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/backing-dev.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-mq.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/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
50 #include "blk-mq-sched.h"
52 #include "blk-rq-qos.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
);
62 DEFINE_IDA(blk_queue_ida
);
65 * For queue allocation
67 struct kmem_cache
*blk_requestq_cachep
;
70 * Controlling structure to kblockd
72 static struct workqueue_struct
*kblockd_workqueue
;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag
, struct request_queue
*q
)
81 set_bit(flag
, &q
->queue_flags
);
83 EXPORT_SYMBOL(blk_queue_flag_set
);
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
90 void blk_queue_flag_clear(unsigned int flag
, struct request_queue
*q
)
92 clear_bit(flag
, &q
->queue_flags
);
94 EXPORT_SYMBOL(blk_queue_flag_clear
);
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
104 bool blk_queue_flag_test_and_set(unsigned int flag
, struct request_queue
*q
)
106 return test_and_set_bit(flag
, &q
->queue_flags
);
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set
);
110 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
112 memset(rq
, 0, sizeof(*rq
));
114 INIT_LIST_HEAD(&rq
->queuelist
);
116 rq
->__sector
= (sector_t
) -1;
117 INIT_HLIST_NODE(&rq
->hash
);
118 RB_CLEAR_NODE(&rq
->rb_node
);
119 rq
->tag
= BLK_MQ_NO_TAG
;
120 rq
->internal_tag
= BLK_MQ_NO_TAG
;
121 rq
->start_time_ns
= ktime_get_ns();
123 refcount_set(&rq
->ref
, 1);
124 blk_crypto_rq_set_defaults(rq
);
126 EXPORT_SYMBOL(blk_rq_init
);
128 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129 static const char *const blk_op_name
[] = {
133 REQ_OP_NAME(DISCARD
),
134 REQ_OP_NAME(SECURE_ERASE
),
135 REQ_OP_NAME(ZONE_RESET
),
136 REQ_OP_NAME(ZONE_RESET_ALL
),
137 REQ_OP_NAME(ZONE_OPEN
),
138 REQ_OP_NAME(ZONE_CLOSE
),
139 REQ_OP_NAME(ZONE_FINISH
),
140 REQ_OP_NAME(ZONE_APPEND
),
141 REQ_OP_NAME(WRITE_SAME
),
142 REQ_OP_NAME(WRITE_ZEROES
),
143 REQ_OP_NAME(SCSI_IN
),
144 REQ_OP_NAME(SCSI_OUT
),
146 REQ_OP_NAME(DRV_OUT
),
151 * blk_op_str - Return string XXX in the REQ_OP_XXX.
154 * Description: Centralize block layer function to convert REQ_OP_XXX into
155 * string format. Useful in the debugging and tracing bio or request. For
156 * invalid REQ_OP_XXX it returns string "UNKNOWN".
158 inline const char *blk_op_str(unsigned int op
)
160 const char *op_str
= "UNKNOWN";
162 if (op
< ARRAY_SIZE(blk_op_name
) && blk_op_name
[op
])
163 op_str
= blk_op_name
[op
];
167 EXPORT_SYMBOL_GPL(blk_op_str
);
169 static const struct {
173 [BLK_STS_OK
] = { 0, "" },
174 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
175 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
176 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
177 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
178 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
179 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
180 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
181 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
182 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
183 [BLK_STS_DEV_RESOURCE
] = { -EBUSY
, "device resource" },
184 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
186 /* device mapper special case, should not leak out: */
187 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
189 /* everything else not covered above: */
190 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
193 blk_status_t
errno_to_blk_status(int errno
)
197 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
198 if (blk_errors
[i
].errno
== errno
)
199 return (__force blk_status_t
)i
;
202 return BLK_STS_IOERR
;
204 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
206 int blk_status_to_errno(blk_status_t status
)
208 int idx
= (__force
int)status
;
210 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
212 return blk_errors
[idx
].errno
;
214 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
216 static void print_req_error(struct request
*req
, blk_status_t status
,
219 int idx
= (__force
int)status
;
221 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
224 printk_ratelimited(KERN_ERR
225 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
226 "phys_seg %u prio class %u\n",
227 caller
, blk_errors
[idx
].name
,
228 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
229 blk_rq_pos(req
), req_op(req
), blk_op_str(req_op(req
)),
230 req
->cmd_flags
& ~REQ_OP_MASK
,
231 req
->nr_phys_segments
,
232 IOPRIO_PRIO_CLASS(req
->ioprio
));
235 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
236 unsigned int nbytes
, blk_status_t error
)
239 bio
->bi_status
= error
;
241 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
242 bio_set_flag(bio
, BIO_QUIET
);
244 bio_advance(bio
, nbytes
);
246 if (req_op(rq
) == REQ_OP_ZONE_APPEND
&& error
== BLK_STS_OK
) {
248 * Partial zone append completions cannot be supported as the
249 * BIO fragments may end up not being written sequentially.
251 if (bio
->bi_iter
.bi_size
)
252 bio
->bi_status
= BLK_STS_IOERR
;
254 bio
->bi_iter
.bi_sector
= rq
->__sector
;
257 /* don't actually finish bio if it's part of flush sequence */
258 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
262 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
264 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
265 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
266 (unsigned long long) rq
->cmd_flags
);
268 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
269 (unsigned long long)blk_rq_pos(rq
),
270 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
271 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
272 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
274 EXPORT_SYMBOL(blk_dump_rq_flags
);
277 * blk_sync_queue - cancel any pending callbacks on a queue
281 * The block layer may perform asynchronous callback activity
282 * on a queue, such as calling the unplug function after a timeout.
283 * A block device may call blk_sync_queue to ensure that any
284 * such activity is cancelled, thus allowing it to release resources
285 * that the callbacks might use. The caller must already have made sure
286 * that its ->submit_bio will not re-add plugging prior to calling
289 * This function does not cancel any asynchronous activity arising
290 * out of elevator or throttling code. That would require elevator_exit()
291 * and blkcg_exit_queue() to be called with queue lock initialized.
294 void blk_sync_queue(struct request_queue
*q
)
296 del_timer_sync(&q
->timeout
);
297 cancel_work_sync(&q
->timeout_work
);
299 EXPORT_SYMBOL(blk_sync_queue
);
302 * blk_set_pm_only - increment pm_only counter
303 * @q: request queue pointer
305 void blk_set_pm_only(struct request_queue
*q
)
307 atomic_inc(&q
->pm_only
);
309 EXPORT_SYMBOL_GPL(blk_set_pm_only
);
311 void blk_clear_pm_only(struct request_queue
*q
)
315 pm_only
= atomic_dec_return(&q
->pm_only
);
316 WARN_ON_ONCE(pm_only
< 0);
318 wake_up_all(&q
->mq_freeze_wq
);
320 EXPORT_SYMBOL_GPL(blk_clear_pm_only
);
323 * blk_put_queue - decrement the request_queue refcount
324 * @q: the request_queue structure to decrement the refcount for
326 * Decrements the refcount of the request_queue kobject. When this reaches 0
327 * we'll have blk_release_queue() called.
329 * Context: Any context, but the last reference must not be dropped from
332 void blk_put_queue(struct request_queue
*q
)
334 kobject_put(&q
->kobj
);
336 EXPORT_SYMBOL(blk_put_queue
);
338 void blk_set_queue_dying(struct request_queue
*q
)
340 blk_queue_flag_set(QUEUE_FLAG_DYING
, q
);
343 * When queue DYING flag is set, we need to block new req
344 * entering queue, so we call blk_freeze_queue_start() to
345 * prevent I/O from crossing blk_queue_enter().
347 blk_freeze_queue_start(q
);
350 blk_mq_wake_waiters(q
);
352 /* Make blk_queue_enter() reexamine the DYING flag. */
353 wake_up_all(&q
->mq_freeze_wq
);
355 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
358 * blk_cleanup_queue - shutdown a request queue
359 * @q: request queue to shutdown
361 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
362 * put it. All future requests will be failed immediately with -ENODEV.
366 void blk_cleanup_queue(struct request_queue
*q
)
368 /* cannot be called from atomic context */
371 WARN_ON_ONCE(blk_queue_registered(q
));
373 /* mark @q DYING, no new request or merges will be allowed afterwards */
374 blk_set_queue_dying(q
);
376 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
377 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
380 * Drain all requests queued before DYING marking. Set DEAD flag to
381 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
382 * after draining finished.
388 blk_queue_flag_set(QUEUE_FLAG_DEAD
, q
);
390 /* for synchronous bio-based driver finish in-flight integrity i/o */
391 blk_flush_integrity();
393 /* @q won't process any more request, flush async actions */
394 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
398 blk_mq_exit_queue(q
);
401 * In theory, request pool of sched_tags belongs to request queue.
402 * However, the current implementation requires tag_set for freeing
403 * requests, so free the pool now.
405 * Queue has become frozen, there can't be any in-queue requests, so
406 * it is safe to free requests now.
408 mutex_lock(&q
->sysfs_lock
);
410 blk_mq_sched_free_requests(q
);
411 mutex_unlock(&q
->sysfs_lock
);
413 percpu_ref_exit(&q
->q_usage_counter
);
415 /* @q is and will stay empty, shutdown and put */
418 EXPORT_SYMBOL(blk_cleanup_queue
);
421 * blk_queue_enter() - try to increase q->q_usage_counter
422 * @q: request queue pointer
423 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
425 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
427 const bool pm
= flags
& BLK_MQ_REQ_PREEMPT
;
430 bool success
= false;
433 if (percpu_ref_tryget_live(&q
->q_usage_counter
)) {
435 * The code that increments the pm_only counter is
436 * responsible for ensuring that that counter is
437 * globally visible before the queue is unfrozen.
439 if (pm
|| !blk_queue_pm_only(q
)) {
442 percpu_ref_put(&q
->q_usage_counter
);
450 if (flags
& BLK_MQ_REQ_NOWAIT
)
454 * read pair of barrier in blk_freeze_queue_start(),
455 * we need to order reading __PERCPU_REF_DEAD flag of
456 * .q_usage_counter and reading .mq_freeze_depth or
457 * queue dying flag, otherwise the following wait may
458 * never return if the two reads are reordered.
462 wait_event(q
->mq_freeze_wq
,
463 (!q
->mq_freeze_depth
&&
464 (pm
|| (blk_pm_request_resume(q
),
465 !blk_queue_pm_only(q
)))) ||
467 if (blk_queue_dying(q
))
472 static inline int bio_queue_enter(struct bio
*bio
)
474 struct request_queue
*q
= bio
->bi_disk
->queue
;
475 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
478 ret
= blk_queue_enter(q
, nowait
? BLK_MQ_REQ_NOWAIT
: 0);
480 if (nowait
&& !blk_queue_dying(q
))
481 bio_wouldblock_error(bio
);
489 void blk_queue_exit(struct request_queue
*q
)
491 percpu_ref_put(&q
->q_usage_counter
);
494 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
496 struct request_queue
*q
=
497 container_of(ref
, struct request_queue
, q_usage_counter
);
499 wake_up_all(&q
->mq_freeze_wq
);
502 static void blk_rq_timed_out_timer(struct timer_list
*t
)
504 struct request_queue
*q
= from_timer(q
, t
, timeout
);
506 kblockd_schedule_work(&q
->timeout_work
);
509 static void blk_timeout_work(struct work_struct
*work
)
513 struct request_queue
*blk_alloc_queue(int node_id
)
515 struct request_queue
*q
;
518 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
519 GFP_KERNEL
| __GFP_ZERO
, node_id
);
523 q
->last_merge
= NULL
;
525 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
529 ret
= bioset_init(&q
->bio_split
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
533 q
->backing_dev_info
= bdi_alloc(node_id
);
534 if (!q
->backing_dev_info
)
537 q
->stats
= blk_alloc_queue_stats();
543 atomic_set(&q
->nr_active_requests_shared_sbitmap
, 0);
545 timer_setup(&q
->backing_dev_info
->laptop_mode_wb_timer
,
546 laptop_mode_timer_fn
, 0);
547 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
548 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
549 INIT_LIST_HEAD(&q
->icq_list
);
550 #ifdef CONFIG_BLK_CGROUP
551 INIT_LIST_HEAD(&q
->blkg_list
);
554 kobject_init(&q
->kobj
, &blk_queue_ktype
);
556 mutex_init(&q
->debugfs_mutex
);
557 mutex_init(&q
->sysfs_lock
);
558 mutex_init(&q
->sysfs_dir_lock
);
559 spin_lock_init(&q
->queue_lock
);
561 init_waitqueue_head(&q
->mq_freeze_wq
);
562 mutex_init(&q
->mq_freeze_lock
);
565 * Init percpu_ref in atomic mode so that it's faster to shutdown.
566 * See blk_register_queue() for details.
568 if (percpu_ref_init(&q
->q_usage_counter
,
569 blk_queue_usage_counter_release
,
570 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
573 if (blkcg_init_queue(q
))
576 blk_queue_dma_alignment(q
, 511);
577 blk_set_default_limits(&q
->limits
);
578 q
->nr_requests
= BLKDEV_MAX_RQ
;
583 percpu_ref_exit(&q
->q_usage_counter
);
585 blk_free_queue_stats(q
->stats
);
587 bdi_put(q
->backing_dev_info
);
589 bioset_exit(&q
->bio_split
);
591 ida_simple_remove(&blk_queue_ida
, q
->id
);
593 kmem_cache_free(blk_requestq_cachep
, q
);
596 EXPORT_SYMBOL(blk_alloc_queue
);
599 * blk_get_queue - increment the request_queue refcount
600 * @q: the request_queue structure to increment the refcount for
602 * Increment the refcount of the request_queue kobject.
604 * Context: Any context.
606 bool blk_get_queue(struct request_queue
*q
)
608 if (likely(!blk_queue_dying(q
))) {
615 EXPORT_SYMBOL(blk_get_queue
);
618 * blk_get_request - allocate a request
619 * @q: request queue to allocate a request for
620 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
621 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
623 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
624 blk_mq_req_flags_t flags
)
628 WARN_ON_ONCE(op
& REQ_NOWAIT
);
629 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
631 req
= blk_mq_alloc_request(q
, op
, flags
);
632 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
633 q
->mq_ops
->initialize_rq_fn(req
);
637 EXPORT_SYMBOL(blk_get_request
);
639 void blk_put_request(struct request
*req
)
641 blk_mq_free_request(req
);
643 EXPORT_SYMBOL(blk_put_request
);
645 static void handle_bad_sector(struct bio
*bio
, sector_t maxsector
)
647 char b
[BDEVNAME_SIZE
];
649 pr_info_ratelimited("attempt to access beyond end of device\n"
650 "%s: rw=%d, want=%llu, limit=%llu\n",
651 bio_devname(bio
, b
), bio
->bi_opf
,
652 bio_end_sector(bio
), maxsector
);
655 #ifdef CONFIG_FAIL_MAKE_REQUEST
657 static DECLARE_FAULT_ATTR(fail_make_request
);
659 static int __init
setup_fail_make_request(char *str
)
661 return setup_fault_attr(&fail_make_request
, str
);
663 __setup("fail_make_request=", setup_fail_make_request
);
665 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
667 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
670 static int __init
fail_make_request_debugfs(void)
672 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
673 NULL
, &fail_make_request
);
675 return PTR_ERR_OR_ZERO(dir
);
678 late_initcall(fail_make_request_debugfs
);
680 #else /* CONFIG_FAIL_MAKE_REQUEST */
682 static inline bool should_fail_request(struct hd_struct
*part
,
688 #endif /* CONFIG_FAIL_MAKE_REQUEST */
690 static inline bool bio_check_ro(struct bio
*bio
, struct hd_struct
*part
)
692 const int op
= bio_op(bio
);
694 if (part
->policy
&& op_is_write(op
)) {
695 char b
[BDEVNAME_SIZE
];
697 if (op_is_flush(bio
->bi_opf
) && !bio_sectors(bio
))
701 "Trying to write to read-only block-device %s (partno %d)\n",
702 bio_devname(bio
, b
), part
->partno
);
703 /* Older lvm-tools actually trigger this */
710 static noinline
int should_fail_bio(struct bio
*bio
)
712 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
716 ALLOW_ERROR_INJECTION(should_fail_bio
, ERRNO
);
719 * Check whether this bio extends beyond the end of the device or partition.
720 * This may well happen - the kernel calls bread() without checking the size of
721 * the device, e.g., when mounting a file system.
723 static inline int bio_check_eod(struct bio
*bio
, sector_t maxsector
)
725 unsigned int nr_sectors
= bio_sectors(bio
);
727 if (nr_sectors
&& maxsector
&&
728 (nr_sectors
> maxsector
||
729 bio
->bi_iter
.bi_sector
> maxsector
- nr_sectors
)) {
730 handle_bad_sector(bio
, maxsector
);
737 * Remap block n of partition p to block n+start(p) of the disk.
739 static inline int blk_partition_remap(struct bio
*bio
)
745 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
748 if (unlikely(should_fail_request(p
, bio
->bi_iter
.bi_size
)))
750 if (unlikely(bio_check_ro(bio
, p
)))
753 if (bio_sectors(bio
)) {
754 if (bio_check_eod(bio
, part_nr_sects_read(p
)))
756 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
757 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
758 bio
->bi_iter
.bi_sector
- p
->start_sect
);
768 * Check write append to a zoned block device.
770 static inline blk_status_t
blk_check_zone_append(struct request_queue
*q
,
773 sector_t pos
= bio
->bi_iter
.bi_sector
;
774 int nr_sectors
= bio_sectors(bio
);
776 /* Only applicable to zoned block devices */
777 if (!blk_queue_is_zoned(q
))
778 return BLK_STS_NOTSUPP
;
780 /* The bio sector must point to the start of a sequential zone */
781 if (pos
& (blk_queue_zone_sectors(q
) - 1) ||
782 !blk_queue_zone_is_seq(q
, pos
))
783 return BLK_STS_IOERR
;
786 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
787 * split and could result in non-contiguous sectors being written in
790 if (nr_sectors
> q
->limits
.chunk_sectors
)
791 return BLK_STS_IOERR
;
793 /* Make sure the BIO is small enough and will not get split */
794 if (nr_sectors
> q
->limits
.max_zone_append_sectors
)
795 return BLK_STS_IOERR
;
797 bio
->bi_opf
|= REQ_NOMERGE
;
802 static noinline_for_stack
bool submit_bio_checks(struct bio
*bio
)
804 struct request_queue
*q
= bio
->bi_disk
->queue
;
805 blk_status_t status
= BLK_STS_IOERR
;
806 struct blk_plug
*plug
;
810 plug
= blk_mq_plug(q
, bio
);
811 if (plug
&& plug
->nowait
)
812 bio
->bi_opf
|= REQ_NOWAIT
;
815 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
816 * if queue does not support NOWAIT.
818 if ((bio
->bi_opf
& REQ_NOWAIT
) && !blk_queue_nowait(q
))
821 if (should_fail_bio(bio
))
824 if (bio
->bi_partno
) {
825 if (unlikely(blk_partition_remap(bio
)))
828 if (unlikely(bio_check_ro(bio
, &bio
->bi_disk
->part0
)))
830 if (unlikely(bio_check_eod(bio
, get_capacity(bio
->bi_disk
))))
835 * Filter flush bio's early so that bio based drivers without flush
836 * support don't have to worry about them.
838 if (op_is_flush(bio
->bi_opf
) &&
839 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
840 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
841 if (!bio_sectors(bio
)) {
847 if (!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
848 bio
->bi_opf
&= ~REQ_HIPRI
;
850 switch (bio_op(bio
)) {
852 if (!blk_queue_discard(q
))
855 case REQ_OP_SECURE_ERASE
:
856 if (!blk_queue_secure_erase(q
))
859 case REQ_OP_WRITE_SAME
:
860 if (!q
->limits
.max_write_same_sectors
)
863 case REQ_OP_ZONE_APPEND
:
864 status
= blk_check_zone_append(q
, bio
);
865 if (status
!= BLK_STS_OK
)
868 case REQ_OP_ZONE_RESET
:
869 case REQ_OP_ZONE_OPEN
:
870 case REQ_OP_ZONE_CLOSE
:
871 case REQ_OP_ZONE_FINISH
:
872 if (!blk_queue_is_zoned(q
))
875 case REQ_OP_ZONE_RESET_ALL
:
876 if (!blk_queue_is_zoned(q
) || !blk_queue_zone_resetall(q
))
879 case REQ_OP_WRITE_ZEROES
:
880 if (!q
->limits
.max_write_zeroes_sectors
)
888 * Various block parts want %current->io_context, so allocate it up
889 * front rather than dealing with lots of pain to allocate it only
890 * where needed. This may fail and the block layer knows how to live
893 if (unlikely(!current
->io_context
))
894 create_task_io_context(current
, GFP_ATOMIC
, q
->node
);
896 if (blk_throtl_bio(bio
)) {
897 blkcg_bio_issue_init(bio
);
901 blk_cgroup_bio_start(bio
);
902 blkcg_bio_issue_init(bio
);
904 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
905 trace_block_bio_queue(q
, bio
);
906 /* Now that enqueuing has been traced, we need to trace
907 * completion as well.
909 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
914 status
= BLK_STS_NOTSUPP
;
916 bio
->bi_status
= status
;
921 static blk_qc_t
__submit_bio(struct bio
*bio
)
923 struct gendisk
*disk
= bio
->bi_disk
;
924 blk_qc_t ret
= BLK_QC_T_NONE
;
926 if (blk_crypto_bio_prep(&bio
)) {
927 if (!disk
->fops
->submit_bio
)
928 return blk_mq_submit_bio(bio
);
929 ret
= disk
->fops
->submit_bio(bio
);
931 blk_queue_exit(disk
->queue
);
936 * The loop in this function may be a bit non-obvious, and so deserves some
939 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
940 * that), so we have a list with a single bio.
941 * - We pretend that we have just taken it off a longer list, so we assign
942 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
943 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
944 * bios through a recursive call to submit_bio_noacct. If it did, we find a
945 * non-NULL value in bio_list and re-enter the loop from the top.
946 * - In this case we really did just take the bio of the top of the list (no
947 * pretending) and so remove it from bio_list, and call into ->submit_bio()
950 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
951 * bio_list_on_stack[1] contains bios that were submitted before the current
952 * ->submit_bio_bio, but that haven't been processed yet.
954 static blk_qc_t
__submit_bio_noacct(struct bio
*bio
)
956 struct bio_list bio_list_on_stack
[2];
957 blk_qc_t ret
= BLK_QC_T_NONE
;
959 BUG_ON(bio
->bi_next
);
961 bio_list_init(&bio_list_on_stack
[0]);
962 current
->bio_list
= bio_list_on_stack
;
965 struct request_queue
*q
= bio
->bi_disk
->queue
;
966 struct bio_list lower
, same
;
968 if (unlikely(bio_queue_enter(bio
) != 0))
972 * Create a fresh bio_list for all subordinate requests.
974 bio_list_on_stack
[1] = bio_list_on_stack
[0];
975 bio_list_init(&bio_list_on_stack
[0]);
977 ret
= __submit_bio(bio
);
980 * Sort new bios into those for a lower level and those for the
983 bio_list_init(&lower
);
984 bio_list_init(&same
);
985 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
986 if (q
== bio
->bi_disk
->queue
)
987 bio_list_add(&same
, bio
);
989 bio_list_add(&lower
, bio
);
992 * Now assemble so we handle the lowest level first.
994 bio_list_merge(&bio_list_on_stack
[0], &lower
);
995 bio_list_merge(&bio_list_on_stack
[0], &same
);
996 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
997 } while ((bio
= bio_list_pop(&bio_list_on_stack
[0])));
999 current
->bio_list
= NULL
;
1003 static blk_qc_t
__submit_bio_noacct_mq(struct bio
*bio
)
1005 struct bio_list bio_list
[2] = { };
1006 blk_qc_t ret
= BLK_QC_T_NONE
;
1008 current
->bio_list
= bio_list
;
1011 struct gendisk
*disk
= bio
->bi_disk
;
1013 if (unlikely(bio_queue_enter(bio
) != 0))
1016 if (!blk_crypto_bio_prep(&bio
)) {
1017 blk_queue_exit(disk
->queue
);
1018 ret
= BLK_QC_T_NONE
;
1022 ret
= blk_mq_submit_bio(bio
);
1023 } while ((bio
= bio_list_pop(&bio_list
[0])));
1025 current
->bio_list
= NULL
;
1030 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1031 * @bio: The bio describing the location in memory and on the device.
1033 * This is a version of submit_bio() that shall only be used for I/O that is
1034 * resubmitted to lower level drivers by stacking block drivers. All file
1035 * systems and other upper level users of the block layer should use
1036 * submit_bio() instead.
1038 blk_qc_t
submit_bio_noacct(struct bio
*bio
)
1040 if (!submit_bio_checks(bio
))
1041 return BLK_QC_T_NONE
;
1044 * We only want one ->submit_bio to be active at a time, else stack
1045 * usage with stacked devices could be a problem. Use current->bio_list
1046 * to collect a list of requests submited by a ->submit_bio method while
1047 * it is active, and then process them after it returned.
1049 if (current
->bio_list
) {
1050 bio_list_add(¤t
->bio_list
[0], bio
);
1051 return BLK_QC_T_NONE
;
1054 if (!bio
->bi_disk
->fops
->submit_bio
)
1055 return __submit_bio_noacct_mq(bio
);
1056 return __submit_bio_noacct(bio
);
1058 EXPORT_SYMBOL(submit_bio_noacct
);
1061 * submit_bio - submit a bio to the block device layer for I/O
1062 * @bio: The &struct bio which describes the I/O
1064 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1065 * fully set up &struct bio that describes the I/O that needs to be done. The
1066 * bio will be send to the device described by the bi_disk and bi_partno fields.
1068 * The success/failure status of the request, along with notification of
1069 * completion, is delivered asynchronously through the ->bi_end_io() callback
1070 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1073 blk_qc_t
submit_bio(struct bio
*bio
)
1075 if (blkcg_punt_bio_submit(bio
))
1076 return BLK_QC_T_NONE
;
1079 * If it's a regular read/write or a barrier with data attached,
1080 * go through the normal accounting stuff before submission.
1082 if (bio_has_data(bio
)) {
1085 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1086 count
= queue_logical_block_size(bio
->bi_disk
->queue
) >> 9;
1088 count
= bio_sectors(bio
);
1090 if (op_is_write(bio_op(bio
))) {
1091 count_vm_events(PGPGOUT
, count
);
1093 task_io_account_read(bio
->bi_iter
.bi_size
);
1094 count_vm_events(PGPGIN
, count
);
1097 if (unlikely(block_dump
)) {
1098 char b
[BDEVNAME_SIZE
];
1099 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1100 current
->comm
, task_pid_nr(current
),
1101 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
1102 (unsigned long long)bio
->bi_iter
.bi_sector
,
1103 bio_devname(bio
, b
), count
);
1108 * If we're reading data that is part of the userspace workingset, count
1109 * submission time as memory stall. When the device is congested, or
1110 * the submitting cgroup IO-throttled, submission can be a significant
1111 * part of overall IO time.
1113 if (unlikely(bio_op(bio
) == REQ_OP_READ
&&
1114 bio_flagged(bio
, BIO_WORKINGSET
))) {
1115 unsigned long pflags
;
1118 psi_memstall_enter(&pflags
);
1119 ret
= submit_bio_noacct(bio
);
1120 psi_memstall_leave(&pflags
);
1125 return submit_bio_noacct(bio
);
1127 EXPORT_SYMBOL(submit_bio
);
1130 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1131 * for the new queue limits
1133 * @rq: the request being checked
1136 * @rq may have been made based on weaker limitations of upper-level queues
1137 * in request stacking drivers, and it may violate the limitation of @q.
1138 * Since the block layer and the underlying device driver trust @rq
1139 * after it is inserted to @q, it should be checked against @q before
1140 * the insertion using this generic function.
1142 * Request stacking drivers like request-based dm may change the queue
1143 * limits when retrying requests on other queues. Those requests need
1144 * to be checked against the new queue limits again during dispatch.
1146 static blk_status_t
blk_cloned_rq_check_limits(struct request_queue
*q
,
1149 unsigned int max_sectors
= blk_queue_get_max_sectors(q
, req_op(rq
));
1151 if (blk_rq_sectors(rq
) > max_sectors
) {
1153 * SCSI device does not have a good way to return if
1154 * Write Same/Zero is actually supported. If a device rejects
1155 * a non-read/write command (discard, write same,etc.) the
1156 * low-level device driver will set the relevant queue limit to
1157 * 0 to prevent blk-lib from issuing more of the offending
1158 * operations. Commands queued prior to the queue limit being
1159 * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O
1160 * errors being propagated to upper layers.
1162 if (max_sectors
== 0)
1163 return BLK_STS_NOTSUPP
;
1165 printk(KERN_ERR
"%s: over max size limit. (%u > %u)\n",
1166 __func__
, blk_rq_sectors(rq
), max_sectors
);
1167 return BLK_STS_IOERR
;
1171 * queue's settings related to segment counting like q->bounce_pfn
1172 * may differ from that of other stacking queues.
1173 * Recalculate it to check the request correctly on this queue's
1176 rq
->nr_phys_segments
= blk_recalc_rq_segments(rq
);
1177 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1178 printk(KERN_ERR
"%s: over max segments limit. (%hu > %hu)\n",
1179 __func__
, rq
->nr_phys_segments
, queue_max_segments(q
));
1180 return BLK_STS_IOERR
;
1187 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1188 * @q: the queue to submit the request
1189 * @rq: the request being queued
1191 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1195 ret
= blk_cloned_rq_check_limits(q
, rq
);
1196 if (ret
!= BLK_STS_OK
)
1200 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1201 return BLK_STS_IOERR
;
1203 if (blk_crypto_insert_cloned_request(rq
))
1204 return BLK_STS_IOERR
;
1206 if (blk_queue_io_stat(q
))
1207 blk_account_io_start(rq
);
1210 * Since we have a scheduler attached on the top device,
1211 * bypass a potential scheduler on the bottom device for
1214 return blk_mq_request_issue_directly(rq
, true);
1216 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1219 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1220 * @rq: request to examine
1223 * A request could be merge of IOs which require different failure
1224 * handling. This function determines the number of bytes which
1225 * can be failed from the beginning of the request without
1226 * crossing into area which need to be retried further.
1229 * The number of bytes to fail.
1231 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1233 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1234 unsigned int bytes
= 0;
1237 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
1238 return blk_rq_bytes(rq
);
1241 * Currently the only 'mixing' which can happen is between
1242 * different fastfail types. We can safely fail portions
1243 * which have all the failfast bits that the first one has -
1244 * the ones which are at least as eager to fail as the first
1247 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1248 if ((bio
->bi_opf
& ff
) != ff
)
1250 bytes
+= bio
->bi_iter
.bi_size
;
1253 /* this could lead to infinite loop */
1254 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1257 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1259 static void update_io_ticks(struct hd_struct
*part
, unsigned long now
, bool end
)
1261 unsigned long stamp
;
1263 stamp
= READ_ONCE(part
->stamp
);
1264 if (unlikely(stamp
!= now
)) {
1265 if (likely(cmpxchg(&part
->stamp
, stamp
, now
) == stamp
))
1266 __part_stat_add(part
, io_ticks
, end
? now
- stamp
: 1);
1269 part
= &part_to_disk(part
)->part0
;
1274 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1276 if (req
->part
&& blk_do_io_stat(req
)) {
1277 const int sgrp
= op_stat_group(req_op(req
));
1278 struct hd_struct
*part
;
1282 part_stat_add(part
, sectors
[sgrp
], bytes
>> 9);
1287 void blk_account_io_done(struct request
*req
, u64 now
)
1290 * Account IO completion. flush_rq isn't accounted as a
1291 * normal IO on queueing nor completion. Accounting the
1292 * containing request is enough.
1294 if (req
->part
&& blk_do_io_stat(req
) &&
1295 !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
1296 const int sgrp
= op_stat_group(req_op(req
));
1297 struct hd_struct
*part
;
1302 update_io_ticks(part
, jiffies
, true);
1303 part_stat_inc(part
, ios
[sgrp
]);
1304 part_stat_add(part
, nsecs
[sgrp
], now
- req
->start_time_ns
);
1307 hd_struct_put(part
);
1311 void blk_account_io_start(struct request
*rq
)
1313 if (!blk_do_io_stat(rq
))
1316 rq
->part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
1319 update_io_ticks(rq
->part
, jiffies
, false);
1323 static unsigned long __part_start_io_acct(struct hd_struct
*part
,
1324 unsigned int sectors
, unsigned int op
)
1326 const int sgrp
= op_stat_group(op
);
1327 unsigned long now
= READ_ONCE(jiffies
);
1330 update_io_ticks(part
, now
, false);
1331 part_stat_inc(part
, ios
[sgrp
]);
1332 part_stat_add(part
, sectors
[sgrp
], sectors
);
1333 part_stat_local_inc(part
, in_flight
[op_is_write(op
)]);
1339 unsigned long part_start_io_acct(struct gendisk
*disk
, struct hd_struct
**part
,
1342 *part
= disk_map_sector_rcu(disk
, bio
->bi_iter
.bi_sector
);
1344 return __part_start_io_acct(*part
, bio_sectors(bio
), bio_op(bio
));
1346 EXPORT_SYMBOL_GPL(part_start_io_acct
);
1348 unsigned long disk_start_io_acct(struct gendisk
*disk
, unsigned int sectors
,
1351 return __part_start_io_acct(&disk
->part0
, sectors
, op
);
1353 EXPORT_SYMBOL(disk_start_io_acct
);
1355 static void __part_end_io_acct(struct hd_struct
*part
, unsigned int op
,
1356 unsigned long start_time
)
1358 const int sgrp
= op_stat_group(op
);
1359 unsigned long now
= READ_ONCE(jiffies
);
1360 unsigned long duration
= now
- start_time
;
1363 update_io_ticks(part
, now
, true);
1364 part_stat_add(part
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1365 part_stat_local_dec(part
, in_flight
[op_is_write(op
)]);
1369 void part_end_io_acct(struct hd_struct
*part
, struct bio
*bio
,
1370 unsigned long start_time
)
1372 __part_end_io_acct(part
, bio_op(bio
), start_time
);
1373 hd_struct_put(part
);
1375 EXPORT_SYMBOL_GPL(part_end_io_acct
);
1377 void disk_end_io_acct(struct gendisk
*disk
, unsigned int op
,
1378 unsigned long start_time
)
1380 __part_end_io_acct(&disk
->part0
, op
, start_time
);
1382 EXPORT_SYMBOL(disk_end_io_acct
);
1385 * Steal bios from a request and add them to a bio list.
1386 * The request must not have been partially completed before.
1388 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
1392 list
->tail
->bi_next
= rq
->bio
;
1394 list
->head
= rq
->bio
;
1395 list
->tail
= rq
->biotail
;
1403 EXPORT_SYMBOL_GPL(blk_steal_bios
);
1406 * blk_update_request - Special helper function for request stacking drivers
1407 * @req: the request being processed
1408 * @error: block status code
1409 * @nr_bytes: number of bytes to complete @req
1412 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1413 * the request structure even if @req doesn't have leftover.
1414 * If @req has leftover, sets it up for the next range of segments.
1416 * This special helper function is only for request stacking drivers
1417 * (e.g. request-based dm) so that they can handle partial completion.
1418 * Actual device drivers should use blk_mq_end_request instead.
1420 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1421 * %false return from this function.
1424 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1425 * blk_rq_bytes() and in blk_update_request().
1428 * %false - this request doesn't have any more data
1429 * %true - this request has more data
1431 bool blk_update_request(struct request
*req
, blk_status_t error
,
1432 unsigned int nr_bytes
)
1436 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
1441 #ifdef CONFIG_BLK_DEV_INTEGRITY
1442 if (blk_integrity_rq(req
) && req_op(req
) == REQ_OP_READ
&&
1443 error
== BLK_STS_OK
)
1444 req
->q
->integrity
.profile
->complete_fn(req
, nr_bytes
);
1447 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
1448 !(req
->rq_flags
& RQF_QUIET
)))
1449 print_req_error(req
, error
, __func__
);
1451 blk_account_io_completion(req
, nr_bytes
);
1455 struct bio
*bio
= req
->bio
;
1456 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
1458 if (bio_bytes
== bio
->bi_iter
.bi_size
)
1459 req
->bio
= bio
->bi_next
;
1461 /* Completion has already been traced */
1462 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1463 req_bio_endio(req
, bio
, bio_bytes
, error
);
1465 total_bytes
+= bio_bytes
;
1466 nr_bytes
-= bio_bytes
;
1477 * Reset counters so that the request stacking driver
1478 * can find how many bytes remain in the request
1481 req
->__data_len
= 0;
1485 req
->__data_len
-= total_bytes
;
1487 /* update sector only for requests with clear definition of sector */
1488 if (!blk_rq_is_passthrough(req
))
1489 req
->__sector
+= total_bytes
>> 9;
1491 /* mixed attributes always follow the first bio */
1492 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
1493 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
1494 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
1497 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
1499 * If total number of sectors is less than the first segment
1500 * size, something has gone terribly wrong.
1502 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
1503 blk_dump_rq_flags(req
, "request botched");
1504 req
->__data_len
= blk_rq_cur_bytes(req
);
1507 /* recalculate the number of segments */
1508 req
->nr_phys_segments
= blk_recalc_rq_segments(req
);
1513 EXPORT_SYMBOL_GPL(blk_update_request
);
1515 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1517 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1518 * @rq: the request to be flushed
1521 * Flush all pages in @rq.
1523 void rq_flush_dcache_pages(struct request
*rq
)
1525 struct req_iterator iter
;
1526 struct bio_vec bvec
;
1528 rq_for_each_segment(bvec
, rq
, iter
)
1529 flush_dcache_page(bvec
.bv_page
);
1531 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
1535 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1536 * @q : the queue of the device being checked
1539 * Check if underlying low-level drivers of a device are busy.
1540 * If the drivers want to export their busy state, they must set own
1541 * exporting function using blk_queue_lld_busy() first.
1543 * Basically, this function is used only by request stacking drivers
1544 * to stop dispatching requests to underlying devices when underlying
1545 * devices are busy. This behavior helps more I/O merging on the queue
1546 * of the request stacking driver and prevents I/O throughput regression
1547 * on burst I/O load.
1550 * 0 - Not busy (The request stacking driver should dispatch request)
1551 * 1 - Busy (The request stacking driver should stop dispatching request)
1553 int blk_lld_busy(struct request_queue
*q
)
1555 if (queue_is_mq(q
) && q
->mq_ops
->busy
)
1556 return q
->mq_ops
->busy(q
);
1560 EXPORT_SYMBOL_GPL(blk_lld_busy
);
1563 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1564 * @rq: the clone request to be cleaned up
1567 * Free all bios in @rq for a cloned request.
1569 void blk_rq_unprep_clone(struct request
*rq
)
1573 while ((bio
= rq
->bio
) != NULL
) {
1574 rq
->bio
= bio
->bi_next
;
1579 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
1582 * blk_rq_prep_clone - Helper function to setup clone request
1583 * @rq: the request to be setup
1584 * @rq_src: original request to be cloned
1585 * @bs: bio_set that bios for clone are allocated from
1586 * @gfp_mask: memory allocation mask for bio
1587 * @bio_ctr: setup function to be called for each clone bio.
1588 * Returns %0 for success, non %0 for failure.
1589 * @data: private data to be passed to @bio_ctr
1592 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1593 * Also, pages which the original bios are pointing to are not copied
1594 * and the cloned bios just point same pages.
1595 * So cloned bios must be completed before original bios, which means
1596 * the caller must complete @rq before @rq_src.
1598 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
1599 struct bio_set
*bs
, gfp_t gfp_mask
,
1600 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
1603 struct bio
*bio
, *bio_src
;
1608 __rq_for_each_bio(bio_src
, rq_src
) {
1609 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
1613 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
1617 rq
->biotail
->bi_next
= bio
;
1620 rq
->bio
= rq
->biotail
= bio
;
1625 /* Copy attributes of the original request to the clone request. */
1626 rq
->__sector
= blk_rq_pos(rq_src
);
1627 rq
->__data_len
= blk_rq_bytes(rq_src
);
1628 if (rq_src
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
1629 rq
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
1630 rq
->special_vec
= rq_src
->special_vec
;
1632 rq
->nr_phys_segments
= rq_src
->nr_phys_segments
;
1633 rq
->ioprio
= rq_src
->ioprio
;
1635 if (rq
->bio
&& blk_crypto_rq_bio_prep(rq
, rq
->bio
, gfp_mask
) < 0)
1643 blk_rq_unprep_clone(rq
);
1647 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
1649 int kblockd_schedule_work(struct work_struct
*work
)
1651 return queue_work(kblockd_workqueue
, work
);
1653 EXPORT_SYMBOL(kblockd_schedule_work
);
1655 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
1656 unsigned long delay
)
1658 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
1660 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
1663 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1664 * @plug: The &struct blk_plug that needs to be initialized
1667 * blk_start_plug() indicates to the block layer an intent by the caller
1668 * to submit multiple I/O requests in a batch. The block layer may use
1669 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1670 * is called. However, the block layer may choose to submit requests
1671 * before a call to blk_finish_plug() if the number of queued I/Os
1672 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1673 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1674 * the task schedules (see below).
1676 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1677 * pending I/O should the task end up blocking between blk_start_plug() and
1678 * blk_finish_plug(). This is important from a performance perspective, but
1679 * also ensures that we don't deadlock. For instance, if the task is blocking
1680 * for a memory allocation, memory reclaim could end up wanting to free a
1681 * page belonging to that request that is currently residing in our private
1682 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1683 * this kind of deadlock.
1685 void blk_start_plug(struct blk_plug
*plug
)
1687 struct task_struct
*tsk
= current
;
1690 * If this is a nested plug, don't actually assign it.
1695 INIT_LIST_HEAD(&plug
->mq_list
);
1696 INIT_LIST_HEAD(&plug
->cb_list
);
1698 plug
->multiple_queues
= false;
1699 plug
->nowait
= false;
1702 * Store ordering should not be needed here, since a potential
1703 * preempt will imply a full memory barrier
1707 EXPORT_SYMBOL(blk_start_plug
);
1709 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
1711 LIST_HEAD(callbacks
);
1713 while (!list_empty(&plug
->cb_list
)) {
1714 list_splice_init(&plug
->cb_list
, &callbacks
);
1716 while (!list_empty(&callbacks
)) {
1717 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
1720 list_del(&cb
->list
);
1721 cb
->callback(cb
, from_schedule
);
1726 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
1729 struct blk_plug
*plug
= current
->plug
;
1730 struct blk_plug_cb
*cb
;
1735 list_for_each_entry(cb
, &plug
->cb_list
, list
)
1736 if (cb
->callback
== unplug
&& cb
->data
== data
)
1739 /* Not currently on the callback list */
1740 BUG_ON(size
< sizeof(*cb
));
1741 cb
= kzalloc(size
, GFP_ATOMIC
);
1744 cb
->callback
= unplug
;
1745 list_add(&cb
->list
, &plug
->cb_list
);
1749 EXPORT_SYMBOL(blk_check_plugged
);
1751 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
1753 flush_plug_callbacks(plug
, from_schedule
);
1755 if (!list_empty(&plug
->mq_list
))
1756 blk_mq_flush_plug_list(plug
, from_schedule
);
1760 * blk_finish_plug - mark the end of a batch of submitted I/O
1761 * @plug: The &struct blk_plug passed to blk_start_plug()
1764 * Indicate that a batch of I/O submissions is complete. This function
1765 * must be paired with an initial call to blk_start_plug(). The intent
1766 * is to allow the block layer to optimize I/O submission. See the
1767 * documentation for blk_start_plug() for more information.
1769 void blk_finish_plug(struct blk_plug
*plug
)
1771 if (plug
!= current
->plug
)
1773 blk_flush_plug_list(plug
, false);
1775 current
->plug
= NULL
;
1777 EXPORT_SYMBOL(blk_finish_plug
);
1779 void blk_io_schedule(void)
1781 /* Prevent hang_check timer from firing at us during very long I/O */
1782 unsigned long timeout
= sysctl_hung_task_timeout_secs
* HZ
/ 2;
1785 io_schedule_timeout(timeout
);
1789 EXPORT_SYMBOL_GPL(blk_io_schedule
);
1791 int __init
blk_dev_init(void)
1793 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
1794 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1795 sizeof_field(struct request
, cmd_flags
));
1796 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1797 sizeof_field(struct bio
, bi_opf
));
1799 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1800 kblockd_workqueue
= alloc_workqueue("kblockd",
1801 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
1802 if (!kblockd_workqueue
)
1803 panic("Failed to create kblockd\n");
1805 blk_requestq_cachep
= kmem_cache_create("request_queue",
1806 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
1808 blk_debugfs_root
= debugfs_create_dir("block", NULL
);