1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/backing-dev.h>
18 #include <linux/bio.h>
19 #include <linux/blkdev.h>
20 #include <linux/blk-mq.h>
21 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/block.h>
48 #include "blk-mq-sched.h"
50 #include "blk-rq-qos.h"
52 #ifdef CONFIG_DEBUG_FS
53 struct dentry *blk_debugfs_root;
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
62 DEFINE_IDA(blk_queue_ida);
65 * For queue allocation
67 struct kmem_cache *blk_requestq_cachep;
70 * Controlling structure to kblockd
72 static struct workqueue_struct *kblockd_workqueue;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 set_bit(flag, &q->queue_flags);
83 EXPORT_SYMBOL(blk_queue_flag_set);
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 clear_bit(flag, &q->queue_flags);
94 EXPORT_SYMBOL(blk_queue_flag_clear);
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106 return test_and_set_bit(flag, &q->queue_flags);
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110 void blk_rq_init(struct request_queue *q, struct request *rq)
112 memset(rq, 0, sizeof(*rq));
114 INIT_LIST_HEAD(&rq->queuelist);
116 rq->__sector = (sector_t) -1;
117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
120 rq->internal_tag = -1;
121 rq->start_time_ns = ktime_get_ns();
123 refcount_set(&rq->ref, 1);
125 EXPORT_SYMBOL(blk_rq_init);
127 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
128 static const char *const blk_op_name[] = {
132 REQ_OP_NAME(DISCARD),
133 REQ_OP_NAME(SECURE_ERASE),
134 REQ_OP_NAME(ZONE_RESET),
135 REQ_OP_NAME(ZONE_RESET_ALL),
136 REQ_OP_NAME(ZONE_OPEN),
137 REQ_OP_NAME(ZONE_CLOSE),
138 REQ_OP_NAME(ZONE_FINISH),
139 REQ_OP_NAME(WRITE_SAME),
140 REQ_OP_NAME(WRITE_ZEROES),
141 REQ_OP_NAME(SCSI_IN),
142 REQ_OP_NAME(SCSI_OUT),
144 REQ_OP_NAME(DRV_OUT),
149 * blk_op_str - Return string XXX in the REQ_OP_XXX.
152 * Description: Centralize block layer function to convert REQ_OP_XXX into
153 * string format. Useful in the debugging and tracing bio or request. For
154 * invalid REQ_OP_XXX it returns string "UNKNOWN".
156 inline const char *blk_op_str(unsigned int op)
158 const char *op_str = "UNKNOWN";
160 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
161 op_str = blk_op_name[op];
165 EXPORT_SYMBOL_GPL(blk_op_str);
167 static const struct {
171 [BLK_STS_OK] = { 0, "" },
172 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
173 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
174 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
175 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
176 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
177 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
178 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
179 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
180 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
181 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
182 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
184 /* device mapper special case, should not leak out: */
185 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
187 /* everything else not covered above: */
188 [BLK_STS_IOERR] = { -EIO, "I/O" },
191 blk_status_t errno_to_blk_status(int errno)
195 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
196 if (blk_errors[i].errno == errno)
197 return (__force blk_status_t)i;
200 return BLK_STS_IOERR;
202 EXPORT_SYMBOL_GPL(errno_to_blk_status);
204 int blk_status_to_errno(blk_status_t status)
206 int idx = (__force int)status;
208 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
210 return blk_errors[idx].errno;
212 EXPORT_SYMBOL_GPL(blk_status_to_errno);
214 static void print_req_error(struct request *req, blk_status_t status,
217 int idx = (__force int)status;
219 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
222 printk_ratelimited(KERN_ERR
223 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
224 "phys_seg %u prio class %u\n",
225 caller, blk_errors[idx].name,
226 req->rq_disk ? req->rq_disk->disk_name : "?",
227 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
228 req->cmd_flags & ~REQ_OP_MASK,
229 req->nr_phys_segments,
230 IOPRIO_PRIO_CLASS(req->ioprio));
233 static void req_bio_endio(struct request *rq, struct bio *bio,
234 unsigned int nbytes, blk_status_t error)
237 bio->bi_status = error;
239 if (unlikely(rq->rq_flags & RQF_QUIET))
240 bio_set_flag(bio, BIO_QUIET);
242 bio_advance(bio, nbytes);
244 /* don't actually finish bio if it's part of flush sequence */
245 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
249 void blk_dump_rq_flags(struct request *rq, char *msg)
251 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
252 rq->rq_disk ? rq->rq_disk->disk_name : "?",
253 (unsigned long long) rq->cmd_flags);
255 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
256 (unsigned long long)blk_rq_pos(rq),
257 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
258 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
259 rq->bio, rq->biotail, blk_rq_bytes(rq));
261 EXPORT_SYMBOL(blk_dump_rq_flags);
264 * blk_sync_queue - cancel any pending callbacks on a queue
268 * The block layer may perform asynchronous callback activity
269 * on a queue, such as calling the unplug function after a timeout.
270 * A block device may call blk_sync_queue to ensure that any
271 * such activity is cancelled, thus allowing it to release resources
272 * that the callbacks might use. The caller must already have made sure
273 * that its ->make_request_fn will not re-add plugging prior to calling
276 * This function does not cancel any asynchronous activity arising
277 * out of elevator or throttling code. That would require elevator_exit()
278 * and blkcg_exit_queue() to be called with queue lock initialized.
281 void blk_sync_queue(struct request_queue *q)
283 del_timer_sync(&q->timeout);
284 cancel_work_sync(&q->timeout_work);
286 EXPORT_SYMBOL(blk_sync_queue);
289 * blk_set_pm_only - increment pm_only counter
290 * @q: request queue pointer
292 void blk_set_pm_only(struct request_queue *q)
294 atomic_inc(&q->pm_only);
296 EXPORT_SYMBOL_GPL(blk_set_pm_only);
298 void blk_clear_pm_only(struct request_queue *q)
302 pm_only = atomic_dec_return(&q->pm_only);
303 WARN_ON_ONCE(pm_only < 0);
305 wake_up_all(&q->mq_freeze_wq);
307 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
309 void blk_put_queue(struct request_queue *q)
311 kobject_put(&q->kobj);
313 EXPORT_SYMBOL(blk_put_queue);
315 void blk_set_queue_dying(struct request_queue *q)
317 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
320 * When queue DYING flag is set, we need to block new req
321 * entering queue, so we call blk_freeze_queue_start() to
322 * prevent I/O from crossing blk_queue_enter().
324 blk_freeze_queue_start(q);
327 blk_mq_wake_waiters(q);
329 /* Make blk_queue_enter() reexamine the DYING flag. */
330 wake_up_all(&q->mq_freeze_wq);
332 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
335 * blk_cleanup_queue - shutdown a request queue
336 * @q: request queue to shutdown
338 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
339 * put it. All future requests will be failed immediately with -ENODEV.
341 void blk_cleanup_queue(struct request_queue *q)
343 WARN_ON_ONCE(blk_queue_registered(q));
345 /* mark @q DYING, no new request or merges will be allowed afterwards */
346 blk_set_queue_dying(q);
348 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
349 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
352 * Drain all requests queued before DYING marking. Set DEAD flag to
353 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
354 * after draining finished.
360 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
362 /* for synchronous bio-based driver finish in-flight integrity i/o */
363 blk_flush_integrity();
365 /* @q won't process any more request, flush async actions */
366 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
370 blk_mq_exit_queue(q);
373 * In theory, request pool of sched_tags belongs to request queue.
374 * However, the current implementation requires tag_set for freeing
375 * requests, so free the pool now.
377 * Queue has become frozen, there can't be any in-queue requests, so
378 * it is safe to free requests now.
380 mutex_lock(&q->sysfs_lock);
382 blk_mq_sched_free_requests(q);
383 mutex_unlock(&q->sysfs_lock);
385 percpu_ref_exit(&q->q_usage_counter);
387 /* @q is and will stay empty, shutdown and put */
390 EXPORT_SYMBOL(blk_cleanup_queue);
393 * blk_queue_enter() - try to increase q->q_usage_counter
394 * @q: request queue pointer
395 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
397 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
399 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
402 bool success = false;
405 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
407 * The code that increments the pm_only counter is
408 * responsible for ensuring that that counter is
409 * globally visible before the queue is unfrozen.
411 if (pm || !blk_queue_pm_only(q)) {
414 percpu_ref_put(&q->q_usage_counter);
422 if (flags & BLK_MQ_REQ_NOWAIT)
426 * read pair of barrier in blk_freeze_queue_start(),
427 * we need to order reading __PERCPU_REF_DEAD flag of
428 * .q_usage_counter and reading .mq_freeze_depth or
429 * queue dying flag, otherwise the following wait may
430 * never return if the two reads are reordered.
434 wait_event(q->mq_freeze_wq,
435 (!q->mq_freeze_depth &&
436 (pm || (blk_pm_request_resume(q),
437 !blk_queue_pm_only(q)))) ||
439 if (blk_queue_dying(q))
444 void blk_queue_exit(struct request_queue *q)
446 percpu_ref_put(&q->q_usage_counter);
449 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
451 struct request_queue *q =
452 container_of(ref, struct request_queue, q_usage_counter);
454 wake_up_all(&q->mq_freeze_wq);
457 static void blk_rq_timed_out_timer(struct timer_list *t)
459 struct request_queue *q = from_timer(q, t, timeout);
461 kblockd_schedule_work(&q->timeout_work);
464 static void blk_timeout_work(struct work_struct *work)
468 struct request_queue *__blk_alloc_queue(int node_id)
470 struct request_queue *q;
473 q = kmem_cache_alloc_node(blk_requestq_cachep,
474 GFP_KERNEL | __GFP_ZERO, node_id);
478 q->last_merge = NULL;
480 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
484 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
488 q->backing_dev_info = bdi_alloc_node(GFP_KERNEL, node_id);
489 if (!q->backing_dev_info)
492 q->stats = blk_alloc_queue_stats();
496 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
497 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
498 q->backing_dev_info->name = "block";
501 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
502 laptop_mode_timer_fn, 0);
503 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
504 INIT_WORK(&q->timeout_work, blk_timeout_work);
505 INIT_LIST_HEAD(&q->icq_list);
506 #ifdef CONFIG_BLK_CGROUP
507 INIT_LIST_HEAD(&q->blkg_list);
510 kobject_init(&q->kobj, &blk_queue_ktype);
512 #ifdef CONFIG_BLK_DEV_IO_TRACE
513 mutex_init(&q->blk_trace_mutex);
515 mutex_init(&q->sysfs_lock);
516 mutex_init(&q->sysfs_dir_lock);
517 spin_lock_init(&q->queue_lock);
519 init_waitqueue_head(&q->mq_freeze_wq);
520 mutex_init(&q->mq_freeze_lock);
523 * Init percpu_ref in atomic mode so that it's faster to shutdown.
524 * See blk_register_queue() for details.
526 if (percpu_ref_init(&q->q_usage_counter,
527 blk_queue_usage_counter_release,
528 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
531 if (blkcg_init_queue(q))
534 blk_queue_dma_alignment(q, 511);
535 blk_set_default_limits(&q->limits);
540 percpu_ref_exit(&q->q_usage_counter);
542 blk_free_queue_stats(q->stats);
544 bdi_put(q->backing_dev_info);
546 bioset_exit(&q->bio_split);
548 ida_simple_remove(&blk_queue_ida, q->id);
550 kmem_cache_free(blk_requestq_cachep, q);
554 struct request_queue *blk_alloc_queue(make_request_fn make_request, int node_id)
556 struct request_queue *q;
558 if (WARN_ON_ONCE(!make_request))
561 q = __blk_alloc_queue(node_id);
564 q->make_request_fn = make_request;
565 q->nr_requests = BLKDEV_MAX_RQ;
568 EXPORT_SYMBOL(blk_alloc_queue);
570 bool blk_get_queue(struct request_queue *q)
572 if (likely(!blk_queue_dying(q))) {
579 EXPORT_SYMBOL(blk_get_queue);
582 * blk_get_request - allocate a request
583 * @q: request queue to allocate a request for
584 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
585 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
587 struct request *blk_get_request(struct request_queue *q, unsigned int op,
588 blk_mq_req_flags_t flags)
592 WARN_ON_ONCE(op & REQ_NOWAIT);
593 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
595 req = blk_mq_alloc_request(q, op, flags);
596 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
597 q->mq_ops->initialize_rq_fn(req);
601 EXPORT_SYMBOL(blk_get_request);
603 void blk_put_request(struct request *req)
605 blk_mq_free_request(req);
607 EXPORT_SYMBOL(blk_put_request);
609 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
610 unsigned int nr_segs)
612 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
614 if (!ll_back_merge_fn(req, bio, nr_segs))
617 trace_block_bio_backmerge(req->q, req, bio);
618 rq_qos_merge(req->q, req, bio);
620 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
621 blk_rq_set_mixed_merge(req);
623 req->biotail->bi_next = bio;
625 req->__data_len += bio->bi_iter.bi_size;
627 blk_account_io_start(req, false);
631 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
632 unsigned int nr_segs)
634 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
636 if (!ll_front_merge_fn(req, bio, nr_segs))
639 trace_block_bio_frontmerge(req->q, req, bio);
640 rq_qos_merge(req->q, req, bio);
642 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
643 blk_rq_set_mixed_merge(req);
645 bio->bi_next = req->bio;
648 req->__sector = bio->bi_iter.bi_sector;
649 req->__data_len += bio->bi_iter.bi_size;
651 blk_account_io_start(req, false);
655 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
658 unsigned short segments = blk_rq_nr_discard_segments(req);
660 if (segments >= queue_max_discard_segments(q))
662 if (blk_rq_sectors(req) + bio_sectors(bio) >
663 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
666 rq_qos_merge(q, req, bio);
668 req->biotail->bi_next = bio;
670 req->__data_len += bio->bi_iter.bi_size;
671 req->nr_phys_segments = segments + 1;
673 blk_account_io_start(req, false);
676 req_set_nomerge(q, req);
681 * blk_attempt_plug_merge - try to merge with %current's plugged list
682 * @q: request_queue new bio is being queued at
683 * @bio: new bio being queued
684 * @nr_segs: number of segments in @bio
685 * @same_queue_rq: pointer to &struct request that gets filled in when
686 * another request associated with @q is found on the plug list
687 * (optional, may be %NULL)
689 * Determine whether @bio being queued on @q can be merged with a request
690 * on %current's plugged list. Returns %true if merge was successful,
693 * Plugging coalesces IOs from the same issuer for the same purpose without
694 * going through @q->queue_lock. As such it's more of an issuing mechanism
695 * than scheduling, and the request, while may have elvpriv data, is not
696 * added on the elevator at this point. In addition, we don't have
697 * reliable access to the elevator outside queue lock. Only check basic
698 * merging parameters without querying the elevator.
700 * Caller must ensure !blk_queue_nomerges(q) beforehand.
702 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
703 unsigned int nr_segs, struct request **same_queue_rq)
705 struct blk_plug *plug;
707 struct list_head *plug_list;
709 plug = blk_mq_plug(q, bio);
713 plug_list = &plug->mq_list;
715 list_for_each_entry_reverse(rq, plug_list, queuelist) {
718 if (rq->q == q && same_queue_rq) {
720 * Only blk-mq multiple hardware queues case checks the
721 * rq in the same queue, there should be only one such
727 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
730 switch (blk_try_merge(rq, bio)) {
731 case ELEVATOR_BACK_MERGE:
732 merged = bio_attempt_back_merge(rq, bio, nr_segs);
734 case ELEVATOR_FRONT_MERGE:
735 merged = bio_attempt_front_merge(rq, bio, nr_segs);
737 case ELEVATOR_DISCARD_MERGE:
738 merged = bio_attempt_discard_merge(q, rq, bio);
751 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
753 char b[BDEVNAME_SIZE];
755 printk(KERN_INFO "attempt to access beyond end of device\n");
756 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
757 bio_devname(bio, b), bio->bi_opf,
758 (unsigned long long)bio_end_sector(bio),
759 (long long)maxsector);
762 #ifdef CONFIG_FAIL_MAKE_REQUEST
764 static DECLARE_FAULT_ATTR(fail_make_request);
766 static int __init setup_fail_make_request(char *str)
768 return setup_fault_attr(&fail_make_request, str);
770 __setup("fail_make_request=", setup_fail_make_request);
772 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
774 return part->make_it_fail && should_fail(&fail_make_request, bytes);
777 static int __init fail_make_request_debugfs(void)
779 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
780 NULL, &fail_make_request);
782 return PTR_ERR_OR_ZERO(dir);
785 late_initcall(fail_make_request_debugfs);
787 #else /* CONFIG_FAIL_MAKE_REQUEST */
789 static inline bool should_fail_request(struct hd_struct *part,
795 #endif /* CONFIG_FAIL_MAKE_REQUEST */
797 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
799 const int op = bio_op(bio);
801 if (part->policy && op_is_write(op)) {
802 char b[BDEVNAME_SIZE];
804 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
808 "generic_make_request: Trying to write "
809 "to read-only block-device %s (partno %d)\n",
810 bio_devname(bio, b), part->partno);
811 /* Older lvm-tools actually trigger this */
818 static noinline int should_fail_bio(struct bio *bio)
820 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
824 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
827 * Check whether this bio extends beyond the end of the device or partition.
828 * This may well happen - the kernel calls bread() without checking the size of
829 * the device, e.g., when mounting a file system.
831 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
833 unsigned int nr_sectors = bio_sectors(bio);
835 if (nr_sectors && maxsector &&
836 (nr_sectors > maxsector ||
837 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
838 handle_bad_sector(bio, maxsector);
845 * Remap block n of partition p to block n+start(p) of the disk.
847 static inline int blk_partition_remap(struct bio *bio)
853 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
856 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
858 if (unlikely(bio_check_ro(bio, p)))
861 if (bio_sectors(bio)) {
862 if (bio_check_eod(bio, part_nr_sects_read(p)))
864 bio->bi_iter.bi_sector += p->start_sect;
865 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
866 bio->bi_iter.bi_sector - p->start_sect);
875 static noinline_for_stack bool
876 generic_make_request_checks(struct bio *bio)
878 struct request_queue *q;
879 int nr_sectors = bio_sectors(bio);
880 blk_status_t status = BLK_STS_IOERR;
881 char b[BDEVNAME_SIZE];
885 q = bio->bi_disk->queue;
888 "generic_make_request: Trying to access "
889 "nonexistent block-device %s (%Lu)\n",
890 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
895 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
896 * if queue is not a request based queue.
898 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
901 if (should_fail_bio(bio))
904 if (bio->bi_partno) {
905 if (unlikely(blk_partition_remap(bio)))
908 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
910 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
915 * Filter flush bio's early so that make_request based
916 * drivers without flush support don't have to worry
919 if (op_is_flush(bio->bi_opf) &&
920 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
921 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
928 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
929 bio->bi_opf &= ~REQ_HIPRI;
931 switch (bio_op(bio)) {
933 if (!blk_queue_discard(q))
936 case REQ_OP_SECURE_ERASE:
937 if (!blk_queue_secure_erase(q))
940 case REQ_OP_WRITE_SAME:
941 if (!q->limits.max_write_same_sectors)
944 case REQ_OP_ZONE_RESET:
945 case REQ_OP_ZONE_OPEN:
946 case REQ_OP_ZONE_CLOSE:
947 case REQ_OP_ZONE_FINISH:
948 if (!blk_queue_is_zoned(q))
951 case REQ_OP_ZONE_RESET_ALL:
952 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
955 case REQ_OP_WRITE_ZEROES:
956 if (!q->limits.max_write_zeroes_sectors)
964 * Various block parts want %current->io_context and lazy ioc
965 * allocation ends up trading a lot of pain for a small amount of
966 * memory. Just allocate it upfront. This may fail and block
967 * layer knows how to live with it.
969 create_io_context(GFP_ATOMIC, q->node);
971 if (!blkcg_bio_issue_check(q, bio))
974 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
975 trace_block_bio_queue(q, bio);
976 /* Now that enqueuing has been traced, we need to trace
977 * completion as well.
979 bio_set_flag(bio, BIO_TRACE_COMPLETION);
984 status = BLK_STS_NOTSUPP;
986 bio->bi_status = status;
992 * generic_make_request - hand a buffer to its device driver for I/O
993 * @bio: The bio describing the location in memory and on the device.
995 * generic_make_request() is used to make I/O requests of block
996 * devices. It is passed a &struct bio, which describes the I/O that needs
999 * generic_make_request() does not return any status. The
1000 * success/failure status of the request, along with notification of
1001 * completion, is delivered asynchronously through the bio->bi_end_io
1002 * function described (one day) else where.
1004 * The caller of generic_make_request must make sure that bi_io_vec
1005 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1006 * set to describe the device address, and the
1007 * bi_end_io and optionally bi_private are set to describe how
1008 * completion notification should be signaled.
1010 * generic_make_request and the drivers it calls may use bi_next if this
1011 * bio happens to be merged with someone else, and may resubmit the bio to
1012 * a lower device by calling into generic_make_request recursively, which
1013 * means the bio should NOT be touched after the call to ->make_request_fn.
1015 blk_qc_t generic_make_request(struct bio *bio)
1018 * bio_list_on_stack[0] contains bios submitted by the current
1020 * bio_list_on_stack[1] contains bios that were submitted before
1021 * the current make_request_fn, but that haven't been processed
1024 struct bio_list bio_list_on_stack[2];
1025 blk_qc_t ret = BLK_QC_T_NONE;
1027 if (!generic_make_request_checks(bio))
1031 * We only want one ->make_request_fn to be active at a time, else
1032 * stack usage with stacked devices could be a problem. So use
1033 * current->bio_list to keep a list of requests submited by a
1034 * make_request_fn function. current->bio_list is also used as a
1035 * flag to say if generic_make_request is currently active in this
1036 * task or not. If it is NULL, then no make_request is active. If
1037 * it is non-NULL, then a make_request is active, and new requests
1038 * should be added at the tail
1040 if (current->bio_list) {
1041 bio_list_add(¤t->bio_list[0], bio);
1045 /* following loop may be a bit non-obvious, and so deserves some
1047 * Before entering the loop, bio->bi_next is NULL (as all callers
1048 * ensure that) so we have a list with a single bio.
1049 * We pretend that we have just taken it off a longer list, so
1050 * we assign bio_list to a pointer to the bio_list_on_stack,
1051 * thus initialising the bio_list of new bios to be
1052 * added. ->make_request() may indeed add some more bios
1053 * through a recursive call to generic_make_request. If it
1054 * did, we find a non-NULL value in bio_list and re-enter the loop
1055 * from the top. In this case we really did just take the bio
1056 * of the top of the list (no pretending) and so remove it from
1057 * bio_list, and call into ->make_request() again.
1059 BUG_ON(bio->bi_next);
1060 bio_list_init(&bio_list_on_stack[0]);
1061 current->bio_list = bio_list_on_stack;
1063 struct request_queue *q = bio->bi_disk->queue;
1064 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1065 BLK_MQ_REQ_NOWAIT : 0;
1067 if (likely(blk_queue_enter(q, flags) == 0)) {
1068 struct bio_list lower, same;
1070 /* Create a fresh bio_list for all subordinate requests */
1071 bio_list_on_stack[1] = bio_list_on_stack[0];
1072 bio_list_init(&bio_list_on_stack[0]);
1073 ret = q->make_request_fn(q, bio);
1077 /* sort new bios into those for a lower level
1078 * and those for the same level
1080 bio_list_init(&lower);
1081 bio_list_init(&same);
1082 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1083 if (q == bio->bi_disk->queue)
1084 bio_list_add(&same, bio);
1086 bio_list_add(&lower, bio);
1087 /* now assemble so we handle the lowest level first */
1088 bio_list_merge(&bio_list_on_stack[0], &lower);
1089 bio_list_merge(&bio_list_on_stack[0], &same);
1090 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1092 if (unlikely(!blk_queue_dying(q) &&
1093 (bio->bi_opf & REQ_NOWAIT)))
1094 bio_wouldblock_error(bio);
1098 bio = bio_list_pop(&bio_list_on_stack[0]);
1100 current->bio_list = NULL; /* deactivate */
1105 EXPORT_SYMBOL(generic_make_request);
1108 * direct_make_request - hand a buffer directly to its device driver for I/O
1109 * @bio: The bio describing the location in memory and on the device.
1111 * This function behaves like generic_make_request(), but does not protect
1112 * against recursion. Must only be used if the called driver is known
1113 * to not call generic_make_request (or direct_make_request) again from
1114 * its make_request function. (Calling direct_make_request again from
1115 * a workqueue is perfectly fine as that doesn't recurse).
1117 blk_qc_t direct_make_request(struct bio *bio)
1119 struct request_queue *q = bio->bi_disk->queue;
1120 bool nowait = bio->bi_opf & REQ_NOWAIT;
1123 if (!generic_make_request_checks(bio))
1124 return BLK_QC_T_NONE;
1126 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1127 if (nowait && !blk_queue_dying(q))
1128 bio_wouldblock_error(bio);
1131 return BLK_QC_T_NONE;
1134 ret = q->make_request_fn(q, bio);
1138 EXPORT_SYMBOL_GPL(direct_make_request);
1141 * submit_bio - submit a bio to the block device layer for I/O
1142 * @bio: The &struct bio which describes the I/O
1144 * submit_bio() is very similar in purpose to generic_make_request(), and
1145 * uses that function to do most of the work. Both are fairly rough
1146 * interfaces; @bio must be presetup and ready for I/O.
1149 blk_qc_t submit_bio(struct bio *bio)
1151 bool workingset_read = false;
1152 unsigned long pflags;
1155 if (blkcg_punt_bio_submit(bio))
1156 return BLK_QC_T_NONE;
1159 * If it's a regular read/write or a barrier with data attached,
1160 * go through the normal accounting stuff before submission.
1162 if (bio_has_data(bio)) {
1165 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1166 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1168 count = bio_sectors(bio);
1170 if (op_is_write(bio_op(bio))) {
1171 count_vm_events(PGPGOUT, count);
1173 if (bio_flagged(bio, BIO_WORKINGSET))
1174 workingset_read = true;
1175 task_io_account_read(bio->bi_iter.bi_size);
1176 count_vm_events(PGPGIN, count);
1179 if (unlikely(block_dump)) {
1180 char b[BDEVNAME_SIZE];
1181 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1182 current->comm, task_pid_nr(current),
1183 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1184 (unsigned long long)bio->bi_iter.bi_sector,
1185 bio_devname(bio, b), count);
1190 * If we're reading data that is part of the userspace
1191 * workingset, count submission time as memory stall. When the
1192 * device is congested, or the submitting cgroup IO-throttled,
1193 * submission can be a significant part of overall IO time.
1195 if (workingset_read)
1196 psi_memstall_enter(&pflags);
1198 ret = generic_make_request(bio);
1200 if (workingset_read)
1201 psi_memstall_leave(&pflags);
1205 EXPORT_SYMBOL(submit_bio);
1208 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1209 * for the new queue limits
1211 * @rq: the request being checked
1214 * @rq may have been made based on weaker limitations of upper-level queues
1215 * in request stacking drivers, and it may violate the limitation of @q.
1216 * Since the block layer and the underlying device driver trust @rq
1217 * after it is inserted to @q, it should be checked against @q before
1218 * the insertion using this generic function.
1220 * Request stacking drivers like request-based dm may change the queue
1221 * limits when retrying requests on other queues. Those requests need
1222 * to be checked against the new queue limits again during dispatch.
1224 static int blk_cloned_rq_check_limits(struct request_queue *q,
1227 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1228 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1229 __func__, blk_rq_sectors(rq),
1230 blk_queue_get_max_sectors(q, req_op(rq)));
1235 * queue's settings related to segment counting like q->bounce_pfn
1236 * may differ from that of other stacking queues.
1237 * Recalculate it to check the request correctly on this queue's
1240 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1241 if (rq->nr_phys_segments > queue_max_segments(q)) {
1242 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1243 __func__, rq->nr_phys_segments, queue_max_segments(q));
1251 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1252 * @q: the queue to submit the request
1253 * @rq: the request being queued
1255 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1257 if (blk_cloned_rq_check_limits(q, rq))
1258 return BLK_STS_IOERR;
1261 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1262 return BLK_STS_IOERR;
1264 if (blk_queue_io_stat(q))
1265 blk_account_io_start(rq, true);
1268 * Since we have a scheduler attached on the top device,
1269 * bypass a potential scheduler on the bottom device for
1272 return blk_mq_request_issue_directly(rq, true);
1274 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1277 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1278 * @rq: request to examine
1281 * A request could be merge of IOs which require different failure
1282 * handling. This function determines the number of bytes which
1283 * can be failed from the beginning of the request without
1284 * crossing into area which need to be retried further.
1287 * The number of bytes to fail.
1289 unsigned int blk_rq_err_bytes(const struct request *rq)
1291 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1292 unsigned int bytes = 0;
1295 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1296 return blk_rq_bytes(rq);
1299 * Currently the only 'mixing' which can happen is between
1300 * different fastfail types. We can safely fail portions
1301 * which have all the failfast bits that the first one has -
1302 * the ones which are at least as eager to fail as the first
1305 for (bio = rq->bio; bio; bio = bio->bi_next) {
1306 if ((bio->bi_opf & ff) != ff)
1308 bytes += bio->bi_iter.bi_size;
1311 /* this could lead to infinite loop */
1312 BUG_ON(blk_rq_bytes(rq) && !bytes);
1315 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1317 void blk_account_io_completion(struct request *req, unsigned int bytes)
1319 if (req->part && blk_do_io_stat(req)) {
1320 const int sgrp = op_stat_group(req_op(req));
1321 struct hd_struct *part;
1325 part_stat_add(part, sectors[sgrp], bytes >> 9);
1330 void blk_account_io_done(struct request *req, u64 now)
1333 * Account IO completion. flush_rq isn't accounted as a
1334 * normal IO on queueing nor completion. Accounting the
1335 * containing request is enough.
1337 if (req->part && blk_do_io_stat(req) &&
1338 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1339 const int sgrp = op_stat_group(req_op(req));
1340 struct hd_struct *part;
1345 update_io_ticks(part, jiffies, true);
1346 part_stat_inc(part, ios[sgrp]);
1347 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1348 part_dec_in_flight(req->q, part, rq_data_dir(req));
1350 hd_struct_put(part);
1355 void blk_account_io_start(struct request *rq, bool new_io)
1357 struct hd_struct *part;
1358 int rw = rq_data_dir(rq);
1360 if (!blk_do_io_stat(rq))
1367 part_stat_inc(part, merges[rw]);
1369 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1370 if (!hd_struct_try_get(part)) {
1372 * The partition is already being removed,
1373 * the request will be accounted on the disk only
1375 * We take a reference on disk->part0 although that
1376 * partition will never be deleted, so we can treat
1377 * it as any other partition.
1379 part = &rq->rq_disk->part0;
1380 hd_struct_get(part);
1382 part_inc_in_flight(rq->q, part, rw);
1386 update_io_ticks(part, jiffies, false);
1392 * Steal bios from a request and add them to a bio list.
1393 * The request must not have been partially completed before.
1395 void blk_steal_bios(struct bio_list *list, struct request *rq)
1399 list->tail->bi_next = rq->bio;
1401 list->head = rq->bio;
1402 list->tail = rq->biotail;
1410 EXPORT_SYMBOL_GPL(blk_steal_bios);
1413 * blk_update_request - Special helper function for request stacking drivers
1414 * @req: the request being processed
1415 * @error: block status code
1416 * @nr_bytes: number of bytes to complete @req
1419 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1420 * the request structure even if @req doesn't have leftover.
1421 * If @req has leftover, sets it up for the next range of segments.
1423 * This special helper function is only for request stacking drivers
1424 * (e.g. request-based dm) so that they can handle partial completion.
1425 * Actual device drivers should use blk_mq_end_request instead.
1427 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1428 * %false return from this function.
1431 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1432 * blk_rq_bytes() and in blk_update_request().
1435 * %false - this request doesn't have any more data
1436 * %true - this request has more data
1438 bool blk_update_request(struct request *req, blk_status_t error,
1439 unsigned int nr_bytes)
1443 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1448 #ifdef CONFIG_BLK_DEV_INTEGRITY
1449 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1450 error == BLK_STS_OK)
1451 req->q->integrity.profile->complete_fn(req, nr_bytes);
1454 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1455 !(req->rq_flags & RQF_QUIET)))
1456 print_req_error(req, error, __func__);
1458 blk_account_io_completion(req, nr_bytes);
1462 struct bio *bio = req->bio;
1463 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1465 if (bio_bytes == bio->bi_iter.bi_size)
1466 req->bio = bio->bi_next;
1468 /* Completion has already been traced */
1469 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1470 req_bio_endio(req, bio, bio_bytes, error);
1472 total_bytes += bio_bytes;
1473 nr_bytes -= bio_bytes;
1484 * Reset counters so that the request stacking driver
1485 * can find how many bytes remain in the request
1488 req->__data_len = 0;
1492 req->__data_len -= total_bytes;
1494 /* update sector only for requests with clear definition of sector */
1495 if (!blk_rq_is_passthrough(req))
1496 req->__sector += total_bytes >> 9;
1498 /* mixed attributes always follow the first bio */
1499 if (req->rq_flags & RQF_MIXED_MERGE) {
1500 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1501 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1504 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1506 * If total number of sectors is less than the first segment
1507 * size, something has gone terribly wrong.
1509 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1510 blk_dump_rq_flags(req, "request botched");
1511 req->__data_len = blk_rq_cur_bytes(req);
1514 /* recalculate the number of segments */
1515 req->nr_phys_segments = blk_recalc_rq_segments(req);
1520 EXPORT_SYMBOL_GPL(blk_update_request);
1522 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1524 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1525 * @rq: the request to be flushed
1528 * Flush all pages in @rq.
1530 void rq_flush_dcache_pages(struct request *rq)
1532 struct req_iterator iter;
1533 struct bio_vec bvec;
1535 rq_for_each_segment(bvec, rq, iter)
1536 flush_dcache_page(bvec.bv_page);
1538 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1542 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1543 * @q : the queue of the device being checked
1546 * Check if underlying low-level drivers of a device are busy.
1547 * If the drivers want to export their busy state, they must set own
1548 * exporting function using blk_queue_lld_busy() first.
1550 * Basically, this function is used only by request stacking drivers
1551 * to stop dispatching requests to underlying devices when underlying
1552 * devices are busy. This behavior helps more I/O merging on the queue
1553 * of the request stacking driver and prevents I/O throughput regression
1554 * on burst I/O load.
1557 * 0 - Not busy (The request stacking driver should dispatch request)
1558 * 1 - Busy (The request stacking driver should stop dispatching request)
1560 int blk_lld_busy(struct request_queue *q)
1562 if (queue_is_mq(q) && q->mq_ops->busy)
1563 return q->mq_ops->busy(q);
1567 EXPORT_SYMBOL_GPL(blk_lld_busy);
1570 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1571 * @rq: the clone request to be cleaned up
1574 * Free all bios in @rq for a cloned request.
1576 void blk_rq_unprep_clone(struct request *rq)
1580 while ((bio = rq->bio) != NULL) {
1581 rq->bio = bio->bi_next;
1586 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1589 * blk_rq_prep_clone - Helper function to setup clone request
1590 * @rq: the request to be setup
1591 * @rq_src: original request to be cloned
1592 * @bs: bio_set that bios for clone are allocated from
1593 * @gfp_mask: memory allocation mask for bio
1594 * @bio_ctr: setup function to be called for each clone bio.
1595 * Returns %0 for success, non %0 for failure.
1596 * @data: private data to be passed to @bio_ctr
1599 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
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 /* Copy attributes of the original request to the clone request. */
1631 rq->__sector = blk_rq_pos(rq_src);
1632 rq->__data_len = blk_rq_bytes(rq_src);
1633 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1634 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1635 rq->special_vec = rq_src->special_vec;
1637 rq->nr_phys_segments = rq_src->nr_phys_segments;
1638 rq->ioprio = rq_src->ioprio;
1639 rq->extra_len = rq_src->extra_len;
1646 blk_rq_unprep_clone(rq);
1650 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1652 int kblockd_schedule_work(struct work_struct *work)
1654 return queue_work(kblockd_workqueue, work);
1656 EXPORT_SYMBOL(kblockd_schedule_work);
1658 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1659 unsigned long delay)
1661 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1663 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1666 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1667 * @plug: The &struct blk_plug that needs to be initialized
1670 * blk_start_plug() indicates to the block layer an intent by the caller
1671 * to submit multiple I/O requests in a batch. The block layer may use
1672 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1673 * is called. However, the block layer may choose to submit requests
1674 * before a call to blk_finish_plug() if the number of queued I/Os
1675 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1676 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1677 * the task schedules (see below).
1679 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1680 * pending I/O should the task end up blocking between blk_start_plug() and
1681 * blk_finish_plug(). This is important from a performance perspective, but
1682 * also ensures that we don't deadlock. For instance, if the task is blocking
1683 * for a memory allocation, memory reclaim could end up wanting to free a
1684 * page belonging to that request that is currently residing in our private
1685 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1686 * this kind of deadlock.
1688 void blk_start_plug(struct blk_plug *plug)
1690 struct task_struct *tsk = current;
1693 * If this is a nested plug, don't actually assign it.
1698 INIT_LIST_HEAD(&plug->mq_list);
1699 INIT_LIST_HEAD(&plug->cb_list);
1701 plug->multiple_queues = false;
1704 * Store ordering should not be needed here, since a potential
1705 * preempt will imply a full memory barrier
1709 EXPORT_SYMBOL(blk_start_plug);
1711 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1713 LIST_HEAD(callbacks);
1715 while (!list_empty(&plug->cb_list)) {
1716 list_splice_init(&plug->cb_list, &callbacks);
1718 while (!list_empty(&callbacks)) {
1719 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1722 list_del(&cb->list);
1723 cb->callback(cb, from_schedule);
1728 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1731 struct blk_plug *plug = current->plug;
1732 struct blk_plug_cb *cb;
1737 list_for_each_entry(cb, &plug->cb_list, list)
1738 if (cb->callback == unplug && cb->data == data)
1741 /* Not currently on the callback list */
1742 BUG_ON(size < sizeof(*cb));
1743 cb = kzalloc(size, GFP_ATOMIC);
1746 cb->callback = unplug;
1747 list_add(&cb->list, &plug->cb_list);
1751 EXPORT_SYMBOL(blk_check_plugged);
1753 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1755 flush_plug_callbacks(plug, from_schedule);
1757 if (!list_empty(&plug->mq_list))
1758 blk_mq_flush_plug_list(plug, from_schedule);
1762 * blk_finish_plug - mark the end of a batch of submitted I/O
1763 * @plug: The &struct blk_plug passed to blk_start_plug()
1766 * Indicate that a batch of I/O submissions is complete. This function
1767 * must be paired with an initial call to blk_start_plug(). The intent
1768 * is to allow the block layer to optimize I/O submission. See the
1769 * documentation for blk_start_plug() for more information.
1771 void blk_finish_plug(struct blk_plug *plug)
1773 if (plug != current->plug)
1775 blk_flush_plug_list(plug, false);
1777 current->plug = NULL;
1779 EXPORT_SYMBOL(blk_finish_plug);
1781 int __init blk_dev_init(void)
1783 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1784 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1785 sizeof_field(struct request, cmd_flags));
1786 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1787 sizeof_field(struct bio, bi_opf));
1789 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1790 kblockd_workqueue = alloc_workqueue("kblockd",
1791 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1792 if (!kblockd_workqueue)
1793 panic("Failed to create kblockd\n");
1795 blk_requestq_cachep = kmem_cache_create("request_queue",
1796 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1798 #ifdef CONFIG_DEBUG_FS
1799 blk_debugfs_root = debugfs_create_dir("block", NULL);