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/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
42 #include <linux/part_stat.h>
43 #include <linux/sched/sysctl.h>
44 #include <linux/blk-crypto.h>
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/block.h>
50 #include "blk-mq-sched.h"
52 #include "blk-throttle.h"
53 #include "blk-rq-qos.h"
55 struct dentry *blk_debugfs_root;
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
62 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
64 DEFINE_IDA(blk_queue_ida);
67 * For queue allocation
69 struct kmem_cache *blk_requestq_cachep;
70 struct kmem_cache *blk_requestq_srcu_cachep;
73 * Controlling structure to kblockd
75 static struct workqueue_struct *kblockd_workqueue;
78 * blk_queue_flag_set - atomically set a queue flag
79 * @flag: flag to be set
82 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
84 set_bit(flag, &q->queue_flags);
86 EXPORT_SYMBOL(blk_queue_flag_set);
89 * blk_queue_flag_clear - atomically clear a queue flag
90 * @flag: flag to be cleared
93 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
95 clear_bit(flag, &q->queue_flags);
97 EXPORT_SYMBOL(blk_queue_flag_clear);
100 * blk_queue_flag_test_and_set - atomically test and set a queue flag
101 * @flag: flag to be set
104 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
105 * the flag was already set.
107 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
109 return test_and_set_bit(flag, &q->queue_flags);
111 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
113 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
114 static const char *const blk_op_name[] = {
118 REQ_OP_NAME(DISCARD),
119 REQ_OP_NAME(SECURE_ERASE),
120 REQ_OP_NAME(ZONE_RESET),
121 REQ_OP_NAME(ZONE_RESET_ALL),
122 REQ_OP_NAME(ZONE_OPEN),
123 REQ_OP_NAME(ZONE_CLOSE),
124 REQ_OP_NAME(ZONE_FINISH),
125 REQ_OP_NAME(ZONE_APPEND),
126 REQ_OP_NAME(WRITE_SAME),
127 REQ_OP_NAME(WRITE_ZEROES),
129 REQ_OP_NAME(DRV_OUT),
134 * blk_op_str - Return string XXX in the REQ_OP_XXX.
137 * Description: Centralize block layer function to convert REQ_OP_XXX into
138 * string format. Useful in the debugging and tracing bio or request. For
139 * invalid REQ_OP_XXX it returns string "UNKNOWN".
141 inline const char *blk_op_str(unsigned int op)
143 const char *op_str = "UNKNOWN";
145 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
146 op_str = blk_op_name[op];
150 EXPORT_SYMBOL_GPL(blk_op_str);
152 static const struct {
156 [BLK_STS_OK] = { 0, "" },
157 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
158 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
159 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
160 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
161 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
162 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
163 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
164 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
165 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
166 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
167 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
169 /* device mapper special case, should not leak out: */
170 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
172 /* zone device specific errors */
173 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
174 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
176 /* everything else not covered above: */
177 [BLK_STS_IOERR] = { -EIO, "I/O" },
180 blk_status_t errno_to_blk_status(int errno)
184 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
185 if (blk_errors[i].errno == errno)
186 return (__force blk_status_t)i;
189 return BLK_STS_IOERR;
191 EXPORT_SYMBOL_GPL(errno_to_blk_status);
193 int blk_status_to_errno(blk_status_t status)
195 int idx = (__force int)status;
197 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
199 return blk_errors[idx].errno;
201 EXPORT_SYMBOL_GPL(blk_status_to_errno);
203 const char *blk_status_to_str(blk_status_t status)
205 int idx = (__force int)status;
207 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
209 return blk_errors[idx].name;
213 * blk_sync_queue - cancel any pending callbacks on a queue
217 * The block layer may perform asynchronous callback activity
218 * on a queue, such as calling the unplug function after a timeout.
219 * A block device may call blk_sync_queue to ensure that any
220 * such activity is cancelled, thus allowing it to release resources
221 * that the callbacks might use. The caller must already have made sure
222 * that its ->submit_bio will not re-add plugging prior to calling
225 * This function does not cancel any asynchronous activity arising
226 * out of elevator or throttling code. That would require elevator_exit()
227 * and blkcg_exit_queue() to be called with queue lock initialized.
230 void blk_sync_queue(struct request_queue *q)
232 del_timer_sync(&q->timeout);
233 cancel_work_sync(&q->timeout_work);
235 EXPORT_SYMBOL(blk_sync_queue);
238 * blk_set_pm_only - increment pm_only counter
239 * @q: request queue pointer
241 void blk_set_pm_only(struct request_queue *q)
243 atomic_inc(&q->pm_only);
245 EXPORT_SYMBOL_GPL(blk_set_pm_only);
247 void blk_clear_pm_only(struct request_queue *q)
251 pm_only = atomic_dec_return(&q->pm_only);
252 WARN_ON_ONCE(pm_only < 0);
254 wake_up_all(&q->mq_freeze_wq);
256 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
259 * blk_put_queue - decrement the request_queue refcount
260 * @q: the request_queue structure to decrement the refcount for
262 * Decrements the refcount of the request_queue kobject. When this reaches 0
263 * we'll have blk_release_queue() called.
265 * Context: Any context, but the last reference must not be dropped from
268 void blk_put_queue(struct request_queue *q)
270 kobject_put(&q->kobj);
272 EXPORT_SYMBOL(blk_put_queue);
274 void blk_queue_start_drain(struct request_queue *q)
277 * When queue DYING flag is set, we need to block new req
278 * entering queue, so we call blk_freeze_queue_start() to
279 * prevent I/O from crossing blk_queue_enter().
281 blk_freeze_queue_start(q);
283 blk_mq_wake_waiters(q);
284 /* Make blk_queue_enter() reexamine the DYING flag. */
285 wake_up_all(&q->mq_freeze_wq);
289 * blk_cleanup_queue - shutdown a request queue
290 * @q: request queue to shutdown
292 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
293 * put it. All future requests will be failed immediately with -ENODEV.
297 void blk_cleanup_queue(struct request_queue *q)
299 /* cannot be called from atomic context */
302 WARN_ON_ONCE(blk_queue_registered(q));
304 /* mark @q DYING, no new request or merges will be allowed afterwards */
305 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
306 blk_queue_start_drain(q);
308 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
309 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
312 * Drain all requests queued before DYING marking. Set DEAD flag to
313 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
314 * after draining finished.
318 /* cleanup rq qos structures for queue without disk */
321 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
324 if (queue_is_mq(q)) {
325 blk_mq_cancel_work_sync(q);
326 blk_mq_exit_queue(q);
330 * In theory, request pool of sched_tags belongs to request queue.
331 * However, the current implementation requires tag_set for freeing
332 * requests, so free the pool now.
334 * Queue has become frozen, there can't be any in-queue requests, so
335 * it is safe to free requests now.
337 mutex_lock(&q->sysfs_lock);
339 blk_mq_sched_free_rqs(q);
340 mutex_unlock(&q->sysfs_lock);
342 percpu_ref_exit(&q->q_usage_counter);
344 /* @q is and will stay empty, shutdown and put */
347 EXPORT_SYMBOL(blk_cleanup_queue);
350 * blk_queue_enter() - try to increase q->q_usage_counter
351 * @q: request queue pointer
352 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
354 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
356 const bool pm = flags & BLK_MQ_REQ_PM;
358 while (!blk_try_enter_queue(q, pm)) {
359 if (flags & BLK_MQ_REQ_NOWAIT)
363 * read pair of barrier in blk_freeze_queue_start(), we need to
364 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
365 * reading .mq_freeze_depth or queue dying flag, otherwise the
366 * following wait may never return if the two reads are
370 wait_event(q->mq_freeze_wq,
371 (!q->mq_freeze_depth &&
372 blk_pm_resume_queue(pm, q)) ||
374 if (blk_queue_dying(q))
381 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
383 while (!blk_try_enter_queue(q, false)) {
384 struct gendisk *disk = bio->bi_bdev->bd_disk;
386 if (bio->bi_opf & REQ_NOWAIT) {
387 if (test_bit(GD_DEAD, &disk->state))
389 bio_wouldblock_error(bio);
394 * read pair of barrier in blk_freeze_queue_start(), we need to
395 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
396 * reading .mq_freeze_depth or queue dying flag, otherwise the
397 * following wait may never return if the two reads are
401 wait_event(q->mq_freeze_wq,
402 (!q->mq_freeze_depth &&
403 blk_pm_resume_queue(false, q)) ||
404 test_bit(GD_DEAD, &disk->state));
405 if (test_bit(GD_DEAD, &disk->state))
415 void blk_queue_exit(struct request_queue *q)
417 percpu_ref_put(&q->q_usage_counter);
420 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
422 struct request_queue *q =
423 container_of(ref, struct request_queue, q_usage_counter);
425 wake_up_all(&q->mq_freeze_wq);
428 static void blk_rq_timed_out_timer(struct timer_list *t)
430 struct request_queue *q = from_timer(q, t, timeout);
432 kblockd_schedule_work(&q->timeout_work);
435 static void blk_timeout_work(struct work_struct *work)
439 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
441 struct request_queue *q;
444 q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
445 GFP_KERNEL | __GFP_ZERO, node_id);
450 blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
451 if (init_srcu_struct(q->srcu) != 0)
455 q->last_merge = NULL;
457 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
461 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
465 q->stats = blk_alloc_queue_stats();
471 atomic_set(&q->nr_active_requests_shared_tags, 0);
473 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
474 INIT_WORK(&q->timeout_work, blk_timeout_work);
475 INIT_LIST_HEAD(&q->icq_list);
476 #ifdef CONFIG_BLK_CGROUP
477 INIT_LIST_HEAD(&q->blkg_list);
480 kobject_init(&q->kobj, &blk_queue_ktype);
482 mutex_init(&q->debugfs_mutex);
483 mutex_init(&q->sysfs_lock);
484 mutex_init(&q->sysfs_dir_lock);
485 spin_lock_init(&q->queue_lock);
487 init_waitqueue_head(&q->mq_freeze_wq);
488 mutex_init(&q->mq_freeze_lock);
491 * Init percpu_ref in atomic mode so that it's faster to shutdown.
492 * See blk_register_queue() for details.
494 if (percpu_ref_init(&q->q_usage_counter,
495 blk_queue_usage_counter_release,
496 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
499 if (blkcg_init_queue(q))
502 blk_queue_dma_alignment(q, 511);
503 blk_set_default_limits(&q->limits);
504 q->nr_requests = BLKDEV_DEFAULT_RQ;
509 percpu_ref_exit(&q->q_usage_counter);
511 blk_free_queue_stats(q->stats);
513 bioset_exit(&q->bio_split);
515 ida_simple_remove(&blk_queue_ida, q->id);
518 cleanup_srcu_struct(q->srcu);
520 kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
525 * blk_get_queue - increment the request_queue refcount
526 * @q: the request_queue structure to increment the refcount for
528 * Increment the refcount of the request_queue kobject.
530 * Context: Any context.
532 bool blk_get_queue(struct request_queue *q)
534 if (likely(!blk_queue_dying(q))) {
541 EXPORT_SYMBOL(blk_get_queue);
543 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
545 char b[BDEVNAME_SIZE];
547 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
548 "%s: rw=%d, want=%llu, limit=%llu\n",
550 bio_devname(bio, b), bio->bi_opf,
551 bio_end_sector(bio), maxsector);
554 #ifdef CONFIG_FAIL_MAKE_REQUEST
556 static DECLARE_FAULT_ATTR(fail_make_request);
558 static int __init setup_fail_make_request(char *str)
560 return setup_fault_attr(&fail_make_request, str);
562 __setup("fail_make_request=", setup_fail_make_request);
564 bool should_fail_request(struct block_device *part, unsigned int bytes)
566 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
569 static int __init fail_make_request_debugfs(void)
571 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
572 NULL, &fail_make_request);
574 return PTR_ERR_OR_ZERO(dir);
577 late_initcall(fail_make_request_debugfs);
578 #endif /* CONFIG_FAIL_MAKE_REQUEST */
580 static inline bool bio_check_ro(struct bio *bio)
582 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
583 char b[BDEVNAME_SIZE];
585 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
589 "Trying to write to read-only block-device %s (partno %d)\n",
590 bio_devname(bio, b), bio->bi_bdev->bd_partno);
591 /* Older lvm-tools actually trigger this */
598 static noinline int should_fail_bio(struct bio *bio)
600 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
604 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
607 * Check whether this bio extends beyond the end of the device or partition.
608 * This may well happen - the kernel calls bread() without checking the size of
609 * the device, e.g., when mounting a file system.
611 static inline int bio_check_eod(struct bio *bio)
613 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
614 unsigned int nr_sectors = bio_sectors(bio);
616 if (nr_sectors && maxsector &&
617 (nr_sectors > maxsector ||
618 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
619 handle_bad_sector(bio, maxsector);
626 * Remap block n of partition p to block n+start(p) of the disk.
628 static int blk_partition_remap(struct bio *bio)
630 struct block_device *p = bio->bi_bdev;
632 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
634 if (bio_sectors(bio)) {
635 bio->bi_iter.bi_sector += p->bd_start_sect;
636 trace_block_bio_remap(bio, p->bd_dev,
637 bio->bi_iter.bi_sector -
640 bio_set_flag(bio, BIO_REMAPPED);
645 * Check write append to a zoned block device.
647 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
650 sector_t pos = bio->bi_iter.bi_sector;
651 int nr_sectors = bio_sectors(bio);
653 /* Only applicable to zoned block devices */
654 if (!blk_queue_is_zoned(q))
655 return BLK_STS_NOTSUPP;
657 /* The bio sector must point to the start of a sequential zone */
658 if (pos & (blk_queue_zone_sectors(q) - 1) ||
659 !blk_queue_zone_is_seq(q, pos))
660 return BLK_STS_IOERR;
663 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
664 * split and could result in non-contiguous sectors being written in
667 if (nr_sectors > q->limits.chunk_sectors)
668 return BLK_STS_IOERR;
670 /* Make sure the BIO is small enough and will not get split */
671 if (nr_sectors > q->limits.max_zone_append_sectors)
672 return BLK_STS_IOERR;
674 bio->bi_opf |= REQ_NOMERGE;
679 noinline_for_stack bool submit_bio_checks(struct bio *bio)
681 struct block_device *bdev = bio->bi_bdev;
682 struct request_queue *q = bdev_get_queue(bdev);
683 blk_status_t status = BLK_STS_IOERR;
684 struct blk_plug *plug;
688 plug = blk_mq_plug(q, bio);
689 if (plug && plug->nowait)
690 bio->bi_opf |= REQ_NOWAIT;
693 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
694 * if queue does not support NOWAIT.
696 if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
699 if (should_fail_bio(bio))
701 if (unlikely(bio_check_ro(bio)))
703 if (!bio_flagged(bio, BIO_REMAPPED)) {
704 if (unlikely(bio_check_eod(bio)))
706 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
711 * Filter flush bio's early so that bio based drivers without flush
712 * support don't have to worry about them.
714 if (op_is_flush(bio->bi_opf) &&
715 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
716 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
717 if (!bio_sectors(bio)) {
723 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
724 bio_clear_polled(bio);
726 switch (bio_op(bio)) {
728 if (!blk_queue_discard(q))
731 case REQ_OP_SECURE_ERASE:
732 if (!blk_queue_secure_erase(q))
735 case REQ_OP_WRITE_SAME:
736 if (!q->limits.max_write_same_sectors)
739 case REQ_OP_ZONE_APPEND:
740 status = blk_check_zone_append(q, bio);
741 if (status != BLK_STS_OK)
744 case REQ_OP_ZONE_RESET:
745 case REQ_OP_ZONE_OPEN:
746 case REQ_OP_ZONE_CLOSE:
747 case REQ_OP_ZONE_FINISH:
748 if (!blk_queue_is_zoned(q))
751 case REQ_OP_ZONE_RESET_ALL:
752 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
755 case REQ_OP_WRITE_ZEROES:
756 if (!q->limits.max_write_zeroes_sectors)
763 if (blk_throtl_bio(bio))
766 blk_cgroup_bio_start(bio);
767 blkcg_bio_issue_init(bio);
769 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
770 trace_block_bio_queue(bio);
771 /* Now that enqueuing has been traced, we need to trace
772 * completion as well.
774 bio_set_flag(bio, BIO_TRACE_COMPLETION);
779 status = BLK_STS_NOTSUPP;
781 bio->bi_status = status;
786 static void __submit_bio_fops(struct gendisk *disk, struct bio *bio)
788 if (blk_crypto_bio_prep(&bio)) {
789 if (likely(bio_queue_enter(bio) == 0)) {
790 disk->fops->submit_bio(bio);
791 blk_queue_exit(disk->queue);
796 static void __submit_bio(struct bio *bio)
798 struct gendisk *disk = bio->bi_bdev->bd_disk;
800 if (unlikely(!submit_bio_checks(bio)))
803 if (!disk->fops->submit_bio)
804 blk_mq_submit_bio(bio);
806 __submit_bio_fops(disk, bio);
810 * The loop in this function may be a bit non-obvious, and so deserves some
813 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
814 * that), so we have a list with a single bio.
815 * - We pretend that we have just taken it off a longer list, so we assign
816 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
817 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
818 * bios through a recursive call to submit_bio_noacct. If it did, we find a
819 * non-NULL value in bio_list and re-enter the loop from the top.
820 * - In this case we really did just take the bio of the top of the list (no
821 * pretending) and so remove it from bio_list, and call into ->submit_bio()
824 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
825 * bio_list_on_stack[1] contains bios that were submitted before the current
826 * ->submit_bio_bio, but that haven't been processed yet.
828 static void __submit_bio_noacct(struct bio *bio)
830 struct bio_list bio_list_on_stack[2];
832 BUG_ON(bio->bi_next);
834 bio_list_init(&bio_list_on_stack[0]);
835 current->bio_list = bio_list_on_stack;
838 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
839 struct bio_list lower, same;
842 * Create a fresh bio_list for all subordinate requests.
844 bio_list_on_stack[1] = bio_list_on_stack[0];
845 bio_list_init(&bio_list_on_stack[0]);
850 * Sort new bios into those for a lower level and those for the
853 bio_list_init(&lower);
854 bio_list_init(&same);
855 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
856 if (q == bdev_get_queue(bio->bi_bdev))
857 bio_list_add(&same, bio);
859 bio_list_add(&lower, bio);
862 * Now assemble so we handle the lowest level first.
864 bio_list_merge(&bio_list_on_stack[0], &lower);
865 bio_list_merge(&bio_list_on_stack[0], &same);
866 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
867 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
869 current->bio_list = NULL;
872 static void __submit_bio_noacct_mq(struct bio *bio)
874 struct bio_list bio_list[2] = { };
876 current->bio_list = bio_list;
880 } while ((bio = bio_list_pop(&bio_list[0])));
882 current->bio_list = NULL;
886 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
887 * @bio: The bio describing the location in memory and on the device.
889 * This is a version of submit_bio() that shall only be used for I/O that is
890 * resubmitted to lower level drivers by stacking block drivers. All file
891 * systems and other upper level users of the block layer should use
892 * submit_bio() instead.
894 void submit_bio_noacct(struct bio *bio)
897 * We only want one ->submit_bio to be active at a time, else stack
898 * usage with stacked devices could be a problem. Use current->bio_list
899 * to collect a list of requests submited by a ->submit_bio method while
900 * it is active, and then process them after it returned.
902 if (current->bio_list)
903 bio_list_add(¤t->bio_list[0], bio);
904 else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
905 __submit_bio_noacct_mq(bio);
907 __submit_bio_noacct(bio);
909 EXPORT_SYMBOL(submit_bio_noacct);
912 * submit_bio - submit a bio to the block device layer for I/O
913 * @bio: The &struct bio which describes the I/O
915 * submit_bio() is used to submit I/O requests to block devices. It is passed a
916 * fully set up &struct bio that describes the I/O that needs to be done. The
917 * bio will be send to the device described by the bi_bdev field.
919 * The success/failure status of the request, along with notification of
920 * completion, is delivered asynchronously through the ->bi_end_io() callback
921 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
924 void submit_bio(struct bio *bio)
926 if (blkcg_punt_bio_submit(bio))
930 * If it's a regular read/write or a barrier with data attached,
931 * go through the normal accounting stuff before submission.
933 if (bio_has_data(bio)) {
936 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
937 count = queue_logical_block_size(
938 bdev_get_queue(bio->bi_bdev)) >> 9;
940 count = bio_sectors(bio);
942 if (op_is_write(bio_op(bio))) {
943 count_vm_events(PGPGOUT, count);
945 task_io_account_read(bio->bi_iter.bi_size);
946 count_vm_events(PGPGIN, count);
951 * If we're reading data that is part of the userspace workingset, count
952 * submission time as memory stall. When the device is congested, or
953 * the submitting cgroup IO-throttled, submission can be a significant
954 * part of overall IO time.
956 if (unlikely(bio_op(bio) == REQ_OP_READ &&
957 bio_flagged(bio, BIO_WORKINGSET))) {
958 unsigned long pflags;
960 psi_memstall_enter(&pflags);
961 submit_bio_noacct(bio);
962 psi_memstall_leave(&pflags);
966 submit_bio_noacct(bio);
968 EXPORT_SYMBOL(submit_bio);
971 * bio_poll - poll for BIO completions
972 * @bio: bio to poll for
973 * @iob: batches of IO
974 * @flags: BLK_POLL_* flags that control the behavior
976 * Poll for completions on queue associated with the bio. Returns number of
977 * completed entries found.
979 * Note: the caller must either be the context that submitted @bio, or
980 * be in a RCU critical section to prevent freeing of @bio.
982 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
984 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
985 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
988 if (cookie == BLK_QC_T_NONE ||
989 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
993 blk_flush_plug(current->plug, false);
995 if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT))
997 if (WARN_ON_ONCE(!queue_is_mq(q)))
998 ret = 0; /* not yet implemented, should not happen */
1000 ret = blk_mq_poll(q, cookie, iob, flags);
1004 EXPORT_SYMBOL_GPL(bio_poll);
1007 * Helper to implement file_operations.iopoll. Requires the bio to be stored
1008 * in iocb->private, and cleared before freeing the bio.
1010 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
1017 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
1018 * point to a freshly allocated bio at this point. If that happens
1019 * we have a few cases to consider:
1021 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
1022 * simply nothing in this case
1023 * 2) the bio points to a not poll enabled device. bio_poll will catch
1025 * 3) the bio points to a poll capable device, including but not
1026 * limited to the one that the original bio pointed to. In this
1027 * case we will call into the actual poll method and poll for I/O,
1028 * even if we don't need to, but it won't cause harm either.
1030 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
1031 * is still allocated. Because partitions hold a reference to the whole
1032 * device bdev and thus disk, the disk is also still valid. Grabbing
1033 * a reference to the queue in bio_poll() ensures the hctxs and requests
1034 * are still valid as well.
1037 bio = READ_ONCE(kiocb->private);
1038 if (bio && bio->bi_bdev)
1039 ret = bio_poll(bio, iob, flags);
1044 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1046 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1048 unsigned long stamp;
1050 stamp = READ_ONCE(part->bd_stamp);
1051 if (unlikely(time_after(now, stamp))) {
1052 if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1053 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1055 if (part->bd_partno) {
1056 part = bdev_whole(part);
1061 static unsigned long __part_start_io_acct(struct block_device *part,
1062 unsigned int sectors, unsigned int op,
1063 unsigned long start_time)
1065 const int sgrp = op_stat_group(op);
1068 update_io_ticks(part, start_time, false);
1069 part_stat_inc(part, ios[sgrp]);
1070 part_stat_add(part, sectors[sgrp], sectors);
1071 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1078 * bio_start_io_acct_time - start I/O accounting for bio based drivers
1079 * @bio: bio to start account for
1080 * @start_time: start time that should be passed back to bio_end_io_acct().
1082 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
1084 __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1085 bio_op(bio), start_time);
1087 EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
1090 * bio_start_io_acct - start I/O accounting for bio based drivers
1091 * @bio: bio to start account for
1093 * Returns the start time that should be passed back to bio_end_io_acct().
1095 unsigned long bio_start_io_acct(struct bio *bio)
1097 return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1098 bio_op(bio), jiffies);
1100 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1102 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1105 return __part_start_io_acct(disk->part0, sectors, op, jiffies);
1107 EXPORT_SYMBOL(disk_start_io_acct);
1109 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1110 unsigned long start_time)
1112 const int sgrp = op_stat_group(op);
1113 unsigned long now = READ_ONCE(jiffies);
1114 unsigned long duration = now - start_time;
1117 update_io_ticks(part, now, true);
1118 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1119 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1123 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1124 struct block_device *orig_bdev)
1126 __part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1128 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1130 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1131 unsigned long start_time)
1133 __part_end_io_acct(disk->part0, op, start_time);
1135 EXPORT_SYMBOL(disk_end_io_acct);
1138 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1139 * @q : the queue of the device being checked
1142 * Check if underlying low-level drivers of a device are busy.
1143 * If the drivers want to export their busy state, they must set own
1144 * exporting function using blk_queue_lld_busy() first.
1146 * Basically, this function is used only by request stacking drivers
1147 * to stop dispatching requests to underlying devices when underlying
1148 * devices are busy. This behavior helps more I/O merging on the queue
1149 * of the request stacking driver and prevents I/O throughput regression
1150 * on burst I/O load.
1153 * 0 - Not busy (The request stacking driver should dispatch request)
1154 * 1 - Busy (The request stacking driver should stop dispatching request)
1156 int blk_lld_busy(struct request_queue *q)
1158 if (queue_is_mq(q) && q->mq_ops->busy)
1159 return q->mq_ops->busy(q);
1163 EXPORT_SYMBOL_GPL(blk_lld_busy);
1165 int kblockd_schedule_work(struct work_struct *work)
1167 return queue_work(kblockd_workqueue, work);
1169 EXPORT_SYMBOL(kblockd_schedule_work);
1171 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1172 unsigned long delay)
1174 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1176 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1178 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1180 struct task_struct *tsk = current;
1183 * If this is a nested plug, don't actually assign it.
1188 plug->mq_list = NULL;
1189 plug->cached_rq = NULL;
1190 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1192 plug->multiple_queues = false;
1193 plug->has_elevator = false;
1194 plug->nowait = false;
1195 INIT_LIST_HEAD(&plug->cb_list);
1198 * Store ordering should not be needed here, since a potential
1199 * preempt will imply a full memory barrier
1205 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1206 * @plug: The &struct blk_plug that needs to be initialized
1209 * blk_start_plug() indicates to the block layer an intent by the caller
1210 * to submit multiple I/O requests in a batch. The block layer may use
1211 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1212 * is called. However, the block layer may choose to submit requests
1213 * before a call to blk_finish_plug() if the number of queued I/Os
1214 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1215 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1216 * the task schedules (see below).
1218 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1219 * pending I/O should the task end up blocking between blk_start_plug() and
1220 * blk_finish_plug(). This is important from a performance perspective, but
1221 * also ensures that we don't deadlock. For instance, if the task is blocking
1222 * for a memory allocation, memory reclaim could end up wanting to free a
1223 * page belonging to that request that is currently residing in our private
1224 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1225 * this kind of deadlock.
1227 void blk_start_plug(struct blk_plug *plug)
1229 blk_start_plug_nr_ios(plug, 1);
1231 EXPORT_SYMBOL(blk_start_plug);
1233 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1235 LIST_HEAD(callbacks);
1237 while (!list_empty(&plug->cb_list)) {
1238 list_splice_init(&plug->cb_list, &callbacks);
1240 while (!list_empty(&callbacks)) {
1241 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1244 list_del(&cb->list);
1245 cb->callback(cb, from_schedule);
1250 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1253 struct blk_plug *plug = current->plug;
1254 struct blk_plug_cb *cb;
1259 list_for_each_entry(cb, &plug->cb_list, list)
1260 if (cb->callback == unplug && cb->data == data)
1263 /* Not currently on the callback list */
1264 BUG_ON(size < sizeof(*cb));
1265 cb = kzalloc(size, GFP_ATOMIC);
1268 cb->callback = unplug;
1269 list_add(&cb->list, &plug->cb_list);
1273 EXPORT_SYMBOL(blk_check_plugged);
1275 void blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1277 if (!list_empty(&plug->cb_list))
1278 flush_plug_callbacks(plug, from_schedule);
1279 if (!rq_list_empty(plug->mq_list))
1280 blk_mq_flush_plug_list(plug, from_schedule);
1282 * Unconditionally flush out cached requests, even if the unplug
1283 * event came from schedule. Since we know hold references to the
1284 * queue for cached requests, we don't want a blocked task holding
1285 * up a queue freeze/quiesce event.
1287 if (unlikely(!rq_list_empty(plug->cached_rq)))
1288 blk_mq_free_plug_rqs(plug);
1292 * blk_finish_plug - mark the end of a batch of submitted I/O
1293 * @plug: The &struct blk_plug passed to blk_start_plug()
1296 * Indicate that a batch of I/O submissions is complete. This function
1297 * must be paired with an initial call to blk_start_plug(). The intent
1298 * is to allow the block layer to optimize I/O submission. See the
1299 * documentation for blk_start_plug() for more information.
1301 void blk_finish_plug(struct blk_plug *plug)
1303 if (plug == current->plug) {
1304 blk_flush_plug(plug, false);
1305 current->plug = NULL;
1308 EXPORT_SYMBOL(blk_finish_plug);
1310 void blk_io_schedule(void)
1312 /* Prevent hang_check timer from firing at us during very long I/O */
1313 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1316 io_schedule_timeout(timeout);
1320 EXPORT_SYMBOL_GPL(blk_io_schedule);
1322 int __init blk_dev_init(void)
1324 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1325 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1326 sizeof_field(struct request, cmd_flags));
1327 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1328 sizeof_field(struct bio, bi_opf));
1329 BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1330 __alignof__(struct request_queue)) !=
1331 sizeof(struct request_queue));
1333 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1334 kblockd_workqueue = alloc_workqueue("kblockd",
1335 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1336 if (!kblockd_workqueue)
1337 panic("Failed to create kblockd\n");
1339 blk_requestq_cachep = kmem_cache_create("request_queue",
1340 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1342 blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1343 sizeof(struct request_queue) +
1344 sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1346 blk_debugfs_root = debugfs_create_dir("block", NULL);