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/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/completion.h>
27 #include <linux/slab.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/task_io_accounting_ops.h>
31 #include <linux/fault-inject.h>
32 #include <linux/list_sort.h>
33 #include <linux/delay.h>
34 #include <linux/ratelimit.h>
35 #include <linux/pm_runtime.h>
36 #include <linux/blk-cgroup.h>
37 #include <linux/debugfs.h>
38 #include <linux/bpf.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/block.h>
45 #include "blk-mq-sched.h"
47 #include "blk-rq-qos.h"
49 #ifdef CONFIG_DEBUG_FS
50 struct dentry
*blk_debugfs_root
;
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
59 DEFINE_IDA(blk_queue_ida
);
62 * For queue allocation
64 struct kmem_cache
*blk_requestq_cachep
;
67 * Controlling structure to kblockd
69 static struct workqueue_struct
*kblockd_workqueue
;
72 * blk_queue_flag_set - atomically set a queue flag
73 * @flag: flag to be set
76 void blk_queue_flag_set(unsigned int flag
, struct request_queue
*q
)
78 set_bit(flag
, &q
->queue_flags
);
80 EXPORT_SYMBOL(blk_queue_flag_set
);
83 * blk_queue_flag_clear - atomically clear a queue flag
84 * @flag: flag to be cleared
87 void blk_queue_flag_clear(unsigned int flag
, struct request_queue
*q
)
89 clear_bit(flag
, &q
->queue_flags
);
91 EXPORT_SYMBOL(blk_queue_flag_clear
);
94 * blk_queue_flag_test_and_set - atomically test and set a queue flag
95 * @flag: flag to be set
98 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
99 * the flag was already set.
101 bool blk_queue_flag_test_and_set(unsigned int flag
, struct request_queue
*q
)
103 return test_and_set_bit(flag
, &q
->queue_flags
);
105 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set
);
107 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
109 memset(rq
, 0, sizeof(*rq
));
111 INIT_LIST_HEAD(&rq
->queuelist
);
113 rq
->__sector
= (sector_t
) -1;
114 INIT_HLIST_NODE(&rq
->hash
);
115 RB_CLEAR_NODE(&rq
->rb_node
);
117 rq
->internal_tag
= -1;
118 rq
->start_time_ns
= ktime_get_ns();
121 EXPORT_SYMBOL(blk_rq_init
);
123 static const struct {
127 [BLK_STS_OK
] = { 0, "" },
128 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
129 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
130 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
131 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
132 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
133 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
134 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
135 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
136 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
137 [BLK_STS_DEV_RESOURCE
] = { -EBUSY
, "device resource" },
138 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
140 /* device mapper special case, should not leak out: */
141 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
143 /* everything else not covered above: */
144 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
147 blk_status_t
errno_to_blk_status(int errno
)
151 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
152 if (blk_errors
[i
].errno
== errno
)
153 return (__force blk_status_t
)i
;
156 return BLK_STS_IOERR
;
158 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
160 int blk_status_to_errno(blk_status_t status
)
162 int idx
= (__force
int)status
;
164 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
166 return blk_errors
[idx
].errno
;
168 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
170 static void print_req_error(struct request
*req
, blk_status_t status
)
172 int idx
= (__force
int)status
;
174 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
177 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu flags %x\n",
178 __func__
, blk_errors
[idx
].name
,
179 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
180 (unsigned long long)blk_rq_pos(req
),
184 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
185 unsigned int nbytes
, blk_status_t error
)
188 bio
->bi_status
= error
;
190 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
191 bio_set_flag(bio
, BIO_QUIET
);
193 bio_advance(bio
, nbytes
);
195 /* don't actually finish bio if it's part of flush sequence */
196 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
200 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
202 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
203 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
204 (unsigned long long) rq
->cmd_flags
);
206 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
207 (unsigned long long)blk_rq_pos(rq
),
208 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
209 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
210 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
212 EXPORT_SYMBOL(blk_dump_rq_flags
);
215 * blk_sync_queue - cancel any pending callbacks on a queue
219 * The block layer may perform asynchronous callback activity
220 * on a queue, such as calling the unplug function after a timeout.
221 * A block device may call blk_sync_queue to ensure that any
222 * such activity is cancelled, thus allowing it to release resources
223 * that the callbacks might use. The caller must already have made sure
224 * that its ->make_request_fn will not re-add plugging prior to calling
227 * This function does not cancel any asynchronous activity arising
228 * out of elevator or throttling code. That would require elevator_exit()
229 * and blkcg_exit_queue() to be called with queue lock initialized.
232 void blk_sync_queue(struct request_queue
*q
)
234 del_timer_sync(&q
->timeout
);
235 cancel_work_sync(&q
->timeout_work
);
237 EXPORT_SYMBOL(blk_sync_queue
);
240 * blk_set_pm_only - increment pm_only counter
241 * @q: request queue pointer
243 void blk_set_pm_only(struct request_queue
*q
)
245 atomic_inc(&q
->pm_only
);
247 EXPORT_SYMBOL_GPL(blk_set_pm_only
);
249 void blk_clear_pm_only(struct request_queue
*q
)
253 pm_only
= atomic_dec_return(&q
->pm_only
);
254 WARN_ON_ONCE(pm_only
< 0);
256 wake_up_all(&q
->mq_freeze_wq
);
258 EXPORT_SYMBOL_GPL(blk_clear_pm_only
);
260 void blk_put_queue(struct request_queue
*q
)
262 kobject_put(&q
->kobj
);
264 EXPORT_SYMBOL(blk_put_queue
);
266 void blk_set_queue_dying(struct request_queue
*q
)
268 blk_queue_flag_set(QUEUE_FLAG_DYING
, q
);
271 * When queue DYING flag is set, we need to block new req
272 * entering queue, so we call blk_freeze_queue_start() to
273 * prevent I/O from crossing blk_queue_enter().
275 blk_freeze_queue_start(q
);
278 blk_mq_wake_waiters(q
);
280 /* Make blk_queue_enter() reexamine the DYING flag. */
281 wake_up_all(&q
->mq_freeze_wq
);
283 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
285 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
286 void blk_exit_queue(struct request_queue
*q
)
289 * Since the I/O scheduler exit code may access cgroup information,
290 * perform I/O scheduler exit before disassociating from the block
295 elevator_exit(q
, q
->elevator
);
300 * Remove all references to @q from the block cgroup controller before
301 * restoring @q->queue_lock to avoid that restoring this pointer causes
302 * e.g. blkcg_print_blkgs() to crash.
307 * Since the cgroup code may dereference the @q->backing_dev_info
308 * pointer, only decrease its reference count after having removed the
309 * association with the block cgroup controller.
311 bdi_put(q
->backing_dev_info
);
315 * blk_cleanup_queue - shutdown a request queue
316 * @q: request queue to shutdown
318 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
319 * put it. All future requests will be failed immediately with -ENODEV.
321 void blk_cleanup_queue(struct request_queue
*q
)
323 /* mark @q DYING, no new request or merges will be allowed afterwards */
324 mutex_lock(&q
->sysfs_lock
);
325 blk_set_queue_dying(q
);
327 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
328 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
329 blk_queue_flag_set(QUEUE_FLAG_DYING
, q
);
330 mutex_unlock(&q
->sysfs_lock
);
333 * Drain all requests queued before DYING marking. Set DEAD flag to
334 * prevent that q->request_fn() gets invoked after draining finished.
340 blk_queue_flag_set(QUEUE_FLAG_DEAD
, q
);
342 /* for synchronous bio-based driver finish in-flight integrity i/o */
343 blk_flush_integrity();
345 /* @q won't process any more request, flush async actions */
346 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
350 * I/O scheduler exit is only safe after the sysfs scheduler attribute
353 WARN_ON_ONCE(q
->kobj
.state_in_sysfs
);
358 blk_mq_exit_queue(q
);
360 percpu_ref_exit(&q
->q_usage_counter
);
362 /* @q is and will stay empty, shutdown and put */
365 EXPORT_SYMBOL(blk_cleanup_queue
);
367 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
369 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
371 EXPORT_SYMBOL(blk_alloc_queue
);
374 * blk_queue_enter() - try to increase q->q_usage_counter
375 * @q: request queue pointer
376 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
378 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
380 const bool pm
= flags
& BLK_MQ_REQ_PREEMPT
;
383 bool success
= false;
386 if (percpu_ref_tryget_live(&q
->q_usage_counter
)) {
388 * The code that increments the pm_only counter is
389 * responsible for ensuring that that counter is
390 * globally visible before the queue is unfrozen.
392 if (pm
|| !blk_queue_pm_only(q
)) {
395 percpu_ref_put(&q
->q_usage_counter
);
403 if (flags
& BLK_MQ_REQ_NOWAIT
)
407 * read pair of barrier in blk_freeze_queue_start(),
408 * we need to order reading __PERCPU_REF_DEAD flag of
409 * .q_usage_counter and reading .mq_freeze_depth or
410 * queue dying flag, otherwise the following wait may
411 * never return if the two reads are reordered.
415 wait_event(q
->mq_freeze_wq
,
416 (!q
->mq_freeze_depth
&&
417 (pm
|| (blk_pm_request_resume(q
),
418 !blk_queue_pm_only(q
)))) ||
420 if (blk_queue_dying(q
))
425 void blk_queue_exit(struct request_queue
*q
)
427 percpu_ref_put(&q
->q_usage_counter
);
430 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
432 struct request_queue
*q
=
433 container_of(ref
, struct request_queue
, q_usage_counter
);
435 wake_up_all(&q
->mq_freeze_wq
);
438 static void blk_rq_timed_out_timer(struct timer_list
*t
)
440 struct request_queue
*q
= from_timer(q
, t
, timeout
);
442 kblockd_schedule_work(&q
->timeout_work
);
445 static void blk_timeout_work(struct work_struct
*work
)
450 * blk_alloc_queue_node - allocate a request queue
451 * @gfp_mask: memory allocation flags
452 * @node_id: NUMA node to allocate memory from
454 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
456 struct request_queue
*q
;
459 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
460 gfp_mask
| __GFP_ZERO
, node_id
);
464 INIT_LIST_HEAD(&q
->queue_head
);
465 q
->last_merge
= NULL
;
467 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
471 ret
= bioset_init(&q
->bio_split
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
475 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
476 if (!q
->backing_dev_info
)
479 q
->stats
= blk_alloc_queue_stats();
483 q
->backing_dev_info
->ra_pages
= VM_READAHEAD_PAGES
;
484 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
485 q
->backing_dev_info
->name
= "block";
488 timer_setup(&q
->backing_dev_info
->laptop_mode_wb_timer
,
489 laptop_mode_timer_fn
, 0);
490 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
491 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
492 INIT_LIST_HEAD(&q
->icq_list
);
493 #ifdef CONFIG_BLK_CGROUP
494 INIT_LIST_HEAD(&q
->blkg_list
);
497 kobject_init(&q
->kobj
, &blk_queue_ktype
);
499 #ifdef CONFIG_BLK_DEV_IO_TRACE
500 mutex_init(&q
->blk_trace_mutex
);
502 mutex_init(&q
->sysfs_lock
);
503 spin_lock_init(&q
->queue_lock
);
505 init_waitqueue_head(&q
->mq_freeze_wq
);
506 mutex_init(&q
->mq_freeze_lock
);
509 * Init percpu_ref in atomic mode so that it's faster to shutdown.
510 * See blk_register_queue() for details.
512 if (percpu_ref_init(&q
->q_usage_counter
,
513 blk_queue_usage_counter_release
,
514 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
517 if (blkcg_init_queue(q
))
523 percpu_ref_exit(&q
->q_usage_counter
);
525 blk_free_queue_stats(q
->stats
);
527 bdi_put(q
->backing_dev_info
);
529 bioset_exit(&q
->bio_split
);
531 ida_simple_remove(&blk_queue_ida
, q
->id
);
533 kmem_cache_free(blk_requestq_cachep
, q
);
536 EXPORT_SYMBOL(blk_alloc_queue_node
);
538 bool blk_get_queue(struct request_queue
*q
)
540 if (likely(!blk_queue_dying(q
))) {
547 EXPORT_SYMBOL(blk_get_queue
);
550 * blk_get_request - allocate a request
551 * @q: request queue to allocate a request for
552 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
553 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
555 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
556 blk_mq_req_flags_t flags
)
560 WARN_ON_ONCE(op
& REQ_NOWAIT
);
561 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
563 req
= blk_mq_alloc_request(q
, op
, flags
);
564 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
565 q
->mq_ops
->initialize_rq_fn(req
);
569 EXPORT_SYMBOL(blk_get_request
);
571 void blk_put_request(struct request
*req
)
573 blk_mq_free_request(req
);
575 EXPORT_SYMBOL(blk_put_request
);
577 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
580 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
582 if (!ll_back_merge_fn(q
, req
, bio
))
585 trace_block_bio_backmerge(q
, req
, bio
);
587 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
588 blk_rq_set_mixed_merge(req
);
590 req
->biotail
->bi_next
= bio
;
592 req
->__data_len
+= bio
->bi_iter
.bi_size
;
594 blk_account_io_start(req
, false);
598 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
601 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
603 if (!ll_front_merge_fn(q
, req
, bio
))
606 trace_block_bio_frontmerge(q
, req
, bio
);
608 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
609 blk_rq_set_mixed_merge(req
);
611 bio
->bi_next
= req
->bio
;
614 req
->__sector
= bio
->bi_iter
.bi_sector
;
615 req
->__data_len
+= bio
->bi_iter
.bi_size
;
617 blk_account_io_start(req
, false);
621 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
624 unsigned short segments
= blk_rq_nr_discard_segments(req
);
626 if (segments
>= queue_max_discard_segments(q
))
628 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
629 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
632 req
->biotail
->bi_next
= bio
;
634 req
->__data_len
+= bio
->bi_iter
.bi_size
;
635 req
->nr_phys_segments
= segments
+ 1;
637 blk_account_io_start(req
, false);
640 req_set_nomerge(q
, req
);
645 * blk_attempt_plug_merge - try to merge with %current's plugged list
646 * @q: request_queue new bio is being queued at
647 * @bio: new bio being queued
648 * @same_queue_rq: pointer to &struct request that gets filled in when
649 * another request associated with @q is found on the plug list
650 * (optional, may be %NULL)
652 * Determine whether @bio being queued on @q can be merged with a request
653 * on %current's plugged list. Returns %true if merge was successful,
656 * Plugging coalesces IOs from the same issuer for the same purpose without
657 * going through @q->queue_lock. As such it's more of an issuing mechanism
658 * than scheduling, and the request, while may have elvpriv data, is not
659 * added on the elevator at this point. In addition, we don't have
660 * reliable access to the elevator outside queue lock. Only check basic
661 * merging parameters without querying the elevator.
663 * Caller must ensure !blk_queue_nomerges(q) beforehand.
665 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
666 struct request
**same_queue_rq
)
668 struct blk_plug
*plug
;
670 struct list_head
*plug_list
;
672 plug
= current
->plug
;
676 plug_list
= &plug
->mq_list
;
678 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
681 if (rq
->q
== q
&& same_queue_rq
) {
683 * Only blk-mq multiple hardware queues case checks the
684 * rq in the same queue, there should be only one such
690 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
693 switch (blk_try_merge(rq
, bio
)) {
694 case ELEVATOR_BACK_MERGE
:
695 merged
= bio_attempt_back_merge(q
, rq
, bio
);
697 case ELEVATOR_FRONT_MERGE
:
698 merged
= bio_attempt_front_merge(q
, rq
, bio
);
700 case ELEVATOR_DISCARD_MERGE
:
701 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
714 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
716 if (bio
->bi_opf
& REQ_RAHEAD
)
717 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
719 req
->__sector
= bio
->bi_iter
.bi_sector
;
720 req
->ioprio
= bio_prio(bio
);
721 req
->write_hint
= bio
->bi_write_hint
;
722 blk_rq_bio_prep(req
->q
, req
, bio
);
724 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
726 static void handle_bad_sector(struct bio
*bio
, sector_t maxsector
)
728 char b
[BDEVNAME_SIZE
];
730 printk(KERN_INFO
"attempt to access beyond end of device\n");
731 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
732 bio_devname(bio
, b
), bio
->bi_opf
,
733 (unsigned long long)bio_end_sector(bio
),
734 (long long)maxsector
);
737 #ifdef CONFIG_FAIL_MAKE_REQUEST
739 static DECLARE_FAULT_ATTR(fail_make_request
);
741 static int __init
setup_fail_make_request(char *str
)
743 return setup_fault_attr(&fail_make_request
, str
);
745 __setup("fail_make_request=", setup_fail_make_request
);
747 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
749 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
752 static int __init
fail_make_request_debugfs(void)
754 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
755 NULL
, &fail_make_request
);
757 return PTR_ERR_OR_ZERO(dir
);
760 late_initcall(fail_make_request_debugfs
);
762 #else /* CONFIG_FAIL_MAKE_REQUEST */
764 static inline bool should_fail_request(struct hd_struct
*part
,
770 #endif /* CONFIG_FAIL_MAKE_REQUEST */
772 static inline bool bio_check_ro(struct bio
*bio
, struct hd_struct
*part
)
774 const int op
= bio_op(bio
);
776 if (part
->policy
&& op_is_write(op
)) {
777 char b
[BDEVNAME_SIZE
];
779 if (op_is_flush(bio
->bi_opf
) && !bio_sectors(bio
))
783 "generic_make_request: Trying to write "
784 "to read-only block-device %s (partno %d)\n",
785 bio_devname(bio
, b
), part
->partno
);
786 /* Older lvm-tools actually trigger this */
793 static noinline
int should_fail_bio(struct bio
*bio
)
795 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
799 ALLOW_ERROR_INJECTION(should_fail_bio
, ERRNO
);
802 * Check whether this bio extends beyond the end of the device or partition.
803 * This may well happen - the kernel calls bread() without checking the size of
804 * the device, e.g., when mounting a file system.
806 static inline int bio_check_eod(struct bio
*bio
, sector_t maxsector
)
808 unsigned int nr_sectors
= bio_sectors(bio
);
810 if (nr_sectors
&& maxsector
&&
811 (nr_sectors
> maxsector
||
812 bio
->bi_iter
.bi_sector
> maxsector
- nr_sectors
)) {
813 handle_bad_sector(bio
, maxsector
);
820 * Remap block n of partition p to block n+start(p) of the disk.
822 static inline int blk_partition_remap(struct bio
*bio
)
828 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
831 if (unlikely(should_fail_request(p
, bio
->bi_iter
.bi_size
)))
833 if (unlikely(bio_check_ro(bio
, p
)))
837 * Zone reset does not include bi_size so bio_sectors() is always 0.
838 * Include a test for the reset op code and perform the remap if needed.
840 if (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
) {
841 if (bio_check_eod(bio
, part_nr_sects_read(p
)))
843 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
844 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
845 bio
->bi_iter
.bi_sector
- p
->start_sect
);
854 static noinline_for_stack
bool
855 generic_make_request_checks(struct bio
*bio
)
857 struct request_queue
*q
;
858 int nr_sectors
= bio_sectors(bio
);
859 blk_status_t status
= BLK_STS_IOERR
;
860 char b
[BDEVNAME_SIZE
];
864 q
= bio
->bi_disk
->queue
;
867 "generic_make_request: Trying to access "
868 "nonexistent block-device %s (%Lu)\n",
869 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
874 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
875 * if queue is not a request based queue.
877 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_mq(q
))
880 if (should_fail_bio(bio
))
883 if (bio
->bi_partno
) {
884 if (unlikely(blk_partition_remap(bio
)))
887 if (unlikely(bio_check_ro(bio
, &bio
->bi_disk
->part0
)))
889 if (unlikely(bio_check_eod(bio
, get_capacity(bio
->bi_disk
))))
894 * Filter flush bio's early so that make_request based
895 * drivers without flush support don't have to worry
898 if (op_is_flush(bio
->bi_opf
) &&
899 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
900 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
907 if (!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
908 bio
->bi_opf
&= ~REQ_HIPRI
;
910 switch (bio_op(bio
)) {
912 if (!blk_queue_discard(q
))
915 case REQ_OP_SECURE_ERASE
:
916 if (!blk_queue_secure_erase(q
))
919 case REQ_OP_WRITE_SAME
:
920 if (!q
->limits
.max_write_same_sectors
)
923 case REQ_OP_ZONE_RESET
:
924 if (!blk_queue_is_zoned(q
))
927 case REQ_OP_WRITE_ZEROES
:
928 if (!q
->limits
.max_write_zeroes_sectors
)
936 * Various block parts want %current->io_context and lazy ioc
937 * allocation ends up trading a lot of pain for a small amount of
938 * memory. Just allocate it upfront. This may fail and block
939 * layer knows how to live with it.
941 create_io_context(GFP_ATOMIC
, q
->node
);
943 if (!blkcg_bio_issue_check(q
, bio
))
946 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
947 trace_block_bio_queue(q
, bio
);
948 /* Now that enqueuing has been traced, we need to trace
949 * completion as well.
951 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
956 status
= BLK_STS_NOTSUPP
;
958 bio
->bi_status
= status
;
964 * generic_make_request - hand a buffer to its device driver for I/O
965 * @bio: The bio describing the location in memory and on the device.
967 * generic_make_request() is used to make I/O requests of block
968 * devices. It is passed a &struct bio, which describes the I/O that needs
971 * generic_make_request() does not return any status. The
972 * success/failure status of the request, along with notification of
973 * completion, is delivered asynchronously through the bio->bi_end_io
974 * function described (one day) else where.
976 * The caller of generic_make_request must make sure that bi_io_vec
977 * are set to describe the memory buffer, and that bi_dev and bi_sector are
978 * set to describe the device address, and the
979 * bi_end_io and optionally bi_private are set to describe how
980 * completion notification should be signaled.
982 * generic_make_request and the drivers it calls may use bi_next if this
983 * bio happens to be merged with someone else, and may resubmit the bio to
984 * a lower device by calling into generic_make_request recursively, which
985 * means the bio should NOT be touched after the call to ->make_request_fn.
987 blk_qc_t
generic_make_request(struct bio
*bio
)
990 * bio_list_on_stack[0] contains bios submitted by the current
992 * bio_list_on_stack[1] contains bios that were submitted before
993 * the current make_request_fn, but that haven't been processed
996 struct bio_list bio_list_on_stack
[2];
997 blk_mq_req_flags_t flags
= 0;
998 struct request_queue
*q
= bio
->bi_disk
->queue
;
999 blk_qc_t ret
= BLK_QC_T_NONE
;
1001 if (bio
->bi_opf
& REQ_NOWAIT
)
1002 flags
= BLK_MQ_REQ_NOWAIT
;
1003 if (bio_flagged(bio
, BIO_QUEUE_ENTERED
))
1004 blk_queue_enter_live(q
);
1005 else if (blk_queue_enter(q
, flags
) < 0) {
1006 if (!blk_queue_dying(q
) && (bio
->bi_opf
& REQ_NOWAIT
))
1007 bio_wouldblock_error(bio
);
1013 if (!generic_make_request_checks(bio
))
1017 * We only want one ->make_request_fn to be active at a time, else
1018 * stack usage with stacked devices could be a problem. So use
1019 * current->bio_list to keep a list of requests submited by a
1020 * make_request_fn function. current->bio_list is also used as a
1021 * flag to say if generic_make_request is currently active in this
1022 * task or not. If it is NULL, then no make_request is active. If
1023 * it is non-NULL, then a make_request is active, and new requests
1024 * should be added at the tail
1026 if (current
->bio_list
) {
1027 bio_list_add(¤t
->bio_list
[0], bio
);
1031 /* following loop may be a bit non-obvious, and so deserves some
1033 * Before entering the loop, bio->bi_next is NULL (as all callers
1034 * ensure that) so we have a list with a single bio.
1035 * We pretend that we have just taken it off a longer list, so
1036 * we assign bio_list to a pointer to the bio_list_on_stack,
1037 * thus initialising the bio_list of new bios to be
1038 * added. ->make_request() may indeed add some more bios
1039 * through a recursive call to generic_make_request. If it
1040 * did, we find a non-NULL value in bio_list and re-enter the loop
1041 * from the top. In this case we really did just take the bio
1042 * of the top of the list (no pretending) and so remove it from
1043 * bio_list, and call into ->make_request() again.
1045 BUG_ON(bio
->bi_next
);
1046 bio_list_init(&bio_list_on_stack
[0]);
1047 current
->bio_list
= bio_list_on_stack
;
1049 bool enter_succeeded
= true;
1051 if (unlikely(q
!= bio
->bi_disk
->queue
)) {
1054 q
= bio
->bi_disk
->queue
;
1056 if (bio
->bi_opf
& REQ_NOWAIT
)
1057 flags
= BLK_MQ_REQ_NOWAIT
;
1058 if (blk_queue_enter(q
, flags
) < 0) {
1059 enter_succeeded
= false;
1064 if (enter_succeeded
) {
1065 struct bio_list lower
, same
;
1067 /* Create a fresh bio_list for all subordinate requests */
1068 bio_list_on_stack
[1] = bio_list_on_stack
[0];
1069 bio_list_init(&bio_list_on_stack
[0]);
1070 ret
= q
->make_request_fn(q
, bio
);
1072 /* sort new bios into those for a lower level
1073 * and those for the same level
1075 bio_list_init(&lower
);
1076 bio_list_init(&same
);
1077 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
1078 if (q
== bio
->bi_disk
->queue
)
1079 bio_list_add(&same
, bio
);
1081 bio_list_add(&lower
, bio
);
1082 /* now assemble so we handle the lowest level first */
1083 bio_list_merge(&bio_list_on_stack
[0], &lower
);
1084 bio_list_merge(&bio_list_on_stack
[0], &same
);
1085 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
1087 if (unlikely(!blk_queue_dying(q
) &&
1088 (bio
->bi_opf
& REQ_NOWAIT
)))
1089 bio_wouldblock_error(bio
);
1093 bio
= bio_list_pop(&bio_list_on_stack
[0]);
1095 current
->bio_list
= NULL
; /* deactivate */
1102 EXPORT_SYMBOL(generic_make_request
);
1105 * direct_make_request - hand a buffer directly to its device driver for I/O
1106 * @bio: The bio describing the location in memory and on the device.
1108 * This function behaves like generic_make_request(), but does not protect
1109 * against recursion. Must only be used if the called driver is known
1110 * to not call generic_make_request (or direct_make_request) again from
1111 * its make_request function. (Calling direct_make_request again from
1112 * a workqueue is perfectly fine as that doesn't recurse).
1114 blk_qc_t
direct_make_request(struct bio
*bio
)
1116 struct request_queue
*q
= bio
->bi_disk
->queue
;
1117 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
1120 if (!generic_make_request_checks(bio
))
1121 return BLK_QC_T_NONE
;
1123 if (unlikely(blk_queue_enter(q
, nowait
? BLK_MQ_REQ_NOWAIT
: 0))) {
1124 if (nowait
&& !blk_queue_dying(q
))
1125 bio
->bi_status
= BLK_STS_AGAIN
;
1127 bio
->bi_status
= BLK_STS_IOERR
;
1129 return BLK_QC_T_NONE
;
1132 ret
= q
->make_request_fn(q
, bio
);
1136 EXPORT_SYMBOL_GPL(direct_make_request
);
1139 * submit_bio - submit a bio to the block device layer for I/O
1140 * @bio: The &struct bio which describes the I/O
1142 * submit_bio() is very similar in purpose to generic_make_request(), and
1143 * uses that function to do most of the work. Both are fairly rough
1144 * interfaces; @bio must be presetup and ready for I/O.
1147 blk_qc_t
submit_bio(struct bio
*bio
)
1150 * If it's a regular read/write or a barrier with data attached,
1151 * go through the normal accounting stuff before submission.
1153 if (bio_has_data(bio
)) {
1156 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1157 count
= queue_logical_block_size(bio
->bi_disk
->queue
) >> 9;
1159 count
= bio_sectors(bio
);
1161 if (op_is_write(bio_op(bio
))) {
1162 count_vm_events(PGPGOUT
, count
);
1164 task_io_account_read(bio
->bi_iter
.bi_size
);
1165 count_vm_events(PGPGIN
, count
);
1168 if (unlikely(block_dump
)) {
1169 char b
[BDEVNAME_SIZE
];
1170 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1171 current
->comm
, task_pid_nr(current
),
1172 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
1173 (unsigned long long)bio
->bi_iter
.bi_sector
,
1174 bio_devname(bio
, b
), count
);
1178 return generic_make_request(bio
);
1180 EXPORT_SYMBOL(submit_bio
);
1183 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1184 * for new the queue limits
1186 * @rq: the request being checked
1189 * @rq may have been made based on weaker limitations of upper-level queues
1190 * in request stacking drivers, and it may violate the limitation of @q.
1191 * Since the block layer and the underlying device driver trust @rq
1192 * after it is inserted to @q, it should be checked against @q before
1193 * the insertion using this generic function.
1195 * Request stacking drivers like request-based dm may change the queue
1196 * limits when retrying requests on other queues. Those requests need
1197 * to be checked against the new queue limits again during dispatch.
1199 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
1202 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
1203 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1208 * queue's settings related to segment counting like q->bounce_pfn
1209 * may differ from that of other stacking queues.
1210 * Recalculate it to check the request correctly on this queue's
1213 blk_recalc_rq_segments(rq
);
1214 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1215 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1223 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1224 * @q: the queue to submit the request
1225 * @rq: the request being queued
1227 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1229 if (blk_cloned_rq_check_limits(q
, rq
))
1230 return BLK_STS_IOERR
;
1233 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1234 return BLK_STS_IOERR
;
1236 if (blk_queue_io_stat(q
))
1237 blk_account_io_start(rq
, true);
1240 * Since we have a scheduler attached on the top device,
1241 * bypass a potential scheduler on the bottom device for
1244 return blk_mq_request_issue_directly(rq
, true);
1246 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1249 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1250 * @rq: request to examine
1253 * A request could be merge of IOs which require different failure
1254 * handling. This function determines the number of bytes which
1255 * can be failed from the beginning of the request without
1256 * crossing into area which need to be retried further.
1259 * The number of bytes to fail.
1261 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1263 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1264 unsigned int bytes
= 0;
1267 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
1268 return blk_rq_bytes(rq
);
1271 * Currently the only 'mixing' which can happen is between
1272 * different fastfail types. We can safely fail portions
1273 * which have all the failfast bits that the first one has -
1274 * the ones which are at least as eager to fail as the first
1277 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1278 if ((bio
->bi_opf
& ff
) != ff
)
1280 bytes
+= bio
->bi_iter
.bi_size
;
1283 /* this could lead to infinite loop */
1284 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1287 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1289 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1291 if (blk_do_io_stat(req
)) {
1292 const int sgrp
= op_stat_group(req_op(req
));
1293 struct hd_struct
*part
;
1297 part_stat_add(part
, sectors
[sgrp
], bytes
>> 9);
1302 void blk_account_io_done(struct request
*req
, u64 now
)
1305 * Account IO completion. flush_rq isn't accounted as a
1306 * normal IO on queueing nor completion. Accounting the
1307 * containing request is enough.
1309 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
1310 const int sgrp
= op_stat_group(req_op(req
));
1311 struct hd_struct
*part
;
1316 update_io_ticks(part
, jiffies
);
1317 part_stat_inc(part
, ios
[sgrp
]);
1318 part_stat_add(part
, nsecs
[sgrp
], now
- req
->start_time_ns
);
1319 part_stat_add(part
, time_in_queue
, nsecs_to_jiffies64(now
- req
->start_time_ns
));
1320 part_dec_in_flight(req
->q
, part
, rq_data_dir(req
));
1322 hd_struct_put(part
);
1327 void blk_account_io_start(struct request
*rq
, bool new_io
)
1329 struct hd_struct
*part
;
1330 int rw
= rq_data_dir(rq
);
1332 if (!blk_do_io_stat(rq
))
1339 part_stat_inc(part
, merges
[rw
]);
1341 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
1342 if (!hd_struct_try_get(part
)) {
1344 * The partition is already being removed,
1345 * the request will be accounted on the disk only
1347 * We take a reference on disk->part0 although that
1348 * partition will never be deleted, so we can treat
1349 * it as any other partition.
1351 part
= &rq
->rq_disk
->part0
;
1352 hd_struct_get(part
);
1354 part_inc_in_flight(rq
->q
, part
, rw
);
1358 update_io_ticks(part
, jiffies
);
1364 * Steal bios from a request and add them to a bio list.
1365 * The request must not have been partially completed before.
1367 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
1371 list
->tail
->bi_next
= rq
->bio
;
1373 list
->head
= rq
->bio
;
1374 list
->tail
= rq
->biotail
;
1382 EXPORT_SYMBOL_GPL(blk_steal_bios
);
1385 * blk_update_request - Special helper function for request stacking drivers
1386 * @req: the request being processed
1387 * @error: block status code
1388 * @nr_bytes: number of bytes to complete @req
1391 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1392 * the request structure even if @req doesn't have leftover.
1393 * If @req has leftover, sets it up for the next range of segments.
1395 * This special helper function is only for request stacking drivers
1396 * (e.g. request-based dm) so that they can handle partial completion.
1397 * Actual device drivers should use blk_end_request instead.
1399 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1400 * %false return from this function.
1403 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1404 * blk_rq_bytes() and in blk_update_request().
1407 * %false - this request doesn't have any more data
1408 * %true - this request has more data
1410 bool blk_update_request(struct request
*req
, blk_status_t error
,
1411 unsigned int nr_bytes
)
1415 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
1420 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
1421 !(req
->rq_flags
& RQF_QUIET
)))
1422 print_req_error(req
, error
);
1424 blk_account_io_completion(req
, nr_bytes
);
1428 struct bio
*bio
= req
->bio
;
1429 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
1431 if (bio_bytes
== bio
->bi_iter
.bi_size
)
1432 req
->bio
= bio
->bi_next
;
1434 /* Completion has already been traced */
1435 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1436 req_bio_endio(req
, bio
, bio_bytes
, error
);
1438 total_bytes
+= bio_bytes
;
1439 nr_bytes
-= bio_bytes
;
1450 * Reset counters so that the request stacking driver
1451 * can find how many bytes remain in the request
1454 req
->__data_len
= 0;
1458 req
->__data_len
-= total_bytes
;
1460 /* update sector only for requests with clear definition of sector */
1461 if (!blk_rq_is_passthrough(req
))
1462 req
->__sector
+= total_bytes
>> 9;
1464 /* mixed attributes always follow the first bio */
1465 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
1466 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
1467 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
1470 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
1472 * If total number of sectors is less than the first segment
1473 * size, something has gone terribly wrong.
1475 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
1476 blk_dump_rq_flags(req
, "request botched");
1477 req
->__data_len
= blk_rq_cur_bytes(req
);
1480 /* recalculate the number of segments */
1481 blk_recalc_rq_segments(req
);
1486 EXPORT_SYMBOL_GPL(blk_update_request
);
1488 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
1491 if (bio_has_data(bio
))
1492 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
1493 else if (bio_op(bio
) == REQ_OP_DISCARD
)
1494 rq
->nr_phys_segments
= 1;
1496 rq
->__data_len
= bio
->bi_iter
.bi_size
;
1497 rq
->bio
= rq
->biotail
= bio
;
1500 rq
->rq_disk
= bio
->bi_disk
;
1503 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1505 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1506 * @rq: the request to be flushed
1509 * Flush all pages in @rq.
1511 void rq_flush_dcache_pages(struct request
*rq
)
1513 struct req_iterator iter
;
1514 struct bio_vec bvec
;
1516 rq_for_each_segment(bvec
, rq
, iter
)
1517 flush_dcache_page(bvec
.bv_page
);
1519 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
1523 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1524 * @q : the queue of the device being checked
1527 * Check if underlying low-level drivers of a device are busy.
1528 * If the drivers want to export their busy state, they must set own
1529 * exporting function using blk_queue_lld_busy() first.
1531 * Basically, this function is used only by request stacking drivers
1532 * to stop dispatching requests to underlying devices when underlying
1533 * devices are busy. This behavior helps more I/O merging on the queue
1534 * of the request stacking driver and prevents I/O throughput regression
1535 * on burst I/O load.
1538 * 0 - Not busy (The request stacking driver should dispatch request)
1539 * 1 - Busy (The request stacking driver should stop dispatching request)
1541 int blk_lld_busy(struct request_queue
*q
)
1543 if (queue_is_mq(q
) && q
->mq_ops
->busy
)
1544 return q
->mq_ops
->busy(q
);
1548 EXPORT_SYMBOL_GPL(blk_lld_busy
);
1551 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1552 * @rq: the clone request to be cleaned up
1555 * Free all bios in @rq for a cloned request.
1557 void blk_rq_unprep_clone(struct request
*rq
)
1561 while ((bio
= rq
->bio
) != NULL
) {
1562 rq
->bio
= bio
->bi_next
;
1567 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
1570 * Copy attributes of the original request to the clone request.
1571 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1573 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
1575 dst
->__sector
= blk_rq_pos(src
);
1576 dst
->__data_len
= blk_rq_bytes(src
);
1577 if (src
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
1578 dst
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
1579 dst
->special_vec
= src
->special_vec
;
1581 dst
->nr_phys_segments
= src
->nr_phys_segments
;
1582 dst
->ioprio
= src
->ioprio
;
1583 dst
->extra_len
= src
->extra_len
;
1587 * blk_rq_prep_clone - Helper function to setup clone request
1588 * @rq: the request to be setup
1589 * @rq_src: original request to be cloned
1590 * @bs: bio_set that bios for clone are allocated from
1591 * @gfp_mask: memory allocation mask for bio
1592 * @bio_ctr: setup function to be called for each clone bio.
1593 * Returns %0 for success, non %0 for failure.
1594 * @data: private data to be passed to @bio_ctr
1597 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1598 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1599 * are not copied, and copying such parts is the caller's responsibility.
1600 * Also, pages which the original bios are pointing to are not copied
1601 * and the cloned bios just point same pages.
1602 * So cloned bios must be completed before original bios, which means
1603 * the caller must complete @rq before @rq_src.
1605 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
1606 struct bio_set
*bs
, gfp_t gfp_mask
,
1607 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
1610 struct bio
*bio
, *bio_src
;
1615 __rq_for_each_bio(bio_src
, rq_src
) {
1616 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
1620 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
1624 rq
->biotail
->bi_next
= bio
;
1627 rq
->bio
= rq
->biotail
= bio
;
1630 __blk_rq_prep_clone(rq
, rq_src
);
1637 blk_rq_unprep_clone(rq
);
1641 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
1643 int kblockd_schedule_work(struct work_struct
*work
)
1645 return queue_work(kblockd_workqueue
, work
);
1647 EXPORT_SYMBOL(kblockd_schedule_work
);
1649 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
1651 return queue_work_on(cpu
, kblockd_workqueue
, work
);
1653 EXPORT_SYMBOL(kblockd_schedule_work_on
);
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;
1701 * Store ordering should not be needed here, since a potential
1702 * preempt will imply a full memory barrier
1706 EXPORT_SYMBOL(blk_start_plug
);
1708 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
1710 LIST_HEAD(callbacks
);
1712 while (!list_empty(&plug
->cb_list
)) {
1713 list_splice_init(&plug
->cb_list
, &callbacks
);
1715 while (!list_empty(&callbacks
)) {
1716 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
1719 list_del(&cb
->list
);
1720 cb
->callback(cb
, from_schedule
);
1725 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
1728 struct blk_plug
*plug
= current
->plug
;
1729 struct blk_plug_cb
*cb
;
1734 list_for_each_entry(cb
, &plug
->cb_list
, list
)
1735 if (cb
->callback
== unplug
&& cb
->data
== data
)
1738 /* Not currently on the callback list */
1739 BUG_ON(size
< sizeof(*cb
));
1740 cb
= kzalloc(size
, GFP_ATOMIC
);
1743 cb
->callback
= unplug
;
1744 list_add(&cb
->list
, &plug
->cb_list
);
1748 EXPORT_SYMBOL(blk_check_plugged
);
1750 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
1752 flush_plug_callbacks(plug
, from_schedule
);
1754 if (!list_empty(&plug
->mq_list
))
1755 blk_mq_flush_plug_list(plug
, from_schedule
);
1759 * blk_finish_plug - mark the end of a batch of submitted I/O
1760 * @plug: The &struct blk_plug passed to blk_start_plug()
1763 * Indicate that a batch of I/O submissions is complete. This function
1764 * must be paired with an initial call to blk_start_plug(). The intent
1765 * is to allow the block layer to optimize I/O submission. See the
1766 * documentation for blk_start_plug() for more information.
1768 void blk_finish_plug(struct blk_plug
*plug
)
1770 if (plug
!= current
->plug
)
1772 blk_flush_plug_list(plug
, false);
1774 current
->plug
= NULL
;
1776 EXPORT_SYMBOL(blk_finish_plug
);
1778 int __init
blk_dev_init(void)
1780 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
1781 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1782 FIELD_SIZEOF(struct request
, cmd_flags
));
1783 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
1784 FIELD_SIZEOF(struct bio
, bi_opf
));
1786 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1787 kblockd_workqueue
= alloc_workqueue("kblockd",
1788 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
1789 if (!kblockd_workqueue
)
1790 panic("Failed to create kblockd\n");
1792 blk_requestq_cachep
= kmem_cache_create("request_queue",
1793 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
1795 #ifdef CONFIG_DEBUG_FS
1796 blk_debugfs_root
= debugfs_create_dir("block", NULL
);