace_block_bio_queue(bio);
729 /*
730 * Now that enqueuing has been traced, we need to trace
731 * completion as well.
732 */
733 bio_set_flag(bio, BIO_TRACE_COMPLETION);
734 }
735
736 /*
737 * We only want one ->submit_bio to be active at a time, else stack
738 * usage with stacked devices could be a problem. Use current->bio_list
739 * to collect a list of requests submited by a ->submit_bio method while
740 * it is active, and then process them after it returned.
741 */
742 if (current->bio_list)
743 bio_list_add(¤t->bio_list[0], bio);
744 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO))
745 __submit_bio_noacct_mq(bio);
746 else
747 __submit_bio_noacct(bio);
748}
749
750static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q,
751 struct bio *bio)
752{
753 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q))
754 return BLK_STS_INVAL;
755
756 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q))
757 return BLK_STS_INVAL;
758
759 return BLK_STS_OK;
760}
761
762/**
763 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
764 * @bio: The bio describing the location in memory and on the device.
765 *
766 * This is a version of submit_bio() that shall only be used for I/O that is
767 * resubmitted to lower level drivers by stacking block drivers. All file
768 * systems and other upper level users of the block layer should use
769 * submit_bio() instead.
770 */
771void submit_bio_noacct(struct bio *bio)
772{
773 struct block_device *bdev = bio->bi_bdev;
774 struct request_queue *q = bdev_get_queue(bdev);
775 blk_status_t status = BLK_STS_IOERR;
776
777 might_sleep();
778
779 /*
780 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
781 * if queue does not support NOWAIT.
782 */
783 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
784 goto not_supported;
785
786 if (should_fail_bio(bio))
787 goto end_io;
788 bio_check_ro(bio);
789 if (!bio_flagged(bio, BIO_REMAPPED)) {
790 if (unlikely(bio_check_eod(bio)))
791 goto end_io;
792 if (bdev_is_partition(bdev) &&
793 unlikely(blk_partition_remap(bio)))
794 goto end_io;
795 }
796
797 /*
798 * Filter flush bio's early so that bio based drivers without flush
799 * support don't have to worry about them.
800 */
801 if (op_is_flush(bio->bi_opf)) {
802 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
803 bio_op(bio) != REQ_OP_ZONE_APPEND))
804 goto end_io;
805 if (!bdev_write_cache(bdev)) {
806 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
807 if (!bio_sectors(bio)) {
808 status = BLK_STS_OK;
809 goto end_io;
810 }
811 }
812 }
813
814 switch (bio_op(bio)) {
815 case REQ_OP_READ:
816 break;
817 case REQ_OP_WRITE:
818 if (bio->bi_opf & REQ_ATOMIC) {
819 status = blk_validate_atomic_write_op_size(q, bio);
820 if (status != BLK_STS_OK)
821 goto end_io;
822 }
823 break;
824 case REQ_OP_FLUSH:
825 /*
826 * REQ_OP_FLUSH can't be submitted through bios, it is only
827 * synthetized in struct request by the flush state machine.
828 */
829 goto not_supported;
830 case REQ_OP_DISCARD:
831 if (!bdev_max_discard_sectors(bdev))
832 goto not_supported;
833 break;
834 case REQ_OP_SECURE_ERASE:
835 if (!bdev_max_secure_erase_sectors(bdev))
836 goto not_supported;
837 break;
838 case REQ_OP_ZONE_APPEND:
839 status = blk_check_zone_append(q, bio);
840 if (status != BLK_STS_OK)
841 goto end_io;
842 break;
843 case REQ_OP_WRITE_ZEROES:
844 if (!q->limits.max_write_zeroes_sectors)
845 goto not_supported;
846 break;
847 case REQ_OP_ZONE_RESET:
848 case REQ_OP_ZONE_OPEN:
849 case REQ_OP_ZONE_CLOSE:
850 case REQ_OP_ZONE_FINISH:
851 case REQ_OP_ZONE_RESET_ALL:
852 if (!bdev_is_zoned(bio->bi_bdev))
853 goto not_supported;
854 break;
855 case REQ_OP_DRV_IN:
856 case REQ_OP_DRV_OUT:
857 /*
858 * Driver private operations are only used with passthrough
859 * requests.
860 */
861 fallthrough;
862 default:
863 goto not_supported;
864 }
865
866 if (blk_throtl_bio(bio))
867 return;
868 submit_bio_noacct_nocheck(bio);
869 return;
870
871not_supported:
872 status = BLK_STS_NOTSUPP;
873end_io:
874 bio->bi_status = status;
875 bio_endio(bio);
876}
877EXPORT_SYMBOL(submit_bio_noacct);
878
879static void bio_set_ioprio(struct bio *bio)
880{
881 /* Nobody set ioprio so far? Initialize it based on task's nice value */
882 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
883 bio->bi_ioprio = get_current_ioprio();
884 blkcg_set_ioprio(bio);
885}
886
887/**
888 * submit_bio - submit a bio to the block device layer for I/O
889 * @bio: The &struct bio which describes the I/O
890 *
891 * submit_bio() is used to submit I/O requests to block devices. It is passed a
892 * fully set up &struct bio that describes the I/O that needs to be done. The
893 * bio will be send to the device described by the bi_bdev field.
894 *
895 * The success/failure status of the request, along with notification of
896 * completion, is delivered asynchronously through the ->bi_end_io() callback
897 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
898 * been called.
899 */
900void submit_bio(struct bio *bio)
901{
902 if (bio_op(bio) == REQ_OP_READ) {
903 task_io_account_read(bio->bi_iter.bi_size);
904 count_vm_events(PGPGIN, bio_sectors(bio));
905 } else if (bio_op(bio) == REQ_OP_WRITE) {
906 count_vm_events(PGPGOUT, bio_sectors(bio));
907 }
908
909 bio_set_ioprio(bio);
910 submit_bio_noacct(bio);
911}
912EXPORT_SYMBOL(submit_bio);
913
914/**
915 * bio_poll - poll for BIO completions
916 * @bio: bio to poll for
917 * @iob: batches of IO
918 * @flags: BLK_POLL_* flags that control the behavior
919 *
920 * Poll for completions on queue associated with the bio. Returns number of
921 * completed entries found.
922 *
923 * Note: the caller must either be the context that submitted @bio, or
924 * be in a RCU critical section to prevent freeing of @bio.
925 */
926int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
927{
928 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
929 struct block_device *bdev;
930 struct request_queue *q;
931 int ret = 0;
932
933 bdev = READ_ONCE(bio->bi_bdev);
934 if (!bdev)
935 return 0;
936
937 q = bdev_get_queue(bdev);
938 if (cookie == BLK_QC_T_NONE)
939 return 0;
940
941 blk_flush_plug(current->plug, false);
942
943 /*
944 * We need to be able to enter a frozen queue, similar to how
945 * timeouts also need to do that. If that is blocked, then we can
946 * have pending IO when a queue freeze is started, and then the
947 * wait for the freeze to finish will wait for polled requests to
948 * timeout as the poller is preventer from entering the queue and
949 * completing them. As long as we prevent new IO from being queued,
950 * that should be all that matters.
951 */
952 if (!percpu_ref_tryget(&q->q_usage_counter))
953 return 0;
954 if (queue_is_mq(q)) {
955 ret = blk_mq_poll(q, cookie, iob, flags);
956 } else {
957 struct gendisk *disk = q->disk;
958
959 if ((q->limits.features & BLK_FEAT_POLL) && disk &&
960 disk->fops->poll_bio)
961 ret = disk->fops->poll_bio(bio, iob, flags);
962 }
963 blk_queue_exit(q);
964 return ret;
965}
966EXPORT_SYMBOL_GPL(bio_poll);
967
968/*
969 * Helper to implement file_operations.iopoll. Requires the bio to be stored
970 * in iocb->private, and cleared before freeing the bio.
971 */
972int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
973 unsigned int flags)
974{
975 struct bio *bio;
976 int ret = 0;
977
978 /*
979 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
980 * point to a freshly allocated bio at this point. If that happens
981 * we have a few cases to consider:
982 *
983 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
984 * simply nothing in this case
985 * 2) the bio points to a not poll enabled device. bio_poll will catch
986 * this and return 0
987 * 3) the bio points to a poll capable device, including but not
988 * limited to the one that the original bio pointed to. In this
989 * case we will call into the actual poll method and poll for I/O,
990 * even if we don't need to, but it won't cause harm either.
991 *
992 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
993 * is still allocated. Because partitions hold a reference to the whole
994 * device bdev and thus disk, the disk is also still valid. Grabbing
995 * a reference to the queue in bio_poll() ensures the hctxs and requests
996 * are still valid as well.
997 */
998 rcu_read_lock();
999 bio = READ_ONCE(kiocb->private);
1000 if (bio)
1001 ret = bio_poll(bio, iob, flags);
1002 rcu_read_unlock();
1003
1004 return ret;
1005}
1006EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1007
1008void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1009{
1010 unsigned long stamp;
1011again:
1012 stamp = READ_ONCE(part->bd_stamp);
1013 if (unlikely(time_after(now, stamp)) &&
1014 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
1015 (end || part_in_flight(part)))
1016 __part_stat_add(part, io_ticks, now - stamp);
1017
1018 if (bdev_is_partition(part)) {
1019 part = bdev_whole(part);
1020 goto again;
1021 }
1022}
1023
1024unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1025 unsigned long start_time)
1026{
1027 part_stat_lock();
1028 update_io_ticks(bdev, start_time, false);
1029 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1030 part_stat_unlock();
1031
1032 return start_time;
1033}
1034EXPORT_SYMBOL(bdev_start_io_acct);
1035
1036/**
1037 * bio_start_io_acct - start I/O accounting for bio based drivers
1038 * @bio: bio to start account for
1039 *
1040 * Returns the start time that should be passed back to bio_end_io_acct().
1041 */
1042unsigned long bio_start_io_acct(struct bio *bio)
1043{
1044 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1045}
1046EXPORT_SYMBOL_GPL(bio_start_io_acct);
1047
1048void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1049 unsigned int sectors, unsigned long start_time)
1050{
1051 const int sgrp = op_stat_group(op);
1052 unsigned long now = READ_ONCE(jiffies);
1053 unsigned long duration = now - start_time;
1054
1055 part_stat_lock();
1056 update_io_ticks(bdev, now, true);
1057 part_stat_inc(bdev, ios[sgrp]);
1058 part_stat_add(bdev, sectors[sgrp], sectors);
1059 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1060 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1061 part_stat_unlock();
1062}
1063EXPORT_SYMBOL(bdev_end_io_acct);
1064
1065void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1066 struct block_device *orig_bdev)
1067{
1068 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1069}
1070EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1071
1072/**
1073 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1074 * @q : the queue of the device being checked
1075 *
1076 * Description:
1077 * Check if underlying low-level drivers of a device are busy.
1078 * If the drivers want to export their busy state, they must set own
1079 * exporting function using blk_queue_lld_busy() first.
1080 *
1081 * Basically, this function is used only by request stacking drivers
1082 * to stop dispatching requests to underlying devices when underlying
1083 * devices are busy. This behavior helps more I/O merging on the queue
1084 * of the request stacking driver and prevents I/O throughput regression
1085 * on burst I/O load.
1086 *
1087 * Return:
1088 * 0 - Not busy (The request stacking driver should dispatch request)
1089 * 1 - Busy (The request stacking driver should stop dispatching request)
1090 */
1091int blk_lld_busy(struct request_queue *q)
1092{
1093 if (queue_is_mq(q) && q->mq_ops->busy)
1094 return q->mq_ops->busy(q);
1095
1096 return 0;
1097}
1098EXPORT_SYMBOL_GPL(blk_lld_busy);
1099
1100int kblockd_schedule_work(struct work_struct *work)
1101{
1102 return queue_work(kblockd_workqueue, work);
1103}
1104EXPORT_SYMBOL(kblockd_schedule_work);
1105
1106int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1107 unsigned long delay)
1108{
1109 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1110}
1111EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1112
1113void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1114{
1115 struct task_struct *tsk = current;
1116
1117 /*
1118 * If this is a nested plug, don't actually assign it.
1119 */
1120 if (tsk->plug)
1121 return;
1122
1123 plug->cur_ktime = 0;
1124 rq_list_init(&plug->mq_list);
1125 rq_list_init(&plug->cached_rqs);
1126 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1127 plug->rq_count = 0;
1128 plug->multiple_queues = false;
1129 plug->has_elevator = false;
1130 INIT_LIST_HEAD(&plug->cb_list);
1131
1132 /*
1133 * Store ordering should not be needed here, since a potential
1134 * preempt will imply a full memory barrier
1135 */
1136 tsk->plug = plug;
1137}
1138
1139/**
1140 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1141 * @plug: The &struct blk_plug that needs to be initialized
1142 *
1143 * Description:
1144 * blk_start_plug() indicates to the block layer an intent by the caller
1145 * to submit multiple I/O requests in a batch. The block layer may use
1146 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1147 * is called. However, the block layer may choose to submit requests
1148 * before a call to blk_finish_plug() if the number of queued I/Os
1149 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1150 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1151 * the task schedules (see below).
1152 *
1153 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1154 * pending I/O should the task end up blocking between blk_start_plug() and
1155 * blk_finish_plug(). This is important from a performance perspective, but
1156 * also ensures that we don't deadlock. For instance, if the task is blocking
1157 * for a memory allocation, memory reclaim could end up wanting to free a
1158 * page belonging to that request that is currently residing in our private
1159 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1160 * this kind of deadlock.
1161 */
1162void blk_start_plug(struct blk_plug *plug)
1163{
1164 blk_start_plug_nr_ios(plug, 1);
1165}
1166EXPORT_SYMBOL(blk_start_plug);
1167
1168static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1169{
1170 LIST_HEAD(callbacks);
1171
1172 while (!list_empty(&plug->cb_list)) {
1173 list_splice_init(&plug->cb_list, &callbacks);
1174
1175 while (!list_empty(&callbacks)) {
1176 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1177 struct blk_plug_cb,
1178 list);
1179 list_del(&cb->list);
1180 cb->callback(cb, from_schedule);
1181 }
1182 }
1183}
1184
1185struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1186 int size)
1187{
1188 struct blk_plug *plug = current->plug;
1189 struct blk_plug_cb *cb;
1190
1191 if (!plug)
1192 return NULL;
1193
1194 list_for_each_entry(cb, &plug->cb_list, list)
1195 if (cb->callback == unplug && cb->data == data)
1196 return cb;
1197
1198 /* Not currently on the callback list */
1199 BUG_ON(size < sizeof(*cb));
1200 cb = kzalloc(size, GFP_ATOMIC);
1201 if (cb) {
1202 cb->data = data;
1203 cb->callback = unplug;
1204 list_add(&cb->list, &plug->cb_list);
1205 }
1206 return cb;
1207}
1208EXPORT_SYMBOL(blk_check_plugged);
1209
1210void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1211{
1212 if (!list_empty(&plug->cb_list))
1213 flush_plug_callbacks(plug, from_schedule);
1214 blk_mq_flush_plug_list(plug, from_schedule);
1215 /*
1216 * Unconditionally flush out cached requests, even if the unplug
1217 * event came from schedule. Since we know hold references to the
1218 * queue for cached requests, we don't want a blocked task holding
1219 * up a queue freeze/quiesce event.
1220 */
1221 if (unlikely(!rq_list_empty(&plug->cached_rqs)))
1222 blk_mq_free_plug_rqs(plug);
1223
1224 plug->cur_ktime = 0;
1225 current->flags &= ~PF_BLOCK_TS;
1226}
1227
1228/**
1229 * blk_finish_plug - mark the end of a batch of submitted I/O
1230 * @plug: The &struct blk_plug passed to blk_start_plug()
1231 *
1232 * Description:
1233 * Indicate that a batch of I/O submissions is complete. This function
1234 * must be paired with an initial call to blk_start_plug(). The intent
1235 * is to allow the block layer to optimize I/O submission. See the
1236 * documentation for blk_start_plug() for more information.
1237 */
1238void blk_finish_plug(struct blk_plug *plug)
1239{
1240 if (plug == current->plug) {
1241 __blk_flush_plug(plug, false);
1242 current->plug = NULL;
1243 }
1244}
1245EXPORT_SYMBOL(blk_finish_plug);
1246
1247void blk_io_schedule(void)
1248{
1249 /* Prevent hang_check timer from firing at us during very long I/O */
1250 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1251
1252 if (timeout)
1253 io_schedule_timeout(timeout);
1254 else
1255 io_schedule();
1256}
1257EXPORT_SYMBOL_GPL(blk_io_schedule);
1258
1259int __init blk_dev_init(void)
1260{
1261 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1262 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1263 sizeof_field(struct request, cmd_flags));
1264 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1265 sizeof_field(struct bio, bi_opf));
1266
1267 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1268 kblockd_workqueue = alloc_workqueue("kblockd",
1269 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1270 if (!kblockd_workqueue)
1271 panic("Failed to create kblockd\n");
1272
1273 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1274
1275 blk_debugfs_root = debugfs_create_dir("block", NULL);
1276
1277 return 0;
1278}