1 // SPDX-License-Identifier: GPL-2.0
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>
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
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>
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>
44 #define CREATE_TRACE_POINTS
45 #include <trace/events/block.h>
48 #include "blk-mq-sched.h"
50 #include "blk-cgroup.h"
51 #include "blk-throttle.h"
53 struct dentry *blk_debugfs_root;
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62 static DEFINE_IDA(blk_queue_ida);
65 * For queue allocation
67 static struct kmem_cache *blk_requestq_cachep;
70 * Controlling structure to kblockd
72 static struct workqueue_struct *kblockd_workqueue;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 set_bit(flag, &q->queue_flags);
83 EXPORT_SYMBOL(blk_queue_flag_set);
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 clear_bit(flag, &q->queue_flags);
94 EXPORT_SYMBOL(blk_queue_flag_clear);
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106 return test_and_set_bit(flag, &q->queue_flags);
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
111 static const char *const blk_op_name[] = {
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),
125 REQ_OP_NAME(DRV_OUT),
130 * blk_op_str - Return string XXX in the REQ_OP_XXX.
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".
137 inline const char *blk_op_str(enum req_op op)
139 const char *op_str = "UNKNOWN";
141 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
142 op_str = blk_op_name[op];
146 EXPORT_SYMBOL_GPL(blk_op_str);
148 static const struct {
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" },
166 /* device mapper special case, should not leak out: */
167 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
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" },
173 /* everything else not covered above: */
174 [BLK_STS_IOERR] = { -EIO, "I/O" },
177 blk_status_t errno_to_blk_status(int errno)
181 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
182 if (blk_errors[i].errno == errno)
183 return (__force blk_status_t)i;
186 return BLK_STS_IOERR;
188 EXPORT_SYMBOL_GPL(errno_to_blk_status);
190 int blk_status_to_errno(blk_status_t status)
192 int idx = (__force int)status;
194 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
196 return blk_errors[idx].errno;
198 EXPORT_SYMBOL_GPL(blk_status_to_errno);
200 const char *blk_status_to_str(blk_status_t status)
202 int idx = (__force int)status;
204 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
206 return blk_errors[idx].name;
210 * blk_sync_queue - cancel any pending callbacks on a queue
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
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.
227 void blk_sync_queue(struct request_queue *q)
229 del_timer_sync(&q->timeout);
230 cancel_work_sync(&q->timeout_work);
232 EXPORT_SYMBOL(blk_sync_queue);
235 * blk_set_pm_only - increment pm_only counter
236 * @q: request queue pointer
238 void blk_set_pm_only(struct request_queue *q)
240 atomic_inc(&q->pm_only);
242 EXPORT_SYMBOL_GPL(blk_set_pm_only);
244 void blk_clear_pm_only(struct request_queue *q)
248 pm_only = atomic_dec_return(&q->pm_only);
249 WARN_ON_ONCE(pm_only < 0);
251 wake_up_all(&q->mq_freeze_wq);
253 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
255 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
257 kmem_cache_free(blk_requestq_cachep,
258 container_of(rcu_head, struct request_queue, rcu_head));
261 static void blk_free_queue(struct request_queue *q)
263 percpu_ref_exit(&q->q_usage_counter);
266 blk_stat_remove_callback(q, q->poll_cb);
267 blk_stat_free_callback(q->poll_cb);
269 blk_free_queue_stats(q->stats);
275 ida_free(&blk_queue_ida, q->id);
276 call_rcu(&q->rcu_head, blk_free_queue_rcu);
280 * blk_put_queue - decrement the request_queue refcount
281 * @q: the request_queue structure to decrement the refcount for
283 * Decrements the refcount of the request_queue and free it when the refcount
286 * Context: Can sleep.
288 void blk_put_queue(struct request_queue *q)
291 if (refcount_dec_and_test(&q->refs))
294 EXPORT_SYMBOL(blk_put_queue);
296 void blk_queue_start_drain(struct request_queue *q)
299 * When queue DYING flag is set, we need to block new req
300 * entering queue, so we call blk_freeze_queue_start() to
301 * prevent I/O from crossing blk_queue_enter().
303 blk_freeze_queue_start(q);
305 blk_mq_wake_waiters(q);
306 /* Make blk_queue_enter() reexamine the DYING flag. */
307 wake_up_all(&q->mq_freeze_wq);
311 * blk_queue_enter() - try to increase q->q_usage_counter
312 * @q: request queue pointer
313 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
315 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
317 const bool pm = flags & BLK_MQ_REQ_PM;
319 while (!blk_try_enter_queue(q, pm)) {
320 if (flags & BLK_MQ_REQ_NOWAIT)
324 * read pair of barrier in blk_freeze_queue_start(), we need to
325 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
326 * reading .mq_freeze_depth or queue dying flag, otherwise the
327 * following wait may never return if the two reads are
331 wait_event(q->mq_freeze_wq,
332 (!q->mq_freeze_depth &&
333 blk_pm_resume_queue(pm, q)) ||
335 if (blk_queue_dying(q))
342 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
344 while (!blk_try_enter_queue(q, false)) {
345 struct gendisk *disk = bio->bi_bdev->bd_disk;
347 if (bio->bi_opf & REQ_NOWAIT) {
348 if (test_bit(GD_DEAD, &disk->state))
350 bio_wouldblock_error(bio);
355 * read pair of barrier in blk_freeze_queue_start(), we need to
356 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
357 * reading .mq_freeze_depth or queue dying flag, otherwise the
358 * following wait may never return if the two reads are
362 wait_event(q->mq_freeze_wq,
363 (!q->mq_freeze_depth &&
364 blk_pm_resume_queue(false, q)) ||
365 test_bit(GD_DEAD, &disk->state));
366 if (test_bit(GD_DEAD, &disk->state))
376 void blk_queue_exit(struct request_queue *q)
378 percpu_ref_put(&q->q_usage_counter);
381 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
383 struct request_queue *q =
384 container_of(ref, struct request_queue, q_usage_counter);
386 wake_up_all(&q->mq_freeze_wq);
389 static void blk_rq_timed_out_timer(struct timer_list *t)
391 struct request_queue *q = from_timer(q, t, timeout);
393 kblockd_schedule_work(&q->timeout_work);
396 static void blk_timeout_work(struct work_struct *work)
400 struct request_queue *blk_alloc_queue(int node_id)
402 struct request_queue *q;
404 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
409 q->last_merge = NULL;
411 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
415 q->stats = blk_alloc_queue_stats();
421 atomic_set(&q->nr_active_requests_shared_tags, 0);
423 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
424 INIT_WORK(&q->timeout_work, blk_timeout_work);
425 INIT_LIST_HEAD(&q->icq_list);
427 refcount_set(&q->refs, 1);
428 mutex_init(&q->debugfs_mutex);
429 mutex_init(&q->sysfs_lock);
430 mutex_init(&q->sysfs_dir_lock);
431 spin_lock_init(&q->queue_lock);
433 init_waitqueue_head(&q->mq_freeze_wq);
434 mutex_init(&q->mq_freeze_lock);
437 * Init percpu_ref in atomic mode so that it's faster to shutdown.
438 * See blk_register_queue() for details.
440 if (percpu_ref_init(&q->q_usage_counter,
441 blk_queue_usage_counter_release,
442 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
445 blk_set_default_limits(&q->limits);
446 q->nr_requests = BLKDEV_DEFAULT_RQ;
451 blk_free_queue_stats(q->stats);
453 ida_free(&blk_queue_ida, q->id);
455 kmem_cache_free(blk_requestq_cachep, q);
460 * blk_get_queue - increment the request_queue refcount
461 * @q: the request_queue structure to increment the refcount for
463 * Increment the refcount of the request_queue kobject.
465 * Context: Any context.
467 bool blk_get_queue(struct request_queue *q)
469 if (unlikely(blk_queue_dying(q)))
471 refcount_inc(&q->refs);
474 EXPORT_SYMBOL(blk_get_queue);
476 #ifdef CONFIG_FAIL_MAKE_REQUEST
478 static DECLARE_FAULT_ATTR(fail_make_request);
480 static int __init setup_fail_make_request(char *str)
482 return setup_fault_attr(&fail_make_request, str);
484 __setup("fail_make_request=", setup_fail_make_request);
486 bool should_fail_request(struct block_device *part, unsigned int bytes)
488 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
491 static int __init fail_make_request_debugfs(void)
493 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
494 NULL, &fail_make_request);
496 return PTR_ERR_OR_ZERO(dir);
499 late_initcall(fail_make_request_debugfs);
500 #endif /* CONFIG_FAIL_MAKE_REQUEST */
502 static inline void bio_check_ro(struct bio *bio)
504 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
505 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
507 pr_warn("Trying to write to read-only block-device %pg\n",
509 /* Older lvm-tools actually trigger this */
513 static noinline int should_fail_bio(struct bio *bio)
515 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
519 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
522 * Check whether this bio extends beyond the end of the device or partition.
523 * This may well happen - the kernel calls bread() without checking the size of
524 * the device, e.g., when mounting a file system.
526 static inline int bio_check_eod(struct bio *bio)
528 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
529 unsigned int nr_sectors = bio_sectors(bio);
531 if (nr_sectors && maxsector &&
532 (nr_sectors > maxsector ||
533 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
534 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
535 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
536 current->comm, bio->bi_bdev, bio->bi_opf,
537 bio->bi_iter.bi_sector, nr_sectors, maxsector);
544 * Remap block n of partition p to block n+start(p) of the disk.
546 static int blk_partition_remap(struct bio *bio)
548 struct block_device *p = bio->bi_bdev;
550 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
552 if (bio_sectors(bio)) {
553 bio->bi_iter.bi_sector += p->bd_start_sect;
554 trace_block_bio_remap(bio, p->bd_dev,
555 bio->bi_iter.bi_sector -
558 bio_set_flag(bio, BIO_REMAPPED);
563 * Check write append to a zoned block device.
565 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
568 int nr_sectors = bio_sectors(bio);
570 /* Only applicable to zoned block devices */
571 if (!bdev_is_zoned(bio->bi_bdev))
572 return BLK_STS_NOTSUPP;
574 /* The bio sector must point to the start of a sequential zone */
575 if (bio->bi_iter.bi_sector & (bdev_zone_sectors(bio->bi_bdev) - 1) ||
576 !bio_zone_is_seq(bio))
577 return BLK_STS_IOERR;
580 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
581 * split and could result in non-contiguous sectors being written in
584 if (nr_sectors > q->limits.chunk_sectors)
585 return BLK_STS_IOERR;
587 /* Make sure the BIO is small enough and will not get split */
588 if (nr_sectors > q->limits.max_zone_append_sectors)
589 return BLK_STS_IOERR;
591 bio->bi_opf |= REQ_NOMERGE;
596 static void __submit_bio(struct bio *bio)
598 struct gendisk *disk = bio->bi_bdev->bd_disk;
600 if (unlikely(!blk_crypto_bio_prep(&bio)))
603 if (!disk->fops->submit_bio) {
604 blk_mq_submit_bio(bio);
605 } else if (likely(bio_queue_enter(bio) == 0)) {
606 disk->fops->submit_bio(bio);
607 blk_queue_exit(disk->queue);
612 * The loop in this function may be a bit non-obvious, and so deserves some
615 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
616 * that), so we have a list with a single bio.
617 * - We pretend that we have just taken it off a longer list, so we assign
618 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
619 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
620 * bios through a recursive call to submit_bio_noacct. If it did, we find a
621 * non-NULL value in bio_list and re-enter the loop from the top.
622 * - In this case we really did just take the bio of the top of the list (no
623 * pretending) and so remove it from bio_list, and call into ->submit_bio()
626 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
627 * bio_list_on_stack[1] contains bios that were submitted before the current
628 * ->submit_bio, but that haven't been processed yet.
630 static void __submit_bio_noacct(struct bio *bio)
632 struct bio_list bio_list_on_stack[2];
634 BUG_ON(bio->bi_next);
636 bio_list_init(&bio_list_on_stack[0]);
637 current->bio_list = bio_list_on_stack;
640 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
641 struct bio_list lower, same;
644 * Create a fresh bio_list for all subordinate requests.
646 bio_list_on_stack[1] = bio_list_on_stack[0];
647 bio_list_init(&bio_list_on_stack[0]);
652 * Sort new bios into those for a lower level and those for the
655 bio_list_init(&lower);
656 bio_list_init(&same);
657 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
658 if (q == bdev_get_queue(bio->bi_bdev))
659 bio_list_add(&same, bio);
661 bio_list_add(&lower, bio);
664 * Now assemble so we handle the lowest level first.
666 bio_list_merge(&bio_list_on_stack[0], &lower);
667 bio_list_merge(&bio_list_on_stack[0], &same);
668 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
669 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
671 current->bio_list = NULL;
674 static void __submit_bio_noacct_mq(struct bio *bio)
676 struct bio_list bio_list[2] = { };
678 current->bio_list = bio_list;
682 } while ((bio = bio_list_pop(&bio_list[0])));
684 current->bio_list = NULL;
687 void submit_bio_noacct_nocheck(struct bio *bio)
690 * We only want one ->submit_bio to be active at a time, else stack
691 * usage with stacked devices could be a problem. Use current->bio_list
692 * to collect a list of requests submited by a ->submit_bio method while
693 * it is active, and then process them after it returned.
695 if (current->bio_list)
696 bio_list_add(¤t->bio_list[0], bio);
697 else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
698 __submit_bio_noacct_mq(bio);
700 __submit_bio_noacct(bio);
704 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
705 * @bio: The bio describing the location in memory and on the device.
707 * This is a version of submit_bio() that shall only be used for I/O that is
708 * resubmitted to lower level drivers by stacking block drivers. All file
709 * systems and other upper level users of the block layer should use
710 * submit_bio() instead.
712 void submit_bio_noacct(struct bio *bio)
714 struct block_device *bdev = bio->bi_bdev;
715 struct request_queue *q = bdev_get_queue(bdev);
716 blk_status_t status = BLK_STS_IOERR;
717 struct blk_plug *plug;
721 plug = blk_mq_plug(bio);
722 if (plug && plug->nowait)
723 bio->bi_opf |= REQ_NOWAIT;
726 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
727 * if queue does not support NOWAIT.
729 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
732 if (should_fail_bio(bio))
735 if (!bio_flagged(bio, BIO_REMAPPED)) {
736 if (unlikely(bio_check_eod(bio)))
738 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
743 * Filter flush bio's early so that bio based drivers without flush
744 * support don't have to worry about them.
746 if (op_is_flush(bio->bi_opf) &&
747 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
748 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
749 if (!bio_sectors(bio)) {
755 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
756 bio_clear_polled(bio);
758 switch (bio_op(bio)) {
760 if (!bdev_max_discard_sectors(bdev))
763 case REQ_OP_SECURE_ERASE:
764 if (!bdev_max_secure_erase_sectors(bdev))
767 case REQ_OP_ZONE_APPEND:
768 status = blk_check_zone_append(q, bio);
769 if (status != BLK_STS_OK)
772 case REQ_OP_ZONE_RESET:
773 case REQ_OP_ZONE_OPEN:
774 case REQ_OP_ZONE_CLOSE:
775 case REQ_OP_ZONE_FINISH:
776 if (!bdev_is_zoned(bio->bi_bdev))
779 case REQ_OP_ZONE_RESET_ALL:
780 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
783 case REQ_OP_WRITE_ZEROES:
784 if (!q->limits.max_write_zeroes_sectors)
791 if (blk_throtl_bio(bio))
794 blk_cgroup_bio_start(bio);
795 blkcg_bio_issue_init(bio);
797 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
798 trace_block_bio_queue(bio);
799 /* Now that enqueuing has been traced, we need to trace
800 * completion as well.
802 bio_set_flag(bio, BIO_TRACE_COMPLETION);
804 submit_bio_noacct_nocheck(bio);
808 status = BLK_STS_NOTSUPP;
810 bio->bi_status = status;
813 EXPORT_SYMBOL(submit_bio_noacct);
816 * submit_bio - submit a bio to the block device layer for I/O
817 * @bio: The &struct bio which describes the I/O
819 * submit_bio() is used to submit I/O requests to block devices. It is passed a
820 * fully set up &struct bio that describes the I/O that needs to be done. The
821 * bio will be send to the device described by the bi_bdev field.
823 * The success/failure status of the request, along with notification of
824 * completion, is delivered asynchronously through the ->bi_end_io() callback
825 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
828 void submit_bio(struct bio *bio)
830 if (blkcg_punt_bio_submit(bio))
833 if (bio_op(bio) == REQ_OP_READ) {
834 task_io_account_read(bio->bi_iter.bi_size);
835 count_vm_events(PGPGIN, bio_sectors(bio));
836 } else if (bio_op(bio) == REQ_OP_WRITE) {
837 count_vm_events(PGPGOUT, bio_sectors(bio));
840 submit_bio_noacct(bio);
842 EXPORT_SYMBOL(submit_bio);
845 * bio_poll - poll for BIO completions
846 * @bio: bio to poll for
847 * @iob: batches of IO
848 * @flags: BLK_POLL_* flags that control the behavior
850 * Poll for completions on queue associated with the bio. Returns number of
851 * completed entries found.
853 * Note: the caller must either be the context that submitted @bio, or
854 * be in a RCU critical section to prevent freeing of @bio.
856 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
858 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
859 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
862 if (cookie == BLK_QC_T_NONE ||
863 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
867 * As the requests that require a zone lock are not plugged in the
868 * first place, directly accessing the plug instead of using
869 * blk_mq_plug() should not have any consequences during flushing for
872 blk_flush_plug(current->plug, false);
874 if (bio_queue_enter(bio))
876 if (queue_is_mq(q)) {
877 ret = blk_mq_poll(q, cookie, iob, flags);
879 struct gendisk *disk = q->disk;
881 if (disk && disk->fops->poll_bio)
882 ret = disk->fops->poll_bio(bio, iob, flags);
887 EXPORT_SYMBOL_GPL(bio_poll);
890 * Helper to implement file_operations.iopoll. Requires the bio to be stored
891 * in iocb->private, and cleared before freeing the bio.
893 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
900 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
901 * point to a freshly allocated bio at this point. If that happens
902 * we have a few cases to consider:
904 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
905 * simply nothing in this case
906 * 2) the bio points to a not poll enabled device. bio_poll will catch
908 * 3) the bio points to a poll capable device, including but not
909 * limited to the one that the original bio pointed to. In this
910 * case we will call into the actual poll method and poll for I/O,
911 * even if we don't need to, but it won't cause harm either.
913 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
914 * is still allocated. Because partitions hold a reference to the whole
915 * device bdev and thus disk, the disk is also still valid. Grabbing
916 * a reference to the queue in bio_poll() ensures the hctxs and requests
917 * are still valid as well.
920 bio = READ_ONCE(kiocb->private);
921 if (bio && bio->bi_bdev)
922 ret = bio_poll(bio, iob, flags);
927 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
929 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
933 stamp = READ_ONCE(part->bd_stamp);
934 if (unlikely(time_after(now, stamp))) {
935 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
936 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
938 if (part->bd_partno) {
939 part = bdev_whole(part);
944 unsigned long bdev_start_io_acct(struct block_device *bdev,
945 unsigned int sectors, enum req_op op,
946 unsigned long start_time)
948 const int sgrp = op_stat_group(op);
951 update_io_ticks(bdev, start_time, false);
952 part_stat_inc(bdev, ios[sgrp]);
953 part_stat_add(bdev, sectors[sgrp], sectors);
954 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
959 EXPORT_SYMBOL(bdev_start_io_acct);
962 * bio_start_io_acct - start I/O accounting for bio based drivers
963 * @bio: bio to start account for
965 * Returns the start time that should be passed back to bio_end_io_acct().
967 unsigned long bio_start_io_acct(struct bio *bio)
969 return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
970 bio_op(bio), jiffies);
972 EXPORT_SYMBOL_GPL(bio_start_io_acct);
974 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
975 unsigned long start_time)
977 const int sgrp = op_stat_group(op);
978 unsigned long now = READ_ONCE(jiffies);
979 unsigned long duration = now - start_time;
982 update_io_ticks(bdev, now, true);
983 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
984 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
987 EXPORT_SYMBOL(bdev_end_io_acct);
989 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
990 struct block_device *orig_bdev)
992 bdev_end_io_acct(orig_bdev, bio_op(bio), start_time);
994 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
997 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
998 * @q : the queue of the device being checked
1001 * Check if underlying low-level drivers of a device are busy.
1002 * If the drivers want to export their busy state, they must set own
1003 * exporting function using blk_queue_lld_busy() first.
1005 * Basically, this function is used only by request stacking drivers
1006 * to stop dispatching requests to underlying devices when underlying
1007 * devices are busy. This behavior helps more I/O merging on the queue
1008 * of the request stacking driver and prevents I/O throughput regression
1009 * on burst I/O load.
1012 * 0 - Not busy (The request stacking driver should dispatch request)
1013 * 1 - Busy (The request stacking driver should stop dispatching request)
1015 int blk_lld_busy(struct request_queue *q)
1017 if (queue_is_mq(q) && q->mq_ops->busy)
1018 return q->mq_ops->busy(q);
1022 EXPORT_SYMBOL_GPL(blk_lld_busy);
1024 int kblockd_schedule_work(struct work_struct *work)
1026 return queue_work(kblockd_workqueue, work);
1028 EXPORT_SYMBOL(kblockd_schedule_work);
1030 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1031 unsigned long delay)
1033 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1035 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1037 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1039 struct task_struct *tsk = current;
1042 * If this is a nested plug, don't actually assign it.
1047 plug->mq_list = NULL;
1048 plug->cached_rq = NULL;
1049 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1051 plug->multiple_queues = false;
1052 plug->has_elevator = false;
1053 plug->nowait = false;
1054 INIT_LIST_HEAD(&plug->cb_list);
1057 * Store ordering should not be needed here, since a potential
1058 * preempt will imply a full memory barrier
1064 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1065 * @plug: The &struct blk_plug that needs to be initialized
1068 * blk_start_plug() indicates to the block layer an intent by the caller
1069 * to submit multiple I/O requests in a batch. The block layer may use
1070 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1071 * is called. However, the block layer may choose to submit requests
1072 * before a call to blk_finish_plug() if the number of queued I/Os
1073 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1074 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1075 * the task schedules (see below).
1077 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1078 * pending I/O should the task end up blocking between blk_start_plug() and
1079 * blk_finish_plug(). This is important from a performance perspective, but
1080 * also ensures that we don't deadlock. For instance, if the task is blocking
1081 * for a memory allocation, memory reclaim could end up wanting to free a
1082 * page belonging to that request that is currently residing in our private
1083 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1084 * this kind of deadlock.
1086 void blk_start_plug(struct blk_plug *plug)
1088 blk_start_plug_nr_ios(plug, 1);
1090 EXPORT_SYMBOL(blk_start_plug);
1092 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1094 LIST_HEAD(callbacks);
1096 while (!list_empty(&plug->cb_list)) {
1097 list_splice_init(&plug->cb_list, &callbacks);
1099 while (!list_empty(&callbacks)) {
1100 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1103 list_del(&cb->list);
1104 cb->callback(cb, from_schedule);
1109 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1112 struct blk_plug *plug = current->plug;
1113 struct blk_plug_cb *cb;
1118 list_for_each_entry(cb, &plug->cb_list, list)
1119 if (cb->callback == unplug && cb->data == data)
1122 /* Not currently on the callback list */
1123 BUG_ON(size < sizeof(*cb));
1124 cb = kzalloc(size, GFP_ATOMIC);
1127 cb->callback = unplug;
1128 list_add(&cb->list, &plug->cb_list);
1132 EXPORT_SYMBOL(blk_check_plugged);
1134 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1136 if (!list_empty(&plug->cb_list))
1137 flush_plug_callbacks(plug, from_schedule);
1138 if (!rq_list_empty(plug->mq_list))
1139 blk_mq_flush_plug_list(plug, from_schedule);
1141 * Unconditionally flush out cached requests, even if the unplug
1142 * event came from schedule. Since we know hold references to the
1143 * queue for cached requests, we don't want a blocked task holding
1144 * up a queue freeze/quiesce event.
1146 if (unlikely(!rq_list_empty(plug->cached_rq)))
1147 blk_mq_free_plug_rqs(plug);
1151 * blk_finish_plug - mark the end of a batch of submitted I/O
1152 * @plug: The &struct blk_plug passed to blk_start_plug()
1155 * Indicate that a batch of I/O submissions is complete. This function
1156 * must be paired with an initial call to blk_start_plug(). The intent
1157 * is to allow the block layer to optimize I/O submission. See the
1158 * documentation for blk_start_plug() for more information.
1160 void blk_finish_plug(struct blk_plug *plug)
1162 if (plug == current->plug) {
1163 __blk_flush_plug(plug, false);
1164 current->plug = NULL;
1167 EXPORT_SYMBOL(blk_finish_plug);
1169 void blk_io_schedule(void)
1171 /* Prevent hang_check timer from firing at us during very long I/O */
1172 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1175 io_schedule_timeout(timeout);
1179 EXPORT_SYMBOL_GPL(blk_io_schedule);
1181 int __init blk_dev_init(void)
1183 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1184 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1185 sizeof_field(struct request, cmd_flags));
1186 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1187 sizeof_field(struct bio, bi_opf));
1189 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1190 kblockd_workqueue = alloc_workqueue("kblockd",
1191 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1192 if (!kblockd_workqueue)
1193 panic("Failed to create kblockd\n");
1195 blk_requestq_cachep = kmem_cache_create("request_queue",
1196 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1198 blk_debugfs_root = debugfs_create_dir("block", NULL);