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-mq.h>
20 #include <linux/blk-pm.h>
21 #include <linux/blk-integrity.h>
22 #include <linux/highmem.h>
24 #include <linux/pagemap.h>
25 #include <linux/kernel_stat.h>
26 #include <linux/string.h>
27 #include <linux/init.h>
28 #include <linux/completion.h>
29 #include <linux/slab.h>
30 #include <linux/swap.h>
31 #include <linux/writeback.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/fault-inject.h>
34 #include <linux/list_sort.h>
35 #include <linux/delay.h>
36 #include <linux/ratelimit.h>
37 #include <linux/pm_runtime.h>
38 #include <linux/blk-cgroup.h>
39 #include <linux/t10-pi.h>
40 #include <linux/debugfs.h>
41 #include <linux/bpf.h>
42 #include <linux/psi.h>
43 #include <linux/sched/sysctl.h>
44 #include <linux/blk-crypto.h>
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/block.h>
51 #include "blk-mq-sched.h"
53 #include "blk-throttle.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
;
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 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
114 memset(rq
, 0, sizeof(*rq
));
116 INIT_LIST_HEAD(&rq
->queuelist
);
118 rq
->__sector
= (sector_t
) -1;
119 INIT_HLIST_NODE(&rq
->hash
);
120 RB_CLEAR_NODE(&rq
->rb_node
);
121 rq
->tag
= BLK_MQ_NO_TAG
;
122 rq
->internal_tag
= BLK_MQ_NO_TAG
;
123 rq
->start_time_ns
= ktime_get_ns();
125 blk_crypto_rq_set_defaults(rq
);
127 EXPORT_SYMBOL(blk_rq_init
);
129 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
130 static const char *const blk_op_name
[] = {
134 REQ_OP_NAME(DISCARD
),
135 REQ_OP_NAME(SECURE_ERASE
),
136 REQ_OP_NAME(ZONE_RESET
),
137 REQ_OP_NAME(ZONE_RESET_ALL
),
138 REQ_OP_NAME(ZONE_OPEN
),
139 REQ_OP_NAME(ZONE_CLOSE
),
140 REQ_OP_NAME(ZONE_FINISH
),
141 REQ_OP_NAME(ZONE_APPEND
),
142 REQ_OP_NAME(WRITE_SAME
),
143 REQ_OP_NAME(WRITE_ZEROES
),
145 REQ_OP_NAME(DRV_OUT
),
150 * blk_op_str - Return string XXX in the REQ_OP_XXX.
153 * Description: Centralize block layer function to convert REQ_OP_XXX into
154 * string format. Useful in the debugging and tracing bio or request. For
155 * invalid REQ_OP_XXX it returns string "UNKNOWN".
157 inline const char *blk_op_str(unsigned int op
)
159 const char *op_str
= "UNKNOWN";
161 if (op
< ARRAY_SIZE(blk_op_name
) && blk_op_name
[op
])
162 op_str
= blk_op_name
[op
];
166 EXPORT_SYMBOL_GPL(blk_op_str
);
168 static const struct {
172 [BLK_STS_OK
] = { 0, "" },
173 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
174 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
175 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
176 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
177 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
178 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
179 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
180 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
181 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
182 [BLK_STS_DEV_RESOURCE
] = { -EBUSY
, "device resource" },
183 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
185 /* device mapper special case, should not leak out: */
186 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
188 /* zone device specific errors */
189 [BLK_STS_ZONE_OPEN_RESOURCE
] = { -ETOOMANYREFS
, "open zones exceeded" },
190 [BLK_STS_ZONE_ACTIVE_RESOURCE
] = { -EOVERFLOW
, "active zones exceeded" },
192 /* everything else not covered above: */
193 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
196 blk_status_t
errno_to_blk_status(int errno
)
200 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
201 if (blk_errors
[i
].errno
== errno
)
202 return (__force blk_status_t
)i
;
205 return BLK_STS_IOERR
;
207 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
209 int blk_status_to_errno(blk_status_t status
)
211 int idx
= (__force
int)status
;
213 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
215 return blk_errors
[idx
].errno
;
217 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
219 void blk_print_req_error(struct request
*req
, blk_status_t status
)
221 int idx
= (__force
int)status
;
223 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
226 printk_ratelimited(KERN_ERR
227 "%s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
228 "phys_seg %u prio class %u\n",
229 blk_errors
[idx
].name
,
230 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
231 blk_rq_pos(req
), req_op(req
), blk_op_str(req_op(req
)),
232 req
->cmd_flags
& ~REQ_OP_MASK
,
233 req
->nr_phys_segments
,
234 IOPRIO_PRIO_CLASS(req
->ioprio
));
237 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
239 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
240 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
241 (unsigned long long) rq
->cmd_flags
);
243 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
244 (unsigned long long)blk_rq_pos(rq
),
245 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
246 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
247 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
249 EXPORT_SYMBOL(blk_dump_rq_flags
);
252 * blk_sync_queue - cancel any pending callbacks on a queue
256 * The block layer may perform asynchronous callback activity
257 * on a queue, such as calling the unplug function after a timeout.
258 * A block device may call blk_sync_queue to ensure that any
259 * such activity is cancelled, thus allowing it to release resources
260 * that the callbacks might use. The caller must already have made sure
261 * that its ->submit_bio will not re-add plugging prior to calling
264 * This function does not cancel any asynchronous activity arising
265 * out of elevator or throttling code. That would require elevator_exit()
266 * and blkcg_exit_queue() to be called with queue lock initialized.
269 void blk_sync_queue(struct request_queue
*q
)
271 del_timer_sync(&q
->timeout
);
272 cancel_work_sync(&q
->timeout_work
);
274 EXPORT_SYMBOL(blk_sync_queue
);
277 * blk_set_pm_only - increment pm_only counter
278 * @q: request queue pointer
280 void blk_set_pm_only(struct request_queue
*q
)
282 atomic_inc(&q
->pm_only
);
284 EXPORT_SYMBOL_GPL(blk_set_pm_only
);
286 void blk_clear_pm_only(struct request_queue
*q
)
290 pm_only
= atomic_dec_return(&q
->pm_only
);
291 WARN_ON_ONCE(pm_only
< 0);
293 wake_up_all(&q
->mq_freeze_wq
);
295 EXPORT_SYMBOL_GPL(blk_clear_pm_only
);
298 * blk_put_queue - decrement the request_queue refcount
299 * @q: the request_queue structure to decrement the refcount for
301 * Decrements the refcount of the request_queue kobject. When this reaches 0
302 * we'll have blk_release_queue() called.
304 * Context: Any context, but the last reference must not be dropped from
307 void blk_put_queue(struct request_queue
*q
)
309 kobject_put(&q
->kobj
);
311 EXPORT_SYMBOL(blk_put_queue
);
313 void blk_queue_start_drain(struct request_queue
*q
)
316 * When queue DYING flag is set, we need to block new req
317 * entering queue, so we call blk_freeze_queue_start() to
318 * prevent I/O from crossing blk_queue_enter().
320 blk_freeze_queue_start(q
);
322 blk_mq_wake_waiters(q
);
323 /* Make blk_queue_enter() reexamine the DYING flag. */
324 wake_up_all(&q
->mq_freeze_wq
);
327 void blk_set_queue_dying(struct request_queue
*q
)
329 blk_queue_flag_set(QUEUE_FLAG_DYING
, q
);
330 blk_queue_start_drain(q
);
332 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
335 * blk_cleanup_queue - shutdown a request queue
336 * @q: request queue to shutdown
338 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
339 * put it. All future requests will be failed immediately with -ENODEV.
343 void blk_cleanup_queue(struct request_queue
*q
)
345 /* cannot be called from atomic context */
348 WARN_ON_ONCE(blk_queue_registered(q
));
350 /* mark @q DYING, no new request or merges will be allowed afterwards */
351 blk_set_queue_dying(q
);
353 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
354 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
357 * Drain all requests queued before DYING marking. Set DEAD flag to
358 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
359 * after draining finished.
363 blk_queue_flag_set(QUEUE_FLAG_DEAD
, q
);
367 blk_mq_exit_queue(q
);
370 * In theory, request pool of sched_tags belongs to request queue.
371 * However, the current implementation requires tag_set for freeing
372 * requests, so free the pool now.
374 * Queue has become frozen, there can't be any in-queue requests, so
375 * it is safe to free requests now.
377 mutex_lock(&q
->sysfs_lock
);
379 blk_mq_sched_free_rqs(q
);
380 mutex_unlock(&q
->sysfs_lock
);
382 percpu_ref_exit(&q
->q_usage_counter
);
384 /* @q is and will stay empty, shutdown and put */
387 EXPORT_SYMBOL(blk_cleanup_queue
);
389 static bool blk_try_enter_queue(struct request_queue
*q
, bool pm
)
392 if (!percpu_ref_tryget_live_rcu(&q
->q_usage_counter
))
396 * The code that increments the pm_only counter must ensure that the
397 * counter is globally visible before the queue is unfrozen.
399 if (blk_queue_pm_only(q
) &&
400 (!pm
|| queue_rpm_status(q
) == RPM_SUSPENDED
))
414 * blk_queue_enter() - try to increase q->q_usage_counter
415 * @q: request queue pointer
416 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
418 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
420 const bool pm
= flags
& BLK_MQ_REQ_PM
;
422 while (!blk_try_enter_queue(q
, pm
)) {
423 if (flags
& BLK_MQ_REQ_NOWAIT
)
427 * read pair of barrier in blk_freeze_queue_start(), we need to
428 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
429 * reading .mq_freeze_depth or queue dying flag, otherwise the
430 * following wait may never return if the two reads are
434 wait_event(q
->mq_freeze_wq
,
435 (!q
->mq_freeze_depth
&&
436 blk_pm_resume_queue(pm
, q
)) ||
438 if (blk_queue_dying(q
))
445 static inline int bio_queue_enter(struct bio
*bio
)
447 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
449 while (!blk_try_enter_queue(q
, false)) {
450 struct gendisk
*disk
= bio
->bi_bdev
->bd_disk
;
452 if (bio
->bi_opf
& REQ_NOWAIT
) {
453 if (test_bit(GD_DEAD
, &disk
->state
))
455 bio_wouldblock_error(bio
);
460 * read pair of barrier in blk_freeze_queue_start(), we need to
461 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
462 * reading .mq_freeze_depth or queue dying flag, otherwise the
463 * following wait may never return if the two reads are
467 wait_event(q
->mq_freeze_wq
,
468 (!q
->mq_freeze_depth
&&
469 blk_pm_resume_queue(false, q
)) ||
470 test_bit(GD_DEAD
, &disk
->state
));
471 if (test_bit(GD_DEAD
, &disk
->state
))
481 void blk_queue_exit(struct request_queue
*q
)
483 percpu_ref_put(&q
->q_usage_counter
);
486 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
488 struct request_queue
*q
=
489 container_of(ref
, struct request_queue
, q_usage_counter
);
491 wake_up_all(&q
->mq_freeze_wq
);
494 static void blk_rq_timed_out_timer(struct timer_list
*t
)
496 struct request_queue
*q
= from_timer(q
, t
, timeout
);
498 kblockd_schedule_work(&q
->timeout_work
);
501 static void blk_timeout_work(struct work_struct
*work
)
505 struct request_queue
*blk_alloc_queue(int node_id
)
507 struct request_queue
*q
;
510 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
511 GFP_KERNEL
| __GFP_ZERO
, node_id
);
515 q
->last_merge
= NULL
;
517 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
521 ret
= bioset_init(&q
->bio_split
, BIO_POOL_SIZE
, 0, 0);
525 q
->stats
= blk_alloc_queue_stats();
531 atomic_set(&q
->nr_active_requests_shared_tags
, 0);
533 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
534 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
535 INIT_LIST_HEAD(&q
->icq_list
);
536 #ifdef CONFIG_BLK_CGROUP
537 INIT_LIST_HEAD(&q
->blkg_list
);
540 kobject_init(&q
->kobj
, &blk_queue_ktype
);
542 mutex_init(&q
->debugfs_mutex
);
543 mutex_init(&q
->sysfs_lock
);
544 mutex_init(&q
->sysfs_dir_lock
);
545 spin_lock_init(&q
->queue_lock
);
547 init_waitqueue_head(&q
->mq_freeze_wq
);
548 mutex_init(&q
->mq_freeze_lock
);
551 * Init percpu_ref in atomic mode so that it's faster to shutdown.
552 * See blk_register_queue() for details.
554 if (percpu_ref_init(&q
->q_usage_counter
,
555 blk_queue_usage_counter_release
,
556 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
559 if (blkcg_init_queue(q
))
562 blk_queue_dma_alignment(q
, 511);
563 blk_set_default_limits(&q
->limits
);
564 q
->nr_requests
= BLKDEV_DEFAULT_RQ
;
569 percpu_ref_exit(&q
->q_usage_counter
);
571 blk_free_queue_stats(q
->stats
);
573 bioset_exit(&q
->bio_split
);
575 ida_simple_remove(&blk_queue_ida
, q
->id
);
577 kmem_cache_free(blk_requestq_cachep
, q
);
582 * blk_get_queue - increment the request_queue refcount
583 * @q: the request_queue structure to increment the refcount for
585 * Increment the refcount of the request_queue kobject.
587 * Context: Any context.
589 bool blk_get_queue(struct request_queue
*q
)
591 if (likely(!blk_queue_dying(q
))) {
598 EXPORT_SYMBOL(blk_get_queue
);
600 static void handle_bad_sector(struct bio
*bio
, sector_t maxsector
)
602 char b
[BDEVNAME_SIZE
];
604 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
605 "%s: rw=%d, want=%llu, limit=%llu\n",
607 bio_devname(bio
, b
), bio
->bi_opf
,
608 bio_end_sector(bio
), maxsector
);
611 #ifdef CONFIG_FAIL_MAKE_REQUEST
613 static DECLARE_FAULT_ATTR(fail_make_request
);
615 static int __init
setup_fail_make_request(char *str
)
617 return setup_fault_attr(&fail_make_request
, str
);
619 __setup("fail_make_request=", setup_fail_make_request
);
621 static bool should_fail_request(struct block_device
*part
, unsigned int bytes
)
623 return part
->bd_make_it_fail
&& should_fail(&fail_make_request
, bytes
);
626 static int __init
fail_make_request_debugfs(void)
628 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
629 NULL
, &fail_make_request
);
631 return PTR_ERR_OR_ZERO(dir
);
634 late_initcall(fail_make_request_debugfs
);
636 #else /* CONFIG_FAIL_MAKE_REQUEST */
638 static inline bool should_fail_request(struct block_device
*part
,
644 #endif /* CONFIG_FAIL_MAKE_REQUEST */
646 static inline bool bio_check_ro(struct bio
*bio
)
648 if (op_is_write(bio_op(bio
)) && bdev_read_only(bio
->bi_bdev
)) {
649 char b
[BDEVNAME_SIZE
];
651 if (op_is_flush(bio
->bi_opf
) && !bio_sectors(bio
))
655 "Trying to write to read-only block-device %s (partno %d)\n",
656 bio_devname(bio
, b
), bio
->bi_bdev
->bd_partno
);
657 /* Older lvm-tools actually trigger this */
664 static noinline
int should_fail_bio(struct bio
*bio
)
666 if (should_fail_request(bdev_whole(bio
->bi_bdev
), bio
->bi_iter
.bi_size
))
670 ALLOW_ERROR_INJECTION(should_fail_bio
, ERRNO
);
673 * Check whether this bio extends beyond the end of the device or partition.
674 * This may well happen - the kernel calls bread() without checking the size of
675 * the device, e.g., when mounting a file system.
677 static inline int bio_check_eod(struct bio
*bio
)
679 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
680 unsigned int nr_sectors
= bio_sectors(bio
);
682 if (nr_sectors
&& maxsector
&&
683 (nr_sectors
> maxsector
||
684 bio
->bi_iter
.bi_sector
> maxsector
- nr_sectors
)) {
685 handle_bad_sector(bio
, maxsector
);
692 * Remap block n of partition p to block n+start(p) of the disk.
694 static int blk_partition_remap(struct bio
*bio
)
696 struct block_device
*p
= bio
->bi_bdev
;
698 if (unlikely(should_fail_request(p
, bio
->bi_iter
.bi_size
)))
700 if (bio_sectors(bio
)) {
701 bio
->bi_iter
.bi_sector
+= p
->bd_start_sect
;
702 trace_block_bio_remap(bio
, p
->bd_dev
,
703 bio
->bi_iter
.bi_sector
-
706 bio_set_flag(bio
, BIO_REMAPPED
);
711 * Check write append to a zoned block device.
713 static inline blk_status_t
blk_check_zone_append(struct request_queue
*q
,
716 sector_t pos
= bio
->bi_iter
.bi_sector
;
717 int nr_sectors
= bio_sectors(bio
);
719 /* Only applicable to zoned block devices */
720 if (!blk_queue_is_zoned(q
))
721 return BLK_STS_NOTSUPP
;
723 /* The bio sector must point to the start of a sequential zone */
724 if (pos
& (blk_queue_zone_sectors(q
) - 1) ||
725 !blk_queue_zone_is_seq(q
, pos
))
726 return BLK_STS_IOERR
;
729 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
730 * split and could result in non-contiguous sectors being written in
733 if (nr_sectors
> q
->limits
.chunk_sectors
)
734 return BLK_STS_IOERR
;
736 /* Make sure the BIO is small enough and will not get split */
737 if (nr_sectors
> q
->limits
.max_zone_append_sectors
)
738 return BLK_STS_IOERR
;
740 bio
->bi_opf
|= REQ_NOMERGE
;
745 static noinline_for_stack
bool submit_bio_checks(struct bio
*bio
)
747 struct block_device
*bdev
= bio
->bi_bdev
;
748 struct request_queue
*q
= bdev_get_queue(bdev
);
749 blk_status_t status
= BLK_STS_IOERR
;
750 struct blk_plug
*plug
;
754 plug
= blk_mq_plug(q
, bio
);
755 if (plug
&& plug
->nowait
)
756 bio
->bi_opf
|= REQ_NOWAIT
;
759 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
760 * if queue does not support NOWAIT.
762 if ((bio
->bi_opf
& REQ_NOWAIT
) && !blk_queue_nowait(q
))
765 if (should_fail_bio(bio
))
767 if (unlikely(bio_check_ro(bio
)))
769 if (!bio_flagged(bio
, BIO_REMAPPED
)) {
770 if (unlikely(bio_check_eod(bio
)))
772 if (bdev
->bd_partno
&& unlikely(blk_partition_remap(bio
)))
777 * Filter flush bio's early so that bio based drivers without flush
778 * support don't have to worry about them.
780 if (op_is_flush(bio
->bi_opf
) &&
781 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
782 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
783 if (!bio_sectors(bio
)) {
789 if (!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
790 bio_clear_polled(bio
);
792 switch (bio_op(bio
)) {
794 if (!blk_queue_discard(q
))
797 case REQ_OP_SECURE_ERASE
:
798 if (!blk_queue_secure_erase(q
))
801 case REQ_OP_WRITE_SAME
:
802 if (!q
->limits
.max_write_same_sectors
)
805 case REQ_OP_ZONE_APPEND
:
806 status
= blk_check_zone_append(q
, bio
);
807 if (status
!= BLK_STS_OK
)
810 case REQ_OP_ZONE_RESET
:
811 case REQ_OP_ZONE_OPEN
:
812 case REQ_OP_ZONE_CLOSE
:
813 case REQ_OP_ZONE_FINISH
:
814 if (!blk_queue_is_zoned(q
))
817 case REQ_OP_ZONE_RESET_ALL
:
818 if (!blk_queue_is_zoned(q
) || !blk_queue_zone_resetall(q
))
821 case REQ_OP_WRITE_ZEROES
:
822 if (!q
->limits
.max_write_zeroes_sectors
)
830 * Various block parts want %current->io_context, so allocate it up
831 * front rather than dealing with lots of pain to allocate it only
832 * where needed. This may fail and the block layer knows how to live
835 if (unlikely(!current
->io_context
))
836 create_task_io_context(current
, GFP_ATOMIC
, q
->node
);
838 if (blk_throtl_bio(bio
)) {
839 blkcg_bio_issue_init(bio
);
843 blk_cgroup_bio_start(bio
);
844 blkcg_bio_issue_init(bio
);
846 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
847 trace_block_bio_queue(bio
);
848 /* Now that enqueuing has been traced, we need to trace
849 * completion as well.
851 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
856 status
= BLK_STS_NOTSUPP
;
858 bio
->bi_status
= status
;
863 static void __submit_bio(struct bio
*bio
)
865 struct gendisk
*disk
= bio
->bi_bdev
->bd_disk
;
867 if (unlikely(bio_queue_enter(bio
) != 0))
870 if (!submit_bio_checks(bio
) || !blk_crypto_bio_prep(&bio
))
872 if (!disk
->fops
->submit_bio
) {
873 blk_mq_submit_bio(bio
);
876 disk
->fops
->submit_bio(bio
);
878 blk_queue_exit(disk
->queue
);
882 * The loop in this function may be a bit non-obvious, and so deserves some
885 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
886 * that), so we have a list with a single bio.
887 * - We pretend that we have just taken it off a longer list, so we assign
888 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
889 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
890 * bios through a recursive call to submit_bio_noacct. If it did, we find a
891 * non-NULL value in bio_list and re-enter the loop from the top.
892 * - In this case we really did just take the bio of the top of the list (no
893 * pretending) and so remove it from bio_list, and call into ->submit_bio()
896 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
897 * bio_list_on_stack[1] contains bios that were submitted before the current
898 * ->submit_bio_bio, but that haven't been processed yet.
900 static void __submit_bio_noacct(struct bio
*bio
)
902 struct bio_list bio_list_on_stack
[2];
904 BUG_ON(bio
->bi_next
);
906 bio_list_init(&bio_list_on_stack
[0]);
907 current
->bio_list
= bio_list_on_stack
;
910 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
911 struct bio_list lower
, same
;
914 * Create a fresh bio_list for all subordinate requests.
916 bio_list_on_stack
[1] = bio_list_on_stack
[0];
917 bio_list_init(&bio_list_on_stack
[0]);
922 * Sort new bios into those for a lower level and those for the
925 bio_list_init(&lower
);
926 bio_list_init(&same
);
927 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
928 if (q
== bdev_get_queue(bio
->bi_bdev
))
929 bio_list_add(&same
, bio
);
931 bio_list_add(&lower
, bio
);
934 * Now assemble so we handle the lowest level first.
936 bio_list_merge(&bio_list_on_stack
[0], &lower
);
937 bio_list_merge(&bio_list_on_stack
[0], &same
);
938 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
939 } while ((bio
= bio_list_pop(&bio_list_on_stack
[0])));
941 current
->bio_list
= NULL
;
944 static void __submit_bio_noacct_mq(struct bio
*bio
)
946 struct bio_list bio_list
[2] = { };
948 current
->bio_list
= bio_list
;
952 } while ((bio
= bio_list_pop(&bio_list
[0])));
954 current
->bio_list
= NULL
;
958 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
959 * @bio: The bio describing the location in memory and on the device.
961 * This is a version of submit_bio() that shall only be used for I/O that is
962 * resubmitted to lower level drivers by stacking block drivers. All file
963 * systems and other upper level users of the block layer should use
964 * submit_bio() instead.
966 void submit_bio_noacct(struct bio
*bio
)
969 * We only want one ->submit_bio to be active at a time, else stack
970 * usage with stacked devices could be a problem. Use current->bio_list
971 * to collect a list of requests submited by a ->submit_bio method while
972 * it is active, and then process them after it returned.
974 if (current
->bio_list
)
975 bio_list_add(¤t
->bio_list
[0], bio
);
976 else if (!bio
->bi_bdev
->bd_disk
->fops
->submit_bio
)
977 __submit_bio_noacct_mq(bio
);
979 __submit_bio_noacct(bio
);
981 EXPORT_SYMBOL(submit_bio_noacct
);
984 * submit_bio - submit a bio to the block device layer for I/O
985 * @bio: The &struct bio which describes the I/O
987 * submit_bio() is used to submit I/O requests to block devices. It is passed a
988 * fully set up &struct bio that describes the I/O that needs to be done. The
989 * bio will be send to the device described by the bi_bdev field.
991 * The success/failure status of the request, along with notification of
992 * completion, is delivered asynchronously through the ->bi_end_io() callback
993 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
996 void submit_bio(struct bio
*bio
)
998 if (blkcg_punt_bio_submit(bio
))
1002 * If it's a regular read/write or a barrier with data attached,
1003 * go through the normal accounting stuff before submission.
1005 if (bio_has_data(bio
)) {
1008 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1009 count
= queue_logical_block_size(
1010 bdev_get_queue(bio
->bi_bdev
)) >> 9;
1012 count
= bio_sectors(bio
);
1014 if (op_is_write(bio_op(bio
))) {
1015 count_vm_events(PGPGOUT
, count
);
1017 task_io_account_read(bio
->bi_iter
.bi_size
);
1018 count_vm_events(PGPGIN
, count
);
1023 * If we're reading data that is part of the userspace workingset, count
1024 * submission time as memory stall. When the device is congested, or
1025 * the submitting cgroup IO-throttled, submission can be a significant
1026 * part of overall IO time.
1028 if (unlikely(bio_op(bio
) == REQ_OP_READ
&&
1029 bio_flagged(bio
, BIO_WORKINGSET
))) {
1030 unsigned long pflags
;
1032 psi_memstall_enter(&pflags
);
1033 submit_bio_noacct(bio
);
1034 psi_memstall_leave(&pflags
);
1038 submit_bio_noacct(bio
);
1040 EXPORT_SYMBOL(submit_bio
);
1043 * bio_poll - poll for BIO completions
1044 * @bio: bio to poll for
1045 * @flags: BLK_POLL_* flags that control the behavior
1047 * Poll for completions on queue associated with the bio. Returns number of
1048 * completed entries found.
1050 * Note: the caller must either be the context that submitted @bio, or
1051 * be in a RCU critical section to prevent freeing of @bio.
1053 int bio_poll(struct bio
*bio
, struct io_comp_batch
*iob
, unsigned int flags
)
1055 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1056 blk_qc_t cookie
= READ_ONCE(bio
->bi_cookie
);
1059 if (cookie
== BLK_QC_T_NONE
||
1060 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
1064 blk_flush_plug(current
->plug
, false);
1066 if (blk_queue_enter(q
, BLK_MQ_REQ_NOWAIT
))
1068 if (WARN_ON_ONCE(!queue_is_mq(q
)))
1069 ret
= 0; /* not yet implemented, should not happen */
1071 ret
= blk_mq_poll(q
, cookie
, iob
, flags
);
1075 EXPORT_SYMBOL_GPL(bio_poll
);
1078 * Helper to implement file_operations.iopoll. Requires the bio to be stored
1079 * in iocb->private, and cleared before freeing the bio.
1081 int iocb_bio_iopoll(struct kiocb
*kiocb
, struct io_comp_batch
*iob
,
1088 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
1089 * point to a freshly allocated bio at this point. If that happens
1090 * we have a few cases to consider:
1092 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
1093 * simply nothing in this case
1094 * 2) the bio points to a not poll enabled device. bio_poll will catch
1096 * 3) the bio points to a poll capable device, including but not
1097 * limited to the one that the original bio pointed to. In this
1098 * case we will call into the actual poll method and poll for I/O,
1099 * even if we don't need to, but it won't cause harm either.
1101 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
1102 * is still allocated. Because partitions hold a reference to the whole
1103 * device bdev and thus disk, the disk is also still valid. Grabbing
1104 * a reference to the queue in bio_poll() ensures the hctxs and requests
1105 * are still valid as well.
1108 bio
= READ_ONCE(kiocb
->private);
1109 if (bio
&& bio
->bi_bdev
)
1110 ret
= bio_poll(bio
, iob
, flags
);
1115 EXPORT_SYMBOL_GPL(iocb_bio_iopoll
);
1118 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1119 * for the new queue limits
1121 * @rq: the request being checked
1124 * @rq may have been made based on weaker limitations of upper-level queues
1125 * in request stacking drivers, and it may violate the limitation of @q.
1126 * Since the block layer and the underlying device driver trust @rq
1127 * after it is inserted to @q, it should be checked against @q before
1128 * the insertion using this generic function.
1130 * Request stacking drivers like request-based dm may change the queue
1131 * limits when retrying requests on other queues. Those requests need
1132 * to be checked against the new queue limits again during dispatch.
1134 static blk_status_t
blk_cloned_rq_check_limits(struct request_queue
*q
,
1137 unsigned int max_sectors
= blk_queue_get_max_sectors(q
, req_op(rq
));
1139 if (blk_rq_sectors(rq
) > max_sectors
) {
1141 * SCSI device does not have a good way to return if
1142 * Write Same/Zero is actually supported. If a device rejects
1143 * a non-read/write command (discard, write same,etc.) the
1144 * low-level device driver will set the relevant queue limit to
1145 * 0 to prevent blk-lib from issuing more of the offending
1146 * operations. Commands queued prior to the queue limit being
1147 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1148 * errors being propagated to upper layers.
1150 if (max_sectors
== 0)
1151 return BLK_STS_NOTSUPP
;
1153 printk(KERN_ERR
"%s: over max size limit. (%u > %u)\n",
1154 __func__
, blk_rq_sectors(rq
), max_sectors
);
1155 return BLK_STS_IOERR
;
1159 * The queue settings related to segment counting may differ from the
1162 rq
->nr_phys_segments
= blk_recalc_rq_segments(rq
);
1163 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1164 printk(KERN_ERR
"%s: over max segments limit. (%hu > %hu)\n",
1165 __func__
, rq
->nr_phys_segments
, queue_max_segments(q
));
1166 return BLK_STS_IOERR
;
1173 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1174 * @q: the queue to submit the request
1175 * @rq: the request being queued
1177 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1181 ret
= blk_cloned_rq_check_limits(q
, rq
);
1182 if (ret
!= BLK_STS_OK
)
1186 should_fail_request(rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1187 return BLK_STS_IOERR
;
1189 if (blk_crypto_insert_cloned_request(rq
))
1190 return BLK_STS_IOERR
;
1192 blk_account_io_start(rq
);
1195 * Since we have a scheduler attached on the top device,
1196 * bypass a potential scheduler on the bottom device for
1199 return blk_mq_request_issue_directly(rq
, true);
1201 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1204 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1205 * @rq: request to examine
1208 * A request could be merge of IOs which require different failure
1209 * handling. This function determines the number of bytes which
1210 * can be failed from the beginning of the request without
1211 * crossing into area which need to be retried further.
1214 * The number of bytes to fail.
1216 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1218 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1219 unsigned int bytes
= 0;
1222 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
1223 return blk_rq_bytes(rq
);
1226 * Currently the only 'mixing' which can happen is between
1227 * different fastfail types. We can safely fail portions
1228 * which have all the failfast bits that the first one has -
1229 * the ones which are at least as eager to fail as the first
1232 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1233 if ((bio
->bi_opf
& ff
) != ff
)
1235 bytes
+= bio
->bi_iter
.bi_size
;
1238 /* this could lead to infinite loop */
1239 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1242 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1244 static void update_io_ticks(struct block_device
*part
, unsigned long now
,
1247 unsigned long stamp
;
1249 stamp
= READ_ONCE(part
->bd_stamp
);
1250 if (unlikely(time_after(now
, stamp
))) {
1251 if (likely(cmpxchg(&part
->bd_stamp
, stamp
, now
) == stamp
))
1252 __part_stat_add(part
, io_ticks
, end
? now
- stamp
: 1);
1254 if (part
->bd_partno
) {
1255 part
= bdev_whole(part
);
1260 void __blk_account_io_done(struct request
*req
, u64 now
)
1262 const int sgrp
= op_stat_group(req_op(req
));
1265 update_io_ticks(req
->part
, jiffies
, true);
1266 part_stat_inc(req
->part
, ios
[sgrp
]);
1267 part_stat_add(req
->part
, nsecs
[sgrp
], now
- req
->start_time_ns
);
1271 void __blk_account_io_start(struct request
*rq
)
1273 /* passthrough requests can hold bios that do not have ->bi_bdev set */
1274 if (rq
->bio
&& rq
->bio
->bi_bdev
)
1275 rq
->part
= rq
->bio
->bi_bdev
;
1277 rq
->part
= rq
->rq_disk
->part0
;
1280 update_io_ticks(rq
->part
, jiffies
, false);
1284 static unsigned long __part_start_io_acct(struct block_device
*part
,
1285 unsigned int sectors
, unsigned int op
)
1287 const int sgrp
= op_stat_group(op
);
1288 unsigned long now
= READ_ONCE(jiffies
);
1291 update_io_ticks(part
, now
, false);
1292 part_stat_inc(part
, ios
[sgrp
]);
1293 part_stat_add(part
, sectors
[sgrp
], sectors
);
1294 part_stat_local_inc(part
, in_flight
[op_is_write(op
)]);
1301 * bio_start_io_acct - start I/O accounting for bio based drivers
1302 * @bio: bio to start account for
1304 * Returns the start time that should be passed back to bio_end_io_acct().
1306 unsigned long bio_start_io_acct(struct bio
*bio
)
1308 return __part_start_io_acct(bio
->bi_bdev
, bio_sectors(bio
), bio_op(bio
));
1310 EXPORT_SYMBOL_GPL(bio_start_io_acct
);
1312 unsigned long disk_start_io_acct(struct gendisk
*disk
, unsigned int sectors
,
1315 return __part_start_io_acct(disk
->part0
, sectors
, op
);
1317 EXPORT_SYMBOL(disk_start_io_acct
);
1319 static void __part_end_io_acct(struct block_device
*part
, unsigned int op
,
1320 unsigned long start_time
)
1322 const int sgrp
= op_stat_group(op
);
1323 unsigned long now
= READ_ONCE(jiffies
);
1324 unsigned long duration
= now
- start_time
;
1327 update_io_ticks(part
, now
, true);
1328 part_stat_add(part
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1329 part_stat_local_dec(part
, in_flight
[op_is_write(op
)]);
1333 void bio_end_io_acct_remapped(struct bio
*bio
, unsigned long start_time
,
1334 struct block_device
*orig_bdev
)
1336 __part_end_io_acct(orig_bdev
, bio_op(bio
), start_time
);
1338 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped
);
1340 void disk_end_io_acct(struct gendisk
*disk
, unsigned int op
,
1341 unsigned long start_time
)
1343 __part_end_io_acct(disk
->part0
, op
, start_time
);
1345 EXPORT_SYMBOL(disk_end_io_acct
);
1348 * Steal bios from a request and add them to a bio list.
1349 * The request must not have been partially completed before.
1351 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
1355 list
->tail
->bi_next
= rq
->bio
;
1357 list
->head
= rq
->bio
;
1358 list
->tail
= rq
->biotail
;
1366 EXPORT_SYMBOL_GPL(blk_steal_bios
);
1368 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1370 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1371 * @rq: the request to be flushed
1374 * Flush all pages in @rq.
1376 void rq_flush_dcache_pages(struct request
*rq
)
1378 struct req_iterator iter
;
1379 struct bio_vec bvec
;
1381 rq_for_each_segment(bvec
, rq
, iter
)
1382 flush_dcache_page(bvec
.bv_page
);
1384 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
1388 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1389 * @q : the queue of the device being checked
1392 * Check if underlying low-level drivers of a device are busy.
1393 * If the drivers want to export their busy state, they must set own
1394 * exporting function using blk_queue_lld_busy() first.
1396 * Basically, this function is used only by request stacking drivers
1397 * to stop dispatching requests to underlying devices when underlying
1398 * devices are busy. This behavior helps more I/O merging on the queue
1399 * of the request stacking driver and prevents I/O throughput regression
1400 * on burst I/O load.
1403 * 0 - Not busy (The request stacking driver should dispatch request)
1404 * 1 - Busy (The request stacking driver should stop dispatching request)
1406 int blk_lld_busy(struct request_queue
*q
)
1408 if (queue_is_mq(q
) && q
->mq_ops
->busy
)
1409 return q
->mq_ops
->busy(q
);
1413 EXPORT_SYMBOL_GPL(blk_lld_busy
);
1416 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1417 * @rq: the clone request to be cleaned up
1420 * Free all bios in @rq for a cloned request.
1422 void blk_rq_unprep_clone(struct request
*rq
)
1426 while ((bio
= rq
->bio
) != NULL
) {
1427 rq
->bio
= bio
->bi_next
;
1432 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
1435 * blk_rq_prep_clone - Helper function to setup clone request
1436 * @rq: the request to be setup
1437 * @rq_src: original request to be cloned
1438 * @bs: bio_set that bios for clone are allocated from
1439 * @gfp_mask: memory allocation mask for bio
1440 * @bio_ctr: setup function to be called for each clone bio.
1441 * Returns %0 for success, non %0 for failure.
1442 * @data: private data to be passed to @bio_ctr
1445 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1446 * Also, pages which the original bios are pointing to are not copied
1447 * and the cloned bios just point same pages.
1448 * So cloned bios must be completed before original bios, which means
1449 * the caller must complete @rq before @rq_src.
1451 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
1452 struct bio_set
*bs
, gfp_t gfp_mask
,
1453 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
1456 struct bio
*bio
, *bio_src
;
1461 __rq_for_each_bio(bio_src
, rq_src
) {
1462 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
1466 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
1470 rq
->biotail
->bi_next
= bio
;
1473 rq
->bio
= rq
->biotail
= bio
;
1478 /* Copy attributes of the original request to the clone request. */
1479 rq
->__sector
= blk_rq_pos(rq_src
);
1480 rq
->__data_len
= blk_rq_bytes(rq_src
);
1481 if (rq_src
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
1482 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
1483 rq
->special_vec
= rq_src
->special_vec
;
1485 rq
->nr_phys_segments
= rq_src
->nr_phys_segments
;
1486 rq
->ioprio
= rq_src
->ioprio
;
1488 if (rq
->bio
&& blk_crypto_rq_bio_prep(rq
, rq
->bio
, gfp_mask
) < 0)
1496 blk_rq_unprep_clone(rq
);
1500 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
1502 int kblockd_schedule_work(struct work_struct
*work
)
1504 return queue_work(kblockd_workqueue
, work
);
1506 EXPORT_SYMBOL(kblockd_schedule_work
);
1508 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
1509 unsigned long delay
)
1511 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
1513 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
1515 void blk_start_plug_nr_ios(struct blk_plug
*plug
, unsigned short nr_ios
)
1517 struct task_struct
*tsk
= current
;
1520 * If this is a nested plug, don't actually assign it.
1525 plug
->mq_list
= NULL
;
1526 plug
->cached_rq
= NULL
;
1527 plug
->nr_ios
= min_t(unsigned short, nr_ios
, BLK_MAX_REQUEST_COUNT
);
1529 plug
->multiple_queues
= false;
1530 plug
->has_elevator
= false;
1531 plug
->nowait
= false;
1532 INIT_LIST_HEAD(&plug
->cb_list
);
1535 * Store ordering should not be needed here, since a potential
1536 * preempt will imply a full memory barrier
1542 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1543 * @plug: The &struct blk_plug that needs to be initialized
1546 * blk_start_plug() indicates to the block layer an intent by the caller
1547 * to submit multiple I/O requests in a batch. The block layer may use
1548 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1549 * is called. However, the block layer may choose to submit requests
1550 * before a call to blk_finish_plug() if the number of queued I/Os
1551 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1552 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1553 * the task schedules (see below).
1555 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1556 * pending I/O should the task end up blocking between blk_start_plug() and
1557 * blk_finish_plug(). This is important from a performance perspective, but
1558 * also ensures that we don't deadlock. For instance, if the task is blocking
1559 * for a memory allocation, memory reclaim could end up wanting to free a
1560 * page belonging to that request that is currently residing in our private
1561 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1562 * this kind of deadlock.
1564 void blk_start_plug(struct blk_plug
*plug
)
1566 blk_start_plug_nr_ios(plug
, 1);
1568 EXPORT_SYMBOL(blk_start_plug
);
1570 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
1572 LIST_HEAD(callbacks
);
1574 while (!list_empty(&plug
->cb_list
)) {
1575 list_splice_init(&plug
->cb_list
, &callbacks
);
1577 while (!list_empty(&callbacks
)) {
1578 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
1581 list_del(&cb
->list
);
1582 cb
->callback(cb
, from_schedule
);
1587 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
1590 struct blk_plug
*plug
= current
->plug
;
1591 struct blk_plug_cb
*cb
;
1596 list_for_each_entry(cb
, &plug
->cb_list
, list
)
1597 if (cb
->callback
== unplug
&& cb
->data
== data
)
1600 /* Not currently on the callback list */
1601 BUG_ON(size
< sizeof(*cb
));
1602 cb
= kzalloc(size
, GFP_ATOMIC
);
1605 cb
->callback
= unplug
;
1606 list_add(&cb
->list
, &plug
->cb_list
);
1610 EXPORT_SYMBOL(blk_check_plugged
);
1612 void blk_flush_plug(struct blk_plug
*plug
, bool from_schedule
)
1614 if (!list_empty(&plug
->cb_list
))
1615 flush_plug_callbacks(plug
, from_schedule
);
1616 if (!rq_list_empty(plug
->mq_list
))
1617 blk_mq_flush_plug_list(plug
, from_schedule
);
1618 if (unlikely(!from_schedule
&& plug
->cached_rq
))
1619 blk_mq_free_plug_rqs(plug
);
1623 * blk_finish_plug - mark the end of a batch of submitted I/O
1624 * @plug: The &struct blk_plug passed to blk_start_plug()
1627 * Indicate that a batch of I/O submissions is complete. This function
1628 * must be paired with an initial call to blk_start_plug(). The intent
1629 * is to allow the block layer to optimize I/O submission. See the
1630 * documentation for blk_start_plug() for more information.
1632 void blk_finish_plug(struct blk_plug
*plug
)
1634 if (plug
== current
->plug
) {
1635 blk_flush_plug(plug
, false);
1636 current
->plug
= NULL
;
1639 EXPORT_SYMBOL(blk_finish_plug
);
1641 void blk_io_schedule(void)
1643 /* Prevent hang_check timer from firing at us during very long I/O */
1644 unsigned long timeout
= sysctl_hung_task_timeout_secs
* HZ
/ 2;
1647 io_schedule_timeout(timeout
);
1651 EXPORT_SYMBOL_GPL(blk_io_schedule
);
1653 int __init
blk_dev_init(void)
1655 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
1656 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1657 sizeof_field(struct request
, cmd_flags
));
1658 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1659 sizeof_field(struct bio
, bi_opf
));
1661 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1662 kblockd_workqueue
= alloc_workqueue("kblockd",
1663 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
1664 if (!kblockd_workqueue
)
1665 panic("Failed to create kblockd\n");
1667 blk_requestq_cachep
= kmem_cache_create("request_queue",
1668 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
1670 blk_debugfs_root
= debugfs_create_dir("block", NULL
);