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 (atomic_read(&q->mq_freeze_depth) == 0 &&
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);
508 * Init percpu_ref in atomic mode so that it's faster to shutdown.
509 * See blk_register_queue() for details.
511 if (percpu_ref_init(&q->q_usage_counter,
512 blk_queue_usage_counter_release,
513 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
516 if (blkcg_init_queue(q))
522 percpu_ref_exit(&q->q_usage_counter);
524 blk_free_queue_stats(q->stats);
526 bdi_put(q->backing_dev_info);
528 bioset_exit(&q->bio_split);
530 ida_simple_remove(&blk_queue_ida, q->id);
532 kmem_cache_free(blk_requestq_cachep, q);
535 EXPORT_SYMBOL(blk_alloc_queue_node);
537 bool blk_get_queue(struct request_queue *q)
539 if (likely(!blk_queue_dying(q))) {
546 EXPORT_SYMBOL(blk_get_queue);
549 * blk_get_request - allocate a request
550 * @q: request queue to allocate a request for
551 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
552 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
554 struct request *blk_get_request(struct request_queue *q, unsigned int op,
555 blk_mq_req_flags_t flags)
559 WARN_ON_ONCE(op & REQ_NOWAIT);
560 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
562 req = blk_mq_alloc_request(q, op, flags);
563 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
564 q->mq_ops->initialize_rq_fn(req);
568 EXPORT_SYMBOL(blk_get_request);
570 void blk_put_request(struct request *req)
572 blk_mq_free_request(req);
574 EXPORT_SYMBOL(blk_put_request);
576 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
579 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
581 if (!ll_back_merge_fn(q, req, bio))
584 trace_block_bio_backmerge(q, req, bio);
586 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
587 blk_rq_set_mixed_merge(req);
589 req->biotail->bi_next = bio;
591 req->__data_len += bio->bi_iter.bi_size;
593 blk_account_io_start(req, false);
597 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
600 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
602 if (!ll_front_merge_fn(q, req, bio))
605 trace_block_bio_frontmerge(q, req, bio);
607 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
608 blk_rq_set_mixed_merge(req);
610 bio->bi_next = req->bio;
613 req->__sector = bio->bi_iter.bi_sector;
614 req->__data_len += bio->bi_iter.bi_size;
616 blk_account_io_start(req, false);
620 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
623 unsigned short segments = blk_rq_nr_discard_segments(req);
625 if (segments >= queue_max_discard_segments(q))
627 if (blk_rq_sectors(req) + bio_sectors(bio) >
628 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
631 req->biotail->bi_next = bio;
633 req->__data_len += bio->bi_iter.bi_size;
634 req->nr_phys_segments = segments + 1;
636 blk_account_io_start(req, false);
639 req_set_nomerge(q, req);
644 * blk_attempt_plug_merge - try to merge with %current's plugged list
645 * @q: request_queue new bio is being queued at
646 * @bio: new bio being queued
647 * @same_queue_rq: pointer to &struct request that gets filled in when
648 * another request associated with @q is found on the plug list
649 * (optional, may be %NULL)
651 * Determine whether @bio being queued on @q can be merged with a request
652 * on %current's plugged list. Returns %true if merge was successful,
655 * Plugging coalesces IOs from the same issuer for the same purpose without
656 * going through @q->queue_lock. As such it's more of an issuing mechanism
657 * than scheduling, and the request, while may have elvpriv data, is not
658 * added on the elevator at this point. In addition, we don't have
659 * reliable access to the elevator outside queue lock. Only check basic
660 * merging parameters without querying the elevator.
662 * Caller must ensure !blk_queue_nomerges(q) beforehand.
664 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
665 struct request **same_queue_rq)
667 struct blk_plug *plug;
669 struct list_head *plug_list;
671 plug = current->plug;
675 plug_list = &plug->mq_list;
677 list_for_each_entry_reverse(rq, plug_list, queuelist) {
680 if (rq->q == q && same_queue_rq) {
682 * Only blk-mq multiple hardware queues case checks the
683 * rq in the same queue, there should be only one such
689 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
692 switch (blk_try_merge(rq, bio)) {
693 case ELEVATOR_BACK_MERGE:
694 merged = bio_attempt_back_merge(q, rq, bio);
696 case ELEVATOR_FRONT_MERGE:
697 merged = bio_attempt_front_merge(q, rq, bio);
699 case ELEVATOR_DISCARD_MERGE:
700 merged = bio_attempt_discard_merge(q, rq, bio);
713 void blk_init_request_from_bio(struct request *req, struct bio *bio)
715 if (bio->bi_opf & REQ_RAHEAD)
716 req->cmd_flags |= REQ_FAILFAST_MASK;
718 req->__sector = bio->bi_iter.bi_sector;
719 req->ioprio = bio_prio(bio);
720 req->write_hint = bio->bi_write_hint;
721 blk_rq_bio_prep(req->q, req, bio);
723 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
725 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
727 char b[BDEVNAME_SIZE];
729 printk(KERN_INFO "attempt to access beyond end of device\n");
730 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
731 bio_devname(bio, b), bio->bi_opf,
732 (unsigned long long)bio_end_sector(bio),
733 (long long)maxsector);
736 #ifdef CONFIG_FAIL_MAKE_REQUEST
738 static DECLARE_FAULT_ATTR(fail_make_request);
740 static int __init setup_fail_make_request(char *str)
742 return setup_fault_attr(&fail_make_request, str);
744 __setup("fail_make_request=", setup_fail_make_request);
746 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
748 return part->make_it_fail && should_fail(&fail_make_request, bytes);
751 static int __init fail_make_request_debugfs(void)
753 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
754 NULL, &fail_make_request);
756 return PTR_ERR_OR_ZERO(dir);
759 late_initcall(fail_make_request_debugfs);
761 #else /* CONFIG_FAIL_MAKE_REQUEST */
763 static inline bool should_fail_request(struct hd_struct *part,
769 #endif /* CONFIG_FAIL_MAKE_REQUEST */
771 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
773 const int op = bio_op(bio);
775 if (part->policy && op_is_write(op)) {
776 char b[BDEVNAME_SIZE];
778 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
782 "generic_make_request: Trying to write "
783 "to read-only block-device %s (partno %d)\n",
784 bio_devname(bio, b), part->partno);
785 /* Older lvm-tools actually trigger this */
792 static noinline int should_fail_bio(struct bio *bio)
794 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
798 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
801 * Check whether this bio extends beyond the end of the device or partition.
802 * This may well happen - the kernel calls bread() without checking the size of
803 * the device, e.g., when mounting a file system.
805 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
807 unsigned int nr_sectors = bio_sectors(bio);
809 if (nr_sectors && maxsector &&
810 (nr_sectors > maxsector ||
811 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
812 handle_bad_sector(bio, maxsector);
819 * Remap block n of partition p to block n+start(p) of the disk.
821 static inline int blk_partition_remap(struct bio *bio)
827 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
830 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
832 if (unlikely(bio_check_ro(bio, p)))
836 * Zone reset does not include bi_size so bio_sectors() is always 0.
837 * Include a test for the reset op code and perform the remap if needed.
839 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
840 if (bio_check_eod(bio, part_nr_sects_read(p)))
842 bio->bi_iter.bi_sector += p->start_sect;
843 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
844 bio->bi_iter.bi_sector - p->start_sect);
853 static noinline_for_stack bool
854 generic_make_request_checks(struct bio *bio)
856 struct request_queue *q;
857 int nr_sectors = bio_sectors(bio);
858 blk_status_t status = BLK_STS_IOERR;
859 char b[BDEVNAME_SIZE];
863 q = bio->bi_disk->queue;
866 "generic_make_request: Trying to access "
867 "nonexistent block-device %s (%Lu)\n",
868 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
873 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
874 * if queue is not a request based queue.
876 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
879 if (should_fail_bio(bio))
882 if (bio->bi_partno) {
883 if (unlikely(blk_partition_remap(bio)))
886 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
888 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
893 * Filter flush bio's early so that make_request based
894 * drivers without flush support don't have to worry
897 if (op_is_flush(bio->bi_opf) &&
898 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
899 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
906 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
907 bio->bi_opf &= ~REQ_HIPRI;
909 switch (bio_op(bio)) {
911 if (!blk_queue_discard(q))
914 case REQ_OP_SECURE_ERASE:
915 if (!blk_queue_secure_erase(q))
918 case REQ_OP_WRITE_SAME:
919 if (!q->limits.max_write_same_sectors)
922 case REQ_OP_ZONE_RESET:
923 if (!blk_queue_is_zoned(q))
926 case REQ_OP_WRITE_ZEROES:
927 if (!q->limits.max_write_zeroes_sectors)
935 * Various block parts want %current->io_context and lazy ioc
936 * allocation ends up trading a lot of pain for a small amount of
937 * memory. Just allocate it upfront. This may fail and block
938 * layer knows how to live with it.
940 create_io_context(GFP_ATOMIC, q->node);
942 if (!blkcg_bio_issue_check(q, bio))
945 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
946 trace_block_bio_queue(q, bio);
947 /* Now that enqueuing has been traced, we need to trace
948 * completion as well.
950 bio_set_flag(bio, BIO_TRACE_COMPLETION);
955 status = BLK_STS_NOTSUPP;
957 bio->bi_status = status;
963 * generic_make_request - hand a buffer to its device driver for I/O
964 * @bio: The bio describing the location in memory and on the device.
966 * generic_make_request() is used to make I/O requests of block
967 * devices. It is passed a &struct bio, which describes the I/O that needs
970 * generic_make_request() does not return any status. The
971 * success/failure status of the request, along with notification of
972 * completion, is delivered asynchronously through the bio->bi_end_io
973 * function described (one day) else where.
975 * The caller of generic_make_request must make sure that bi_io_vec
976 * are set to describe the memory buffer, and that bi_dev and bi_sector are
977 * set to describe the device address, and the
978 * bi_end_io and optionally bi_private are set to describe how
979 * completion notification should be signaled.
981 * generic_make_request and the drivers it calls may use bi_next if this
982 * bio happens to be merged with someone else, and may resubmit the bio to
983 * a lower device by calling into generic_make_request recursively, which
984 * means the bio should NOT be touched after the call to ->make_request_fn.
986 blk_qc_t generic_make_request(struct bio *bio)
989 * bio_list_on_stack[0] contains bios submitted by the current
991 * bio_list_on_stack[1] contains bios that were submitted before
992 * the current make_request_fn, but that haven't been processed
995 struct bio_list bio_list_on_stack[2];
996 blk_mq_req_flags_t flags = 0;
997 struct request_queue *q = bio->bi_disk->queue;
998 blk_qc_t ret = BLK_QC_T_NONE;
1000 if (bio->bi_opf & REQ_NOWAIT)
1001 flags = BLK_MQ_REQ_NOWAIT;
1002 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
1003 blk_queue_enter_live(q);
1004 else if (blk_queue_enter(q, flags) < 0) {
1005 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
1006 bio_wouldblock_error(bio);
1012 if (!generic_make_request_checks(bio))
1016 * We only want one ->make_request_fn to be active at a time, else
1017 * stack usage with stacked devices could be a problem. So use
1018 * current->bio_list to keep a list of requests submited by a
1019 * make_request_fn function. current->bio_list is also used as a
1020 * flag to say if generic_make_request is currently active in this
1021 * task or not. If it is NULL, then no make_request is active. If
1022 * it is non-NULL, then a make_request is active, and new requests
1023 * should be added at the tail
1025 if (current->bio_list) {
1026 bio_list_add(¤t->bio_list[0], bio);
1030 /* following loop may be a bit non-obvious, and so deserves some
1032 * Before entering the loop, bio->bi_next is NULL (as all callers
1033 * ensure that) so we have a list with a single bio.
1034 * We pretend that we have just taken it off a longer list, so
1035 * we assign bio_list to a pointer to the bio_list_on_stack,
1036 * thus initialising the bio_list of new bios to be
1037 * added. ->make_request() may indeed add some more bios
1038 * through a recursive call to generic_make_request. If it
1039 * did, we find a non-NULL value in bio_list and re-enter the loop
1040 * from the top. In this case we really did just take the bio
1041 * of the top of the list (no pretending) and so remove it from
1042 * bio_list, and call into ->make_request() again.
1044 BUG_ON(bio->bi_next);
1045 bio_list_init(&bio_list_on_stack[0]);
1046 current->bio_list = bio_list_on_stack;
1048 bool enter_succeeded = true;
1050 if (unlikely(q != bio->bi_disk->queue)) {
1053 q = bio->bi_disk->queue;
1055 if (bio->bi_opf & REQ_NOWAIT)
1056 flags = BLK_MQ_REQ_NOWAIT;
1057 if (blk_queue_enter(q, flags) < 0) {
1058 enter_succeeded = false;
1063 if (enter_succeeded) {
1064 struct bio_list lower, same;
1066 /* Create a fresh bio_list for all subordinate requests */
1067 bio_list_on_stack[1] = bio_list_on_stack[0];
1068 bio_list_init(&bio_list_on_stack[0]);
1069 ret = q->make_request_fn(q, bio);
1071 /* sort new bios into those for a lower level
1072 * and those for the same level
1074 bio_list_init(&lower);
1075 bio_list_init(&same);
1076 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1077 if (q == bio->bi_disk->queue)
1078 bio_list_add(&same, bio);
1080 bio_list_add(&lower, bio);
1081 /* now assemble so we handle the lowest level first */
1082 bio_list_merge(&bio_list_on_stack[0], &lower);
1083 bio_list_merge(&bio_list_on_stack[0], &same);
1084 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1086 if (unlikely(!blk_queue_dying(q) &&
1087 (bio->bi_opf & REQ_NOWAIT)))
1088 bio_wouldblock_error(bio);
1092 bio = bio_list_pop(&bio_list_on_stack[0]);
1094 current->bio_list = NULL; /* deactivate */
1101 EXPORT_SYMBOL(generic_make_request);
1104 * direct_make_request - hand a buffer directly to its device driver for I/O
1105 * @bio: The bio describing the location in memory and on the device.
1107 * This function behaves like generic_make_request(), but does not protect
1108 * against recursion. Must only be used if the called driver is known
1109 * to not call generic_make_request (or direct_make_request) again from
1110 * its make_request function. (Calling direct_make_request again from
1111 * a workqueue is perfectly fine as that doesn't recurse).
1113 blk_qc_t direct_make_request(struct bio *bio)
1115 struct request_queue *q = bio->bi_disk->queue;
1116 bool nowait = bio->bi_opf & REQ_NOWAIT;
1119 if (!generic_make_request_checks(bio))
1120 return BLK_QC_T_NONE;
1122 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1123 if (nowait && !blk_queue_dying(q))
1124 bio->bi_status = BLK_STS_AGAIN;
1126 bio->bi_status = BLK_STS_IOERR;
1128 return BLK_QC_T_NONE;
1131 ret = q->make_request_fn(q, bio);
1135 EXPORT_SYMBOL_GPL(direct_make_request);
1138 * submit_bio - submit a bio to the block device layer for I/O
1139 * @bio: The &struct bio which describes the I/O
1141 * submit_bio() is very similar in purpose to generic_make_request(), and
1142 * uses that function to do most of the work. Both are fairly rough
1143 * interfaces; @bio must be presetup and ready for I/O.
1146 blk_qc_t submit_bio(struct bio *bio)
1149 * If it's a regular read/write or a barrier with data attached,
1150 * go through the normal accounting stuff before submission.
1152 if (bio_has_data(bio)) {
1155 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1156 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1158 count = bio_sectors(bio);
1160 if (op_is_write(bio_op(bio))) {
1161 count_vm_events(PGPGOUT, count);
1163 task_io_account_read(bio->bi_iter.bi_size);
1164 count_vm_events(PGPGIN, count);
1167 if (unlikely(block_dump)) {
1168 char b[BDEVNAME_SIZE];
1169 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1170 current->comm, task_pid_nr(current),
1171 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1172 (unsigned long long)bio->bi_iter.bi_sector,
1173 bio_devname(bio, b), count);
1177 return generic_make_request(bio);
1179 EXPORT_SYMBOL(submit_bio);
1182 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1183 * for new the queue limits
1185 * @rq: the request being checked
1188 * @rq may have been made based on weaker limitations of upper-level queues
1189 * in request stacking drivers, and it may violate the limitation of @q.
1190 * Since the block layer and the underlying device driver trust @rq
1191 * after it is inserted to @q, it should be checked against @q before
1192 * the insertion using this generic function.
1194 * Request stacking drivers like request-based dm may change the queue
1195 * limits when retrying requests on other queues. Those requests need
1196 * to be checked against the new queue limits again during dispatch.
1198 static int blk_cloned_rq_check_limits(struct request_queue *q,
1201 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1202 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1207 * queue's settings related to segment counting like q->bounce_pfn
1208 * may differ from that of other stacking queues.
1209 * Recalculate it to check the request correctly on this queue's
1212 blk_recalc_rq_segments(rq);
1213 if (rq->nr_phys_segments > queue_max_segments(q)) {
1214 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1222 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1223 * @q: the queue to submit the request
1224 * @rq: the request being queued
1226 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1228 if (blk_cloned_rq_check_limits(q, rq))
1229 return BLK_STS_IOERR;
1232 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1233 return BLK_STS_IOERR;
1235 if (blk_queue_io_stat(q))
1236 blk_account_io_start(rq, true);
1239 * Since we have a scheduler attached on the top device,
1240 * bypass a potential scheduler on the bottom device for
1243 return blk_mq_request_issue_directly(rq, true);
1245 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1248 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1249 * @rq: request to examine
1252 * A request could be merge of IOs which require different failure
1253 * handling. This function determines the number of bytes which
1254 * can be failed from the beginning of the request without
1255 * crossing into area which need to be retried further.
1258 * The number of bytes to fail.
1260 unsigned int blk_rq_err_bytes(const struct request *rq)
1262 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1263 unsigned int bytes = 0;
1266 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1267 return blk_rq_bytes(rq);
1270 * Currently the only 'mixing' which can happen is between
1271 * different fastfail types. We can safely fail portions
1272 * which have all the failfast bits that the first one has -
1273 * the ones which are at least as eager to fail as the first
1276 for (bio = rq->bio; bio; bio = bio->bi_next) {
1277 if ((bio->bi_opf & ff) != ff)
1279 bytes += bio->bi_iter.bi_size;
1282 /* this could lead to infinite loop */
1283 BUG_ON(blk_rq_bytes(rq) && !bytes);
1286 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1288 void blk_account_io_completion(struct request *req, unsigned int bytes)
1290 if (blk_do_io_stat(req)) {
1291 const int sgrp = op_stat_group(req_op(req));
1292 struct hd_struct *part;
1296 part_stat_add(part, sectors[sgrp], bytes >> 9);
1301 void blk_account_io_done(struct request *req, u64 now)
1304 * Account IO completion. flush_rq isn't accounted as a
1305 * normal IO on queueing nor completion. Accounting the
1306 * containing request is enough.
1308 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1309 const int sgrp = op_stat_group(req_op(req));
1310 struct hd_struct *part;
1315 update_io_ticks(part, jiffies);
1316 part_stat_inc(part, ios[sgrp]);
1317 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1318 part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1319 part_dec_in_flight(req->q, part, rq_data_dir(req));
1321 hd_struct_put(part);
1326 void blk_account_io_start(struct request *rq, bool new_io)
1328 struct hd_struct *part;
1329 int rw = rq_data_dir(rq);
1331 if (!blk_do_io_stat(rq))
1338 part_stat_inc(part, merges[rw]);
1340 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1341 if (!hd_struct_try_get(part)) {
1343 * The partition is already being removed,
1344 * the request will be accounted on the disk only
1346 * We take a reference on disk->part0 although that
1347 * partition will never be deleted, so we can treat
1348 * it as any other partition.
1350 part = &rq->rq_disk->part0;
1351 hd_struct_get(part);
1353 part_inc_in_flight(rq->q, part, rw);
1357 update_io_ticks(part, jiffies);
1363 * Steal bios from a request and add them to a bio list.
1364 * The request must not have been partially completed before.
1366 void blk_steal_bios(struct bio_list *list, struct request *rq)
1370 list->tail->bi_next = rq->bio;
1372 list->head = rq->bio;
1373 list->tail = rq->biotail;
1381 EXPORT_SYMBOL_GPL(blk_steal_bios);
1384 * blk_update_request - Special helper function for request stacking drivers
1385 * @req: the request being processed
1386 * @error: block status code
1387 * @nr_bytes: number of bytes to complete @req
1390 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1391 * the request structure even if @req doesn't have leftover.
1392 * If @req has leftover, sets it up for the next range of segments.
1394 * This special helper function is only for request stacking drivers
1395 * (e.g. request-based dm) so that they can handle partial completion.
1396 * Actual device drivers should use blk_end_request instead.
1398 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1399 * %false return from this function.
1402 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1403 * blk_rq_bytes() and in blk_update_request().
1406 * %false - this request doesn't have any more data
1407 * %true - this request has more data
1409 bool blk_update_request(struct request *req, blk_status_t error,
1410 unsigned int nr_bytes)
1414 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1419 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1420 !(req->rq_flags & RQF_QUIET)))
1421 print_req_error(req, error);
1423 blk_account_io_completion(req, nr_bytes);
1427 struct bio *bio = req->bio;
1428 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1430 if (bio_bytes == bio->bi_iter.bi_size)
1431 req->bio = bio->bi_next;
1433 /* Completion has already been traced */
1434 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1435 req_bio_endio(req, bio, bio_bytes, error);
1437 total_bytes += bio_bytes;
1438 nr_bytes -= bio_bytes;
1449 * Reset counters so that the request stacking driver
1450 * can find how many bytes remain in the request
1453 req->__data_len = 0;
1457 req->__data_len -= total_bytes;
1459 /* update sector only for requests with clear definition of sector */
1460 if (!blk_rq_is_passthrough(req))
1461 req->__sector += total_bytes >> 9;
1463 /* mixed attributes always follow the first bio */
1464 if (req->rq_flags & RQF_MIXED_MERGE) {
1465 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1466 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1469 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1471 * If total number of sectors is less than the first segment
1472 * size, something has gone terribly wrong.
1474 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1475 blk_dump_rq_flags(req, "request botched");
1476 req->__data_len = blk_rq_cur_bytes(req);
1479 /* recalculate the number of segments */
1480 blk_recalc_rq_segments(req);
1485 EXPORT_SYMBOL_GPL(blk_update_request);
1487 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1490 if (bio_has_data(bio))
1491 rq->nr_phys_segments = bio_phys_segments(q, bio);
1492 else if (bio_op(bio) == REQ_OP_DISCARD)
1493 rq->nr_phys_segments = 1;
1495 rq->__data_len = bio->bi_iter.bi_size;
1496 rq->bio = rq->biotail = bio;
1499 rq->rq_disk = bio->bi_disk;
1502 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1504 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1505 * @rq: the request to be flushed
1508 * Flush all pages in @rq.
1510 void rq_flush_dcache_pages(struct request *rq)
1512 struct req_iterator iter;
1513 struct bio_vec bvec;
1515 rq_for_each_segment(bvec, rq, iter)
1516 flush_dcache_page(bvec.bv_page);
1518 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1522 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1523 * @q : the queue of the device being checked
1526 * Check if underlying low-level drivers of a device are busy.
1527 * If the drivers want to export their busy state, they must set own
1528 * exporting function using blk_queue_lld_busy() first.
1530 * Basically, this function is used only by request stacking drivers
1531 * to stop dispatching requests to underlying devices when underlying
1532 * devices are busy. This behavior helps more I/O merging on the queue
1533 * of the request stacking driver and prevents I/O throughput regression
1534 * on burst I/O load.
1537 * 0 - Not busy (The request stacking driver should dispatch request)
1538 * 1 - Busy (The request stacking driver should stop dispatching request)
1540 int blk_lld_busy(struct request_queue *q)
1542 if (queue_is_mq(q) && q->mq_ops->busy)
1543 return q->mq_ops->busy(q);
1547 EXPORT_SYMBOL_GPL(blk_lld_busy);
1550 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1551 * @rq: the clone request to be cleaned up
1554 * Free all bios in @rq for a cloned request.
1556 void blk_rq_unprep_clone(struct request *rq)
1560 while ((bio = rq->bio) != NULL) {
1561 rq->bio = bio->bi_next;
1566 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1569 * Copy attributes of the original request to the clone request.
1570 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1572 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1574 dst->__sector = blk_rq_pos(src);
1575 dst->__data_len = blk_rq_bytes(src);
1576 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1577 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1578 dst->special_vec = src->special_vec;
1580 dst->nr_phys_segments = src->nr_phys_segments;
1581 dst->ioprio = src->ioprio;
1582 dst->extra_len = src->extra_len;
1586 * blk_rq_prep_clone - Helper function to setup clone request
1587 * @rq: the request to be setup
1588 * @rq_src: original request to be cloned
1589 * @bs: bio_set that bios for clone are allocated from
1590 * @gfp_mask: memory allocation mask for bio
1591 * @bio_ctr: setup function to be called for each clone bio.
1592 * Returns %0 for success, non %0 for failure.
1593 * @data: private data to be passed to @bio_ctr
1596 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1597 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1598 * are not copied, and copying such parts is the caller's responsibility.
1599 * Also, pages which the original bios are pointing to are not copied
1600 * and the cloned bios just point same pages.
1601 * So cloned bios must be completed before original bios, which means
1602 * the caller must complete @rq before @rq_src.
1604 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1605 struct bio_set *bs, gfp_t gfp_mask,
1606 int (*bio_ctr)(struct bio *, struct bio *, void *),
1609 struct bio *bio, *bio_src;
1614 __rq_for_each_bio(bio_src, rq_src) {
1615 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1619 if (bio_ctr && bio_ctr(bio, bio_src, data))
1623 rq->biotail->bi_next = bio;
1626 rq->bio = rq->biotail = bio;
1629 __blk_rq_prep_clone(rq, rq_src);
1636 blk_rq_unprep_clone(rq);
1640 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1642 int kblockd_schedule_work(struct work_struct *work)
1644 return queue_work(kblockd_workqueue, work);
1646 EXPORT_SYMBOL(kblockd_schedule_work);
1648 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1650 return queue_work_on(cpu, kblockd_workqueue, work);
1652 EXPORT_SYMBOL(kblockd_schedule_work_on);
1654 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1655 unsigned long delay)
1657 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1659 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1662 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1663 * @plug: The &struct blk_plug that needs to be initialized
1666 * blk_start_plug() indicates to the block layer an intent by the caller
1667 * to submit multiple I/O requests in a batch. The block layer may use
1668 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1669 * is called. However, the block layer may choose to submit requests
1670 * before a call to blk_finish_plug() if the number of queued I/Os
1671 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1672 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1673 * the task schedules (see below).
1675 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1676 * pending I/O should the task end up blocking between blk_start_plug() and
1677 * blk_finish_plug(). This is important from a performance perspective, but
1678 * also ensures that we don't deadlock. For instance, if the task is blocking
1679 * for a memory allocation, memory reclaim could end up wanting to free a
1680 * page belonging to that request that is currently residing in our private
1681 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1682 * this kind of deadlock.
1684 void blk_start_plug(struct blk_plug *plug)
1686 struct task_struct *tsk = current;
1689 * If this is a nested plug, don't actually assign it.
1694 INIT_LIST_HEAD(&plug->mq_list);
1695 INIT_LIST_HEAD(&plug->cb_list);
1697 plug->multiple_queues = false;
1700 * Store ordering should not be needed here, since a potential
1701 * preempt will imply a full memory barrier
1705 EXPORT_SYMBOL(blk_start_plug);
1707 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1709 LIST_HEAD(callbacks);
1711 while (!list_empty(&plug->cb_list)) {
1712 list_splice_init(&plug->cb_list, &callbacks);
1714 while (!list_empty(&callbacks)) {
1715 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1718 list_del(&cb->list);
1719 cb->callback(cb, from_schedule);
1724 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1727 struct blk_plug *plug = current->plug;
1728 struct blk_plug_cb *cb;
1733 list_for_each_entry(cb, &plug->cb_list, list)
1734 if (cb->callback == unplug && cb->data == data)
1737 /* Not currently on the callback list */
1738 BUG_ON(size < sizeof(*cb));
1739 cb = kzalloc(size, GFP_ATOMIC);
1742 cb->callback = unplug;
1743 list_add(&cb->list, &plug->cb_list);
1747 EXPORT_SYMBOL(blk_check_plugged);
1749 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1751 flush_plug_callbacks(plug, from_schedule);
1753 if (!list_empty(&plug->mq_list))
1754 blk_mq_flush_plug_list(plug, from_schedule);
1758 * blk_finish_plug - mark the end of a batch of submitted I/O
1759 * @plug: The &struct blk_plug passed to blk_start_plug()
1762 * Indicate that a batch of I/O submissions is complete. This function
1763 * must be paired with an initial call to blk_start_plug(). The intent
1764 * is to allow the block layer to optimize I/O submission. See the
1765 * documentation for blk_start_plug() for more information.
1767 void blk_finish_plug(struct blk_plug *plug)
1769 if (plug != current->plug)
1771 blk_flush_plug_list(plug, false);
1773 current->plug = NULL;
1775 EXPORT_SYMBOL(blk_finish_plug);
1777 int __init blk_dev_init(void)
1779 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1780 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1781 FIELD_SIZEOF(struct request, cmd_flags));
1782 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1783 FIELD_SIZEOF(struct bio, bi_opf));
1785 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1786 kblockd_workqueue = alloc_workqueue("kblockd",
1787 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1788 if (!kblockd_workqueue)
1789 panic("Failed to create kblockd\n");
1791 blk_requestq_cachep = kmem_cache_create("request_queue",
1792 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1794 #ifdef CONFIG_DEBUG_FS
1795 blk_debugfs_root = debugfs_create_dir("block", NULL);