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/psi.h>
41 #include <linux/part_stat.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
49 #include "blk-mq-sched.h"
51 #include "blk-cgroup.h"
52 #include "blk-throttle.h"
54 struct dentry
*blk_debugfs_root
;
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert
);
63 DEFINE_IDA(blk_queue_ida
);
66 * For queue allocation
68 struct kmem_cache
*blk_requestq_cachep
;
69 struct kmem_cache
*blk_requestq_srcu_cachep
;
72 * Controlling structure to kblockd
74 static struct workqueue_struct
*kblockd_workqueue
;
77 * blk_queue_flag_set - atomically set a queue flag
78 * @flag: flag to be set
81 void blk_queue_flag_set(unsigned int flag
, struct request_queue
*q
)
83 set_bit(flag
, &q
->queue_flags
);
85 EXPORT_SYMBOL(blk_queue_flag_set
);
88 * blk_queue_flag_clear - atomically clear a queue flag
89 * @flag: flag to be cleared
92 void blk_queue_flag_clear(unsigned int flag
, struct request_queue
*q
)
94 clear_bit(flag
, &q
->queue_flags
);
96 EXPORT_SYMBOL(blk_queue_flag_clear
);
99 * blk_queue_flag_test_and_set - atomically test and set a queue flag
100 * @flag: flag to be set
103 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
104 * the flag was already set.
106 bool blk_queue_flag_test_and_set(unsigned int flag
, struct request_queue
*q
)
108 return test_and_set_bit(flag
, &q
->queue_flags
);
110 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set
);
112 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
113 static const char *const blk_op_name
[] = {
117 REQ_OP_NAME(DISCARD
),
118 REQ_OP_NAME(SECURE_ERASE
),
119 REQ_OP_NAME(ZONE_RESET
),
120 REQ_OP_NAME(ZONE_RESET_ALL
),
121 REQ_OP_NAME(ZONE_OPEN
),
122 REQ_OP_NAME(ZONE_CLOSE
),
123 REQ_OP_NAME(ZONE_FINISH
),
124 REQ_OP_NAME(ZONE_APPEND
),
125 REQ_OP_NAME(WRITE_ZEROES
),
127 REQ_OP_NAME(DRV_OUT
),
132 * blk_op_str - Return string XXX in the REQ_OP_XXX.
135 * Description: Centralize block layer function to convert REQ_OP_XXX into
136 * string format. Useful in the debugging and tracing bio or request. For
137 * invalid REQ_OP_XXX it returns string "UNKNOWN".
139 inline const char *blk_op_str(unsigned int op
)
141 const char *op_str
= "UNKNOWN";
143 if (op
< ARRAY_SIZE(blk_op_name
) && blk_op_name
[op
])
144 op_str
= blk_op_name
[op
];
148 EXPORT_SYMBOL_GPL(blk_op_str
);
150 static const struct {
154 [BLK_STS_OK
] = { 0, "" },
155 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
156 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
157 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
158 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
159 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
160 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
161 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
162 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
163 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
164 [BLK_STS_DEV_RESOURCE
] = { -EBUSY
, "device resource" },
165 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
166 [BLK_STS_OFFLINE
] = { -ENODEV
, "device offline" },
168 /* device mapper special case, should not leak out: */
169 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
171 /* zone device specific errors */
172 [BLK_STS_ZONE_OPEN_RESOURCE
] = { -ETOOMANYREFS
, "open zones exceeded" },
173 [BLK_STS_ZONE_ACTIVE_RESOURCE
] = { -EOVERFLOW
, "active zones exceeded" },
175 /* everything else not covered above: */
176 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
179 blk_status_t
errno_to_blk_status(int errno
)
183 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
184 if (blk_errors
[i
].errno
== errno
)
185 return (__force blk_status_t
)i
;
188 return BLK_STS_IOERR
;
190 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
192 int blk_status_to_errno(blk_status_t status
)
194 int idx
= (__force
int)status
;
196 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
198 return blk_errors
[idx
].errno
;
200 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
202 const char *blk_status_to_str(blk_status_t status
)
204 int idx
= (__force
int)status
;
206 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
208 return blk_errors
[idx
].name
;
212 * blk_sync_queue - cancel any pending callbacks on a queue
216 * The block layer may perform asynchronous callback activity
217 * on a queue, such as calling the unplug function after a timeout.
218 * A block device may call blk_sync_queue to ensure that any
219 * such activity is cancelled, thus allowing it to release resources
220 * that the callbacks might use. The caller must already have made sure
221 * that its ->submit_bio will not re-add plugging prior to calling
224 * This function does not cancel any asynchronous activity arising
225 * out of elevator or throttling code. That would require elevator_exit()
226 * and blkcg_exit_queue() to be called with queue lock initialized.
229 void blk_sync_queue(struct request_queue
*q
)
231 del_timer_sync(&q
->timeout
);
232 cancel_work_sync(&q
->timeout_work
);
234 EXPORT_SYMBOL(blk_sync_queue
);
237 * blk_set_pm_only - increment pm_only counter
238 * @q: request queue pointer
240 void blk_set_pm_only(struct request_queue
*q
)
242 atomic_inc(&q
->pm_only
);
244 EXPORT_SYMBOL_GPL(blk_set_pm_only
);
246 void blk_clear_pm_only(struct request_queue
*q
)
250 pm_only
= atomic_dec_return(&q
->pm_only
);
251 WARN_ON_ONCE(pm_only
< 0);
253 wake_up_all(&q
->mq_freeze_wq
);
255 EXPORT_SYMBOL_GPL(blk_clear_pm_only
);
258 * blk_put_queue - decrement the request_queue refcount
259 * @q: the request_queue structure to decrement the refcount for
261 * Decrements the refcount of the request_queue kobject. When this reaches 0
262 * we'll have blk_release_queue() called.
264 * Context: Any context, but the last reference must not be dropped from
267 void blk_put_queue(struct request_queue
*q
)
269 kobject_put(&q
->kobj
);
271 EXPORT_SYMBOL(blk_put_queue
);
273 void blk_queue_start_drain(struct request_queue
*q
)
276 * When queue DYING flag is set, we need to block new req
277 * entering queue, so we call blk_freeze_queue_start() to
278 * prevent I/O from crossing blk_queue_enter().
280 blk_freeze_queue_start(q
);
282 blk_mq_wake_waiters(q
);
283 /* Make blk_queue_enter() reexamine the DYING flag. */
284 wake_up_all(&q
->mq_freeze_wq
);
288 * blk_cleanup_queue - shutdown a request queue
289 * @q: request queue to shutdown
291 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
292 * put it. All future requests will be failed immediately with -ENODEV.
296 void blk_cleanup_queue(struct request_queue
*q
)
298 /* cannot be called from atomic context */
301 WARN_ON_ONCE(blk_queue_registered(q
));
303 /* mark @q DYING, no new request or merges will be allowed afterwards */
304 blk_queue_flag_set(QUEUE_FLAG_DYING
, q
);
305 blk_queue_start_drain(q
);
307 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
308 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
311 * Drain all requests queued before DYING marking. Set DEAD flag to
312 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
313 * after draining finished.
317 blk_queue_flag_set(QUEUE_FLAG_DEAD
, q
);
320 if (queue_is_mq(q
)) {
321 blk_mq_cancel_work_sync(q
);
322 blk_mq_exit_queue(q
);
325 /* @q is and will stay empty, shutdown and put */
328 EXPORT_SYMBOL(blk_cleanup_queue
);
331 * blk_queue_enter() - try to increase q->q_usage_counter
332 * @q: request queue pointer
333 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
335 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
337 const bool pm
= flags
& BLK_MQ_REQ_PM
;
339 while (!blk_try_enter_queue(q
, pm
)) {
340 if (flags
& BLK_MQ_REQ_NOWAIT
)
344 * read pair of barrier in blk_freeze_queue_start(), we need to
345 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
346 * reading .mq_freeze_depth or queue dying flag, otherwise the
347 * following wait may never return if the two reads are
351 wait_event(q
->mq_freeze_wq
,
352 (!q
->mq_freeze_depth
&&
353 blk_pm_resume_queue(pm
, q
)) ||
355 if (blk_queue_dying(q
))
362 int __bio_queue_enter(struct request_queue
*q
, struct bio
*bio
)
364 while (!blk_try_enter_queue(q
, false)) {
365 struct gendisk
*disk
= bio
->bi_bdev
->bd_disk
;
367 if (bio
->bi_opf
& REQ_NOWAIT
) {
368 if (test_bit(GD_DEAD
, &disk
->state
))
370 bio_wouldblock_error(bio
);
375 * read pair of barrier in blk_freeze_queue_start(), we need to
376 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
377 * reading .mq_freeze_depth or queue dying flag, otherwise the
378 * following wait may never return if the two reads are
382 wait_event(q
->mq_freeze_wq
,
383 (!q
->mq_freeze_depth
&&
384 blk_pm_resume_queue(false, q
)) ||
385 test_bit(GD_DEAD
, &disk
->state
));
386 if (test_bit(GD_DEAD
, &disk
->state
))
396 void blk_queue_exit(struct request_queue
*q
)
398 percpu_ref_put(&q
->q_usage_counter
);
401 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
403 struct request_queue
*q
=
404 container_of(ref
, struct request_queue
, q_usage_counter
);
406 wake_up_all(&q
->mq_freeze_wq
);
409 static void blk_rq_timed_out_timer(struct timer_list
*t
)
411 struct request_queue
*q
= from_timer(q
, t
, timeout
);
413 kblockd_schedule_work(&q
->timeout_work
);
416 static void blk_timeout_work(struct work_struct
*work
)
420 struct request_queue
*blk_alloc_queue(int node_id
, bool alloc_srcu
)
422 struct request_queue
*q
;
425 q
= kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu
),
426 GFP_KERNEL
| __GFP_ZERO
, node_id
);
431 blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU
, q
);
432 if (init_srcu_struct(q
->srcu
) != 0)
436 q
->last_merge
= NULL
;
438 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
442 ret
= bioset_init(&q
->bio_split
, BIO_POOL_SIZE
, 0, 0);
446 q
->stats
= blk_alloc_queue_stats();
452 atomic_set(&q
->nr_active_requests_shared_tags
, 0);
454 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
455 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
456 INIT_LIST_HEAD(&q
->icq_list
);
458 kobject_init(&q
->kobj
, &blk_queue_ktype
);
460 mutex_init(&q
->debugfs_mutex
);
461 mutex_init(&q
->sysfs_lock
);
462 mutex_init(&q
->sysfs_dir_lock
);
463 spin_lock_init(&q
->queue_lock
);
465 init_waitqueue_head(&q
->mq_freeze_wq
);
466 mutex_init(&q
->mq_freeze_lock
);
469 * Init percpu_ref in atomic mode so that it's faster to shutdown.
470 * See blk_register_queue() for details.
472 if (percpu_ref_init(&q
->q_usage_counter
,
473 blk_queue_usage_counter_release
,
474 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
477 blk_queue_dma_alignment(q
, 511);
478 blk_set_default_limits(&q
->limits
);
479 q
->nr_requests
= BLKDEV_DEFAULT_RQ
;
484 blk_free_queue_stats(q
->stats
);
486 bioset_exit(&q
->bio_split
);
488 ida_simple_remove(&blk_queue_ida
, q
->id
);
491 cleanup_srcu_struct(q
->srcu
);
493 kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu
), q
);
498 * blk_get_queue - increment the request_queue refcount
499 * @q: the request_queue structure to increment the refcount for
501 * Increment the refcount of the request_queue kobject.
503 * Context: Any context.
505 bool blk_get_queue(struct request_queue
*q
)
507 if (likely(!blk_queue_dying(q
))) {
514 EXPORT_SYMBOL(blk_get_queue
);
516 #ifdef CONFIG_FAIL_MAKE_REQUEST
518 static DECLARE_FAULT_ATTR(fail_make_request
);
520 static int __init
setup_fail_make_request(char *str
)
522 return setup_fault_attr(&fail_make_request
, str
);
524 __setup("fail_make_request=", setup_fail_make_request
);
526 bool should_fail_request(struct block_device
*part
, unsigned int bytes
)
528 return part
->bd_make_it_fail
&& should_fail(&fail_make_request
, bytes
);
531 static int __init
fail_make_request_debugfs(void)
533 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
534 NULL
, &fail_make_request
);
536 return PTR_ERR_OR_ZERO(dir
);
539 late_initcall(fail_make_request_debugfs
);
540 #endif /* CONFIG_FAIL_MAKE_REQUEST */
542 static inline bool bio_check_ro(struct bio
*bio
)
544 if (op_is_write(bio_op(bio
)) && bdev_read_only(bio
->bi_bdev
)) {
545 if (op_is_flush(bio
->bi_opf
) && !bio_sectors(bio
))
547 pr_warn("Trying to write to read-only block-device %pg\n",
549 /* Older lvm-tools actually trigger this */
556 static noinline
int should_fail_bio(struct bio
*bio
)
558 if (should_fail_request(bdev_whole(bio
->bi_bdev
), bio
->bi_iter
.bi_size
))
562 ALLOW_ERROR_INJECTION(should_fail_bio
, ERRNO
);
565 * Check whether this bio extends beyond the end of the device or partition.
566 * This may well happen - the kernel calls bread() without checking the size of
567 * the device, e.g., when mounting a file system.
569 static inline int bio_check_eod(struct bio
*bio
)
571 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
572 unsigned int nr_sectors
= bio_sectors(bio
);
574 if (nr_sectors
&& maxsector
&&
575 (nr_sectors
> maxsector
||
576 bio
->bi_iter
.bi_sector
> maxsector
- nr_sectors
)) {
577 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
578 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
579 current
->comm
, bio
->bi_bdev
, bio
->bi_opf
,
580 bio
->bi_iter
.bi_sector
, nr_sectors
, maxsector
);
587 * Remap block n of partition p to block n+start(p) of the disk.
589 static int blk_partition_remap(struct bio
*bio
)
591 struct block_device
*p
= bio
->bi_bdev
;
593 if (unlikely(should_fail_request(p
, bio
->bi_iter
.bi_size
)))
595 if (bio_sectors(bio
)) {
596 bio
->bi_iter
.bi_sector
+= p
->bd_start_sect
;
597 trace_block_bio_remap(bio
, p
->bd_dev
,
598 bio
->bi_iter
.bi_sector
-
601 bio_set_flag(bio
, BIO_REMAPPED
);
606 * Check write append to a zoned block device.
608 static inline blk_status_t
blk_check_zone_append(struct request_queue
*q
,
611 sector_t pos
= bio
->bi_iter
.bi_sector
;
612 int nr_sectors
= bio_sectors(bio
);
614 /* Only applicable to zoned block devices */
615 if (!blk_queue_is_zoned(q
))
616 return BLK_STS_NOTSUPP
;
618 /* The bio sector must point to the start of a sequential zone */
619 if (pos
& (blk_queue_zone_sectors(q
) - 1) ||
620 !blk_queue_zone_is_seq(q
, pos
))
621 return BLK_STS_IOERR
;
624 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
625 * split and could result in non-contiguous sectors being written in
628 if (nr_sectors
> q
->limits
.chunk_sectors
)
629 return BLK_STS_IOERR
;
631 /* Make sure the BIO is small enough and will not get split */
632 if (nr_sectors
> q
->limits
.max_zone_append_sectors
)
633 return BLK_STS_IOERR
;
635 bio
->bi_opf
|= REQ_NOMERGE
;
640 static void __submit_bio(struct bio
*bio
)
642 struct gendisk
*disk
= bio
->bi_bdev
->bd_disk
;
644 if (unlikely(!blk_crypto_bio_prep(&bio
)))
647 if (!disk
->fops
->submit_bio
) {
648 blk_mq_submit_bio(bio
);
649 } else if (likely(bio_queue_enter(bio
) == 0)) {
650 disk
->fops
->submit_bio(bio
);
651 blk_queue_exit(disk
->queue
);
656 * The loop in this function may be a bit non-obvious, and so deserves some
659 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
660 * that), so we have a list with a single bio.
661 * - We pretend that we have just taken it off a longer list, so we assign
662 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
663 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
664 * bios through a recursive call to submit_bio_noacct. If it did, we find a
665 * non-NULL value in bio_list and re-enter the loop from the top.
666 * - In this case we really did just take the bio of the top of the list (no
667 * pretending) and so remove it from bio_list, and call into ->submit_bio()
670 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
671 * bio_list_on_stack[1] contains bios that were submitted before the current
672 * ->submit_bio, but that haven't been processed yet.
674 static void __submit_bio_noacct(struct bio
*bio
)
676 struct bio_list bio_list_on_stack
[2];
678 BUG_ON(bio
->bi_next
);
680 bio_list_init(&bio_list_on_stack
[0]);
681 current
->bio_list
= bio_list_on_stack
;
684 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
685 struct bio_list lower
, same
;
688 * Create a fresh bio_list for all subordinate requests.
690 bio_list_on_stack
[1] = bio_list_on_stack
[0];
691 bio_list_init(&bio_list_on_stack
[0]);
696 * Sort new bios into those for a lower level and those for the
699 bio_list_init(&lower
);
700 bio_list_init(&same
);
701 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
702 if (q
== bdev_get_queue(bio
->bi_bdev
))
703 bio_list_add(&same
, bio
);
705 bio_list_add(&lower
, bio
);
708 * Now assemble so we handle the lowest level first.
710 bio_list_merge(&bio_list_on_stack
[0], &lower
);
711 bio_list_merge(&bio_list_on_stack
[0], &same
);
712 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
713 } while ((bio
= bio_list_pop(&bio_list_on_stack
[0])));
715 current
->bio_list
= NULL
;
718 static void __submit_bio_noacct_mq(struct bio
*bio
)
720 struct bio_list bio_list
[2] = { };
722 current
->bio_list
= bio_list
;
726 } while ((bio
= bio_list_pop(&bio_list
[0])));
728 current
->bio_list
= NULL
;
731 void submit_bio_noacct_nocheck(struct bio
*bio
)
734 * We only want one ->submit_bio to be active at a time, else stack
735 * usage with stacked devices could be a problem. Use current->bio_list
736 * to collect a list of requests submited by a ->submit_bio method while
737 * it is active, and then process them after it returned.
739 if (current
->bio_list
)
740 bio_list_add(¤t
->bio_list
[0], bio
);
741 else if (!bio
->bi_bdev
->bd_disk
->fops
->submit_bio
)
742 __submit_bio_noacct_mq(bio
);
744 __submit_bio_noacct(bio
);
748 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
749 * @bio: The bio describing the location in memory and on the device.
751 * This is a version of submit_bio() that shall only be used for I/O that is
752 * resubmitted to lower level drivers by stacking block drivers. All file
753 * systems and other upper level users of the block layer should use
754 * submit_bio() instead.
756 void submit_bio_noacct(struct bio
*bio
)
758 struct block_device
*bdev
= bio
->bi_bdev
;
759 struct request_queue
*q
= bdev_get_queue(bdev
);
760 blk_status_t status
= BLK_STS_IOERR
;
761 struct blk_plug
*plug
;
765 plug
= blk_mq_plug(q
, bio
);
766 if (plug
&& plug
->nowait
)
767 bio
->bi_opf
|= REQ_NOWAIT
;
770 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
771 * if queue does not support NOWAIT.
773 if ((bio
->bi_opf
& REQ_NOWAIT
) && !blk_queue_nowait(q
))
776 if (should_fail_bio(bio
))
778 if (unlikely(bio_check_ro(bio
)))
780 if (!bio_flagged(bio
, BIO_REMAPPED
)) {
781 if (unlikely(bio_check_eod(bio
)))
783 if (bdev
->bd_partno
&& unlikely(blk_partition_remap(bio
)))
788 * Filter flush bio's early so that bio based drivers without flush
789 * support don't have to worry about them.
791 if (op_is_flush(bio
->bi_opf
) &&
792 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
793 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
794 if (!bio_sectors(bio
)) {
800 if (!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
801 bio_clear_polled(bio
);
803 switch (bio_op(bio
)) {
805 if (!bdev_max_discard_sectors(bdev
))
808 case REQ_OP_SECURE_ERASE
:
809 if (!bdev_max_secure_erase_sectors(bdev
))
812 case REQ_OP_ZONE_APPEND
:
813 status
= blk_check_zone_append(q
, bio
);
814 if (status
!= BLK_STS_OK
)
817 case REQ_OP_ZONE_RESET
:
818 case REQ_OP_ZONE_OPEN
:
819 case REQ_OP_ZONE_CLOSE
:
820 case REQ_OP_ZONE_FINISH
:
821 if (!blk_queue_is_zoned(q
))
824 case REQ_OP_ZONE_RESET_ALL
:
825 if (!blk_queue_is_zoned(q
) || !blk_queue_zone_resetall(q
))
828 case REQ_OP_WRITE_ZEROES
:
829 if (!q
->limits
.max_write_zeroes_sectors
)
836 if (blk_throtl_bio(bio
))
839 blk_cgroup_bio_start(bio
);
840 blkcg_bio_issue_init(bio
);
842 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
843 trace_block_bio_queue(bio
);
844 /* Now that enqueuing has been traced, we need to trace
845 * completion as well.
847 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
849 submit_bio_noacct_nocheck(bio
);
853 status
= BLK_STS_NOTSUPP
;
855 bio
->bi_status
= status
;
858 EXPORT_SYMBOL(submit_bio_noacct
);
861 * submit_bio - submit a bio to the block device layer for I/O
862 * @bio: The &struct bio which describes the I/O
864 * submit_bio() is used to submit I/O requests to block devices. It is passed a
865 * fully set up &struct bio that describes the I/O that needs to be done. The
866 * bio will be send to the device described by the bi_bdev field.
868 * The success/failure status of the request, along with notification of
869 * completion, is delivered asynchronously through the ->bi_end_io() callback
870 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
873 void submit_bio(struct bio
*bio
)
875 if (blkcg_punt_bio_submit(bio
))
878 if (bio_op(bio
) == REQ_OP_READ
) {
879 task_io_account_read(bio
->bi_iter
.bi_size
);
880 count_vm_events(PGPGIN
, bio_sectors(bio
));
881 } else if (bio_op(bio
) == REQ_OP_WRITE
) {
882 count_vm_events(PGPGOUT
, bio_sectors(bio
));
886 * If we're reading data that is part of the userspace workingset, count
887 * submission time as memory stall. When the device is congested, or
888 * the submitting cgroup IO-throttled, submission can be a significant
889 * part of overall IO time.
891 if (unlikely(bio_op(bio
) == REQ_OP_READ
&&
892 bio_flagged(bio
, BIO_WORKINGSET
))) {
893 unsigned long pflags
;
895 psi_memstall_enter(&pflags
);
896 submit_bio_noacct(bio
);
897 psi_memstall_leave(&pflags
);
901 submit_bio_noacct(bio
);
903 EXPORT_SYMBOL(submit_bio
);
906 * bio_poll - poll for BIO completions
907 * @bio: bio to poll for
908 * @iob: batches of IO
909 * @flags: BLK_POLL_* flags that control the behavior
911 * Poll for completions on queue associated with the bio. Returns number of
912 * completed entries found.
914 * Note: the caller must either be the context that submitted @bio, or
915 * be in a RCU critical section to prevent freeing of @bio.
917 int bio_poll(struct bio
*bio
, struct io_comp_batch
*iob
, unsigned int flags
)
919 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
920 blk_qc_t cookie
= READ_ONCE(bio
->bi_cookie
);
923 if (cookie
== BLK_QC_T_NONE
||
924 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
927 blk_flush_plug(current
->plug
, false);
929 if (bio_queue_enter(bio
))
931 if (queue_is_mq(q
)) {
932 ret
= blk_mq_poll(q
, cookie
, iob
, flags
);
934 struct gendisk
*disk
= q
->disk
;
936 if (disk
&& disk
->fops
->poll_bio
)
937 ret
= disk
->fops
->poll_bio(bio
, iob
, flags
);
942 EXPORT_SYMBOL_GPL(bio_poll
);
945 * Helper to implement file_operations.iopoll. Requires the bio to be stored
946 * in iocb->private, and cleared before freeing the bio.
948 int iocb_bio_iopoll(struct kiocb
*kiocb
, struct io_comp_batch
*iob
,
955 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
956 * point to a freshly allocated bio at this point. If that happens
957 * we have a few cases to consider:
959 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
960 * simply nothing in this case
961 * 2) the bio points to a not poll enabled device. bio_poll will catch
963 * 3) the bio points to a poll capable device, including but not
964 * limited to the one that the original bio pointed to. In this
965 * case we will call into the actual poll method and poll for I/O,
966 * even if we don't need to, but it won't cause harm either.
968 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
969 * is still allocated. Because partitions hold a reference to the whole
970 * device bdev and thus disk, the disk is also still valid. Grabbing
971 * a reference to the queue in bio_poll() ensures the hctxs and requests
972 * are still valid as well.
975 bio
= READ_ONCE(kiocb
->private);
976 if (bio
&& bio
->bi_bdev
)
977 ret
= bio_poll(bio
, iob
, flags
);
982 EXPORT_SYMBOL_GPL(iocb_bio_iopoll
);
984 void update_io_ticks(struct block_device
*part
, unsigned long now
, bool end
)
988 stamp
= READ_ONCE(part
->bd_stamp
);
989 if (unlikely(time_after(now
, stamp
))) {
990 if (likely(cmpxchg(&part
->bd_stamp
, stamp
, now
) == stamp
))
991 __part_stat_add(part
, io_ticks
, end
? now
- stamp
: 1);
993 if (part
->bd_partno
) {
994 part
= bdev_whole(part
);
999 unsigned long bdev_start_io_acct(struct block_device
*bdev
,
1000 unsigned int sectors
, unsigned int op
,
1001 unsigned long start_time
)
1003 const int sgrp
= op_stat_group(op
);
1006 update_io_ticks(bdev
, start_time
, false);
1007 part_stat_inc(bdev
, ios
[sgrp
]);
1008 part_stat_add(bdev
, sectors
[sgrp
], sectors
);
1009 part_stat_local_inc(bdev
, in_flight
[op_is_write(op
)]);
1014 EXPORT_SYMBOL(bdev_start_io_acct
);
1017 * bio_start_io_acct_time - start I/O accounting for bio based drivers
1018 * @bio: bio to start account for
1019 * @start_time: start time that should be passed back to bio_end_io_acct().
1021 void bio_start_io_acct_time(struct bio
*bio
, unsigned long start_time
)
1023 bdev_start_io_acct(bio
->bi_bdev
, bio_sectors(bio
),
1024 bio_op(bio
), start_time
);
1026 EXPORT_SYMBOL_GPL(bio_start_io_acct_time
);
1029 * bio_start_io_acct - start I/O accounting for bio based drivers
1030 * @bio: bio to start account for
1032 * Returns the start time that should be passed back to bio_end_io_acct().
1034 unsigned long bio_start_io_acct(struct bio
*bio
)
1036 return bdev_start_io_acct(bio
->bi_bdev
, bio_sectors(bio
),
1037 bio_op(bio
), jiffies
);
1039 EXPORT_SYMBOL_GPL(bio_start_io_acct
);
1041 void bdev_end_io_acct(struct block_device
*bdev
, unsigned int op
,
1042 unsigned long start_time
)
1044 const int sgrp
= op_stat_group(op
);
1045 unsigned long now
= READ_ONCE(jiffies
);
1046 unsigned long duration
= now
- start_time
;
1049 update_io_ticks(bdev
, now
, true);
1050 part_stat_add(bdev
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1051 part_stat_local_dec(bdev
, in_flight
[op_is_write(op
)]);
1054 EXPORT_SYMBOL(bdev_end_io_acct
);
1056 void bio_end_io_acct_remapped(struct bio
*bio
, unsigned long start_time
,
1057 struct block_device
*orig_bdev
)
1059 bdev_end_io_acct(orig_bdev
, bio_op(bio
), start_time
);
1061 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped
);
1064 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1065 * @q : the queue of the device being checked
1068 * Check if underlying low-level drivers of a device are busy.
1069 * If the drivers want to export their busy state, they must set own
1070 * exporting function using blk_queue_lld_busy() first.
1072 * Basically, this function is used only by request stacking drivers
1073 * to stop dispatching requests to underlying devices when underlying
1074 * devices are busy. This behavior helps more I/O merging on the queue
1075 * of the request stacking driver and prevents I/O throughput regression
1076 * on burst I/O load.
1079 * 0 - Not busy (The request stacking driver should dispatch request)
1080 * 1 - Busy (The request stacking driver should stop dispatching request)
1082 int blk_lld_busy(struct request_queue
*q
)
1084 if (queue_is_mq(q
) && q
->mq_ops
->busy
)
1085 return q
->mq_ops
->busy(q
);
1089 EXPORT_SYMBOL_GPL(blk_lld_busy
);
1091 int kblockd_schedule_work(struct work_struct
*work
)
1093 return queue_work(kblockd_workqueue
, work
);
1095 EXPORT_SYMBOL(kblockd_schedule_work
);
1097 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
1098 unsigned long delay
)
1100 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
1102 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
1104 void blk_start_plug_nr_ios(struct blk_plug
*plug
, unsigned short nr_ios
)
1106 struct task_struct
*tsk
= current
;
1109 * If this is a nested plug, don't actually assign it.
1114 plug
->mq_list
= NULL
;
1115 plug
->cached_rq
= NULL
;
1116 plug
->nr_ios
= min_t(unsigned short, nr_ios
, BLK_MAX_REQUEST_COUNT
);
1118 plug
->multiple_queues
= false;
1119 plug
->has_elevator
= false;
1120 plug
->nowait
= false;
1121 INIT_LIST_HEAD(&plug
->cb_list
);
1124 * Store ordering should not be needed here, since a potential
1125 * preempt will imply a full memory barrier
1131 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1132 * @plug: The &struct blk_plug that needs to be initialized
1135 * blk_start_plug() indicates to the block layer an intent by the caller
1136 * to submit multiple I/O requests in a batch. The block layer may use
1137 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1138 * is called. However, the block layer may choose to submit requests
1139 * before a call to blk_finish_plug() if the number of queued I/Os
1140 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1141 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1142 * the task schedules (see below).
1144 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1145 * pending I/O should the task end up blocking between blk_start_plug() and
1146 * blk_finish_plug(). This is important from a performance perspective, but
1147 * also ensures that we don't deadlock. For instance, if the task is blocking
1148 * for a memory allocation, memory reclaim could end up wanting to free a
1149 * page belonging to that request that is currently residing in our private
1150 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1151 * this kind of deadlock.
1153 void blk_start_plug(struct blk_plug
*plug
)
1155 blk_start_plug_nr_ios(plug
, 1);
1157 EXPORT_SYMBOL(blk_start_plug
);
1159 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
1161 LIST_HEAD(callbacks
);
1163 while (!list_empty(&plug
->cb_list
)) {
1164 list_splice_init(&plug
->cb_list
, &callbacks
);
1166 while (!list_empty(&callbacks
)) {
1167 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
1170 list_del(&cb
->list
);
1171 cb
->callback(cb
, from_schedule
);
1176 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
1179 struct blk_plug
*plug
= current
->plug
;
1180 struct blk_plug_cb
*cb
;
1185 list_for_each_entry(cb
, &plug
->cb_list
, list
)
1186 if (cb
->callback
== unplug
&& cb
->data
== data
)
1189 /* Not currently on the callback list */
1190 BUG_ON(size
< sizeof(*cb
));
1191 cb
= kzalloc(size
, GFP_ATOMIC
);
1194 cb
->callback
= unplug
;
1195 list_add(&cb
->list
, &plug
->cb_list
);
1199 EXPORT_SYMBOL(blk_check_plugged
);
1201 void __blk_flush_plug(struct blk_plug
*plug
, bool from_schedule
)
1203 if (!list_empty(&plug
->cb_list
))
1204 flush_plug_callbacks(plug
, from_schedule
);
1205 if (!rq_list_empty(plug
->mq_list
))
1206 blk_mq_flush_plug_list(plug
, from_schedule
);
1208 * Unconditionally flush out cached requests, even if the unplug
1209 * event came from schedule. Since we know hold references to the
1210 * queue for cached requests, we don't want a blocked task holding
1211 * up a queue freeze/quiesce event.
1213 if (unlikely(!rq_list_empty(plug
->cached_rq
)))
1214 blk_mq_free_plug_rqs(plug
);
1218 * blk_finish_plug - mark the end of a batch of submitted I/O
1219 * @plug: The &struct blk_plug passed to blk_start_plug()
1222 * Indicate that a batch of I/O submissions is complete. This function
1223 * must be paired with an initial call to blk_start_plug(). The intent
1224 * is to allow the block layer to optimize I/O submission. See the
1225 * documentation for blk_start_plug() for more information.
1227 void blk_finish_plug(struct blk_plug
*plug
)
1229 if (plug
== current
->plug
) {
1230 __blk_flush_plug(plug
, false);
1231 current
->plug
= NULL
;
1234 EXPORT_SYMBOL(blk_finish_plug
);
1236 void blk_io_schedule(void)
1238 /* Prevent hang_check timer from firing at us during very long I/O */
1239 unsigned long timeout
= sysctl_hung_task_timeout_secs
* HZ
/ 2;
1242 io_schedule_timeout(timeout
);
1246 EXPORT_SYMBOL_GPL(blk_io_schedule
);
1248 int __init
blk_dev_init(void)
1250 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
1251 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1252 sizeof_field(struct request
, cmd_flags
));
1253 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1254 sizeof_field(struct bio
, bi_opf
));
1255 BUILD_BUG_ON(ALIGN(offsetof(struct request_queue
, srcu
),
1256 __alignof__(struct request_queue
)) !=
1257 sizeof(struct request_queue
));
1259 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1260 kblockd_workqueue
= alloc_workqueue("kblockd",
1261 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
1262 if (!kblockd_workqueue
)
1263 panic("Failed to create kblockd\n");
1265 blk_requestq_cachep
= kmem_cache_create("request_queue",
1266 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
1268 blk_requestq_srcu_cachep
= kmem_cache_create("request_queue_srcu",
1269 sizeof(struct request_queue
) +
1270 sizeof(struct srcu_struct
), 0, SLAB_PANIC
, NULL
);
1272 blk_debugfs_root
= debugfs_create_dir("block", NULL
);