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/t10-pi.h>
38 #include <linux/debugfs.h>
39 #include <linux/bpf.h>
40 #include <linux/psi.h>
41 #include <linux/sched/sysctl.h>
42 #include <linux/blk-crypto.h>
44 #define CREATE_TRACE_POINTS
45 #include <trace/events/block.h>
49 #include "blk-mq-sched.h"
51 #include "blk-rq-qos.h"
53 #ifdef CONFIG_DEBUG_FS
54 struct dentry *blk_debugfs_root;
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
63 DEFINE_IDA(blk_queue_ida);
66 * For queue allocation
68 struct kmem_cache *blk_requestq_cachep;
71 * Controlling structure to kblockd
73 static struct workqueue_struct *kblockd_workqueue;
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
80 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
82 set_bit(flag, &q->queue_flags);
84 EXPORT_SYMBOL(blk_queue_flag_set);
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
91 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
93 clear_bit(flag, &q->queue_flags);
95 EXPORT_SYMBOL(blk_queue_flag_clear);
98 * blk_queue_flag_test_and_set - atomically test and set a queue flag
99 * @flag: flag to be set
102 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
103 * the flag was already set.
105 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
107 return test_and_set_bit(flag, &q->queue_flags);
109 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
111 void blk_rq_init(struct request_queue *q, struct request *rq)
113 memset(rq, 0, sizeof(*rq));
115 INIT_LIST_HEAD(&rq->queuelist);
117 rq->__sector = (sector_t) -1;
118 INIT_HLIST_NODE(&rq->hash);
119 RB_CLEAR_NODE(&rq->rb_node);
121 rq->internal_tag = -1;
122 rq->start_time_ns = ktime_get_ns();
124 refcount_set(&rq->ref, 1);
125 blk_crypto_rq_set_defaults(rq);
127 EXPORT_SYMBOL(blk_rq_init);
129 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
130 static const char *const blk_op_name[] = {
134 REQ_OP_NAME(DISCARD),
135 REQ_OP_NAME(SECURE_ERASE),
136 REQ_OP_NAME(ZONE_RESET),
137 REQ_OP_NAME(ZONE_RESET_ALL),
138 REQ_OP_NAME(ZONE_OPEN),
139 REQ_OP_NAME(ZONE_CLOSE),
140 REQ_OP_NAME(ZONE_FINISH),
141 REQ_OP_NAME(ZONE_APPEND),
142 REQ_OP_NAME(WRITE_SAME),
143 REQ_OP_NAME(WRITE_ZEROES),
144 REQ_OP_NAME(SCSI_IN),
145 REQ_OP_NAME(SCSI_OUT),
147 REQ_OP_NAME(DRV_OUT),
152 * blk_op_str - Return string XXX in the REQ_OP_XXX.
155 * Description: Centralize block layer function to convert REQ_OP_XXX into
156 * string format. Useful in the debugging and tracing bio or request. For
157 * invalid REQ_OP_XXX it returns string "UNKNOWN".
159 inline const char *blk_op_str(unsigned int op)
161 const char *op_str = "UNKNOWN";
163 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
164 op_str = blk_op_name[op];
168 EXPORT_SYMBOL_GPL(blk_op_str);
170 static const struct {
174 [BLK_STS_OK] = { 0, "" },
175 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
176 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
177 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
178 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
179 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
180 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
181 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
182 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
183 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
184 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
185 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
187 /* device mapper special case, should not leak out: */
188 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
190 /* everything else not covered above: */
191 [BLK_STS_IOERR] = { -EIO, "I/O" },
194 blk_status_t errno_to_blk_status(int errno)
198 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
199 if (blk_errors[i].errno == errno)
200 return (__force blk_status_t)i;
203 return BLK_STS_IOERR;
205 EXPORT_SYMBOL_GPL(errno_to_blk_status);
207 int blk_status_to_errno(blk_status_t status)
209 int idx = (__force int)status;
211 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
213 return blk_errors[idx].errno;
215 EXPORT_SYMBOL_GPL(blk_status_to_errno);
217 static void print_req_error(struct request *req, blk_status_t status,
220 int idx = (__force int)status;
222 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
225 printk_ratelimited(KERN_ERR
226 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
227 "phys_seg %u prio class %u\n",
228 caller, blk_errors[idx].name,
229 req->rq_disk ? req->rq_disk->disk_name : "?",
230 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
231 req->cmd_flags & ~REQ_OP_MASK,
232 req->nr_phys_segments,
233 IOPRIO_PRIO_CLASS(req->ioprio));
236 static void req_bio_endio(struct request *rq, struct bio *bio,
237 unsigned int nbytes, blk_status_t error)
240 bio->bi_status = error;
242 if (unlikely(rq->rq_flags & RQF_QUIET))
243 bio_set_flag(bio, BIO_QUIET);
245 bio_advance(bio, nbytes);
247 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
249 * Partial zone append completions cannot be supported as the
250 * BIO fragments may end up not being written sequentially.
252 if (bio->bi_iter.bi_size)
253 bio->bi_status = BLK_STS_IOERR;
255 bio->bi_iter.bi_sector = rq->__sector;
258 /* don't actually finish bio if it's part of flush sequence */
259 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
263 void blk_dump_rq_flags(struct request *rq, char *msg)
265 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
266 rq->rq_disk ? rq->rq_disk->disk_name : "?",
267 (unsigned long long) rq->cmd_flags);
269 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
270 (unsigned long long)blk_rq_pos(rq),
271 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
272 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
273 rq->bio, rq->biotail, blk_rq_bytes(rq));
275 EXPORT_SYMBOL(blk_dump_rq_flags);
278 * blk_sync_queue - cancel any pending callbacks on a queue
282 * The block layer may perform asynchronous callback activity
283 * on a queue, such as calling the unplug function after a timeout.
284 * A block device may call blk_sync_queue to ensure that any
285 * such activity is cancelled, thus allowing it to release resources
286 * that the callbacks might use. The caller must already have made sure
287 * that its ->make_request_fn will not re-add plugging prior to calling
290 * This function does not cancel any asynchronous activity arising
291 * out of elevator or throttling code. That would require elevator_exit()
292 * and blkcg_exit_queue() to be called with queue lock initialized.
295 void blk_sync_queue(struct request_queue *q)
297 del_timer_sync(&q->timeout);
298 cancel_work_sync(&q->timeout_work);
300 EXPORT_SYMBOL(blk_sync_queue);
303 * blk_set_pm_only - increment pm_only counter
304 * @q: request queue pointer
306 void blk_set_pm_only(struct request_queue *q)
308 atomic_inc(&q->pm_only);
310 EXPORT_SYMBOL_GPL(blk_set_pm_only);
312 void blk_clear_pm_only(struct request_queue *q)
316 pm_only = atomic_dec_return(&q->pm_only);
317 WARN_ON_ONCE(pm_only < 0);
319 wake_up_all(&q->mq_freeze_wq);
321 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
323 void blk_put_queue(struct request_queue *q)
325 kobject_put(&q->kobj);
327 EXPORT_SYMBOL(blk_put_queue);
329 void blk_set_queue_dying(struct request_queue *q)
331 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
334 * When queue DYING flag is set, we need to block new req
335 * entering queue, so we call blk_freeze_queue_start() to
336 * prevent I/O from crossing blk_queue_enter().
338 blk_freeze_queue_start(q);
341 blk_mq_wake_waiters(q);
343 /* Make blk_queue_enter() reexamine the DYING flag. */
344 wake_up_all(&q->mq_freeze_wq);
346 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
349 * blk_cleanup_queue - shutdown a request queue
350 * @q: request queue to shutdown
352 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
353 * put it. All future requests will be failed immediately with -ENODEV.
355 void blk_cleanup_queue(struct request_queue *q)
357 WARN_ON_ONCE(blk_queue_registered(q));
359 /* mark @q DYING, no new request or merges will be allowed afterwards */
360 blk_set_queue_dying(q);
362 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
363 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
366 * Drain all requests queued before DYING marking. Set DEAD flag to
367 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
368 * after draining finished.
374 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
376 /* for synchronous bio-based driver finish in-flight integrity i/o */
377 blk_flush_integrity();
379 /* @q won't process any more request, flush async actions */
380 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
384 blk_mq_exit_queue(q);
387 * In theory, request pool of sched_tags belongs to request queue.
388 * However, the current implementation requires tag_set for freeing
389 * requests, so free the pool now.
391 * Queue has become frozen, there can't be any in-queue requests, so
392 * it is safe to free requests now.
394 mutex_lock(&q->sysfs_lock);
396 blk_mq_sched_free_requests(q);
397 mutex_unlock(&q->sysfs_lock);
399 percpu_ref_exit(&q->q_usage_counter);
401 /* @q is and will stay empty, shutdown and put */
404 EXPORT_SYMBOL(blk_cleanup_queue);
407 * blk_queue_enter() - try to increase q->q_usage_counter
408 * @q: request queue pointer
409 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
411 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
413 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
416 bool success = false;
419 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
421 * The code that increments the pm_only counter is
422 * responsible for ensuring that that counter is
423 * globally visible before the queue is unfrozen.
425 if (pm || !blk_queue_pm_only(q)) {
428 percpu_ref_put(&q->q_usage_counter);
436 if (flags & BLK_MQ_REQ_NOWAIT)
440 * read pair of barrier in blk_freeze_queue_start(),
441 * we need to order reading __PERCPU_REF_DEAD flag of
442 * .q_usage_counter and reading .mq_freeze_depth or
443 * queue dying flag, otherwise the following wait may
444 * never return if the two reads are reordered.
448 wait_event(q->mq_freeze_wq,
449 (!q->mq_freeze_depth &&
450 (pm || (blk_pm_request_resume(q),
451 !blk_queue_pm_only(q)))) ||
453 if (blk_queue_dying(q))
458 static inline int bio_queue_enter(struct bio *bio)
460 struct request_queue *q = bio->bi_disk->queue;
461 bool nowait = bio->bi_opf & REQ_NOWAIT;
464 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
466 if (nowait && !blk_queue_dying(q))
467 bio_wouldblock_error(bio);
475 void blk_queue_exit(struct request_queue *q)
477 percpu_ref_put(&q->q_usage_counter);
480 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
482 struct request_queue *q =
483 container_of(ref, struct request_queue, q_usage_counter);
485 wake_up_all(&q->mq_freeze_wq);
488 static void blk_rq_timed_out_timer(struct timer_list *t)
490 struct request_queue *q = from_timer(q, t, timeout);
492 kblockd_schedule_work(&q->timeout_work);
495 static void blk_timeout_work(struct work_struct *work)
499 struct request_queue *__blk_alloc_queue(int node_id)
501 struct request_queue *q;
504 q = kmem_cache_alloc_node(blk_requestq_cachep,
505 GFP_KERNEL | __GFP_ZERO, node_id);
509 q->last_merge = NULL;
511 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
515 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
519 q->backing_dev_info = bdi_alloc(node_id);
520 if (!q->backing_dev_info)
523 q->stats = blk_alloc_queue_stats();
527 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
528 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
531 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
532 laptop_mode_timer_fn, 0);
533 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
534 INIT_WORK(&q->timeout_work, blk_timeout_work);
535 INIT_LIST_HEAD(&q->icq_list);
536 #ifdef CONFIG_BLK_CGROUP
537 INIT_LIST_HEAD(&q->blkg_list);
540 kobject_init(&q->kobj, &blk_queue_ktype);
542 #ifdef CONFIG_BLK_DEV_IO_TRACE
543 mutex_init(&q->blk_trace_mutex);
545 mutex_init(&q->sysfs_lock);
546 mutex_init(&q->sysfs_dir_lock);
547 spin_lock_init(&q->queue_lock);
549 init_waitqueue_head(&q->mq_freeze_wq);
550 mutex_init(&q->mq_freeze_lock);
553 * Init percpu_ref in atomic mode so that it's faster to shutdown.
554 * See blk_register_queue() for details.
556 if (percpu_ref_init(&q->q_usage_counter,
557 blk_queue_usage_counter_release,
558 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
561 if (blkcg_init_queue(q))
564 blk_queue_dma_alignment(q, 511);
565 blk_set_default_limits(&q->limits);
570 percpu_ref_exit(&q->q_usage_counter);
572 blk_free_queue_stats(q->stats);
574 bdi_put(q->backing_dev_info);
576 bioset_exit(&q->bio_split);
578 ida_simple_remove(&blk_queue_ida, q->id);
580 kmem_cache_free(blk_requestq_cachep, q);
584 struct request_queue *blk_alloc_queue(make_request_fn make_request, int node_id)
586 struct request_queue *q;
588 if (WARN_ON_ONCE(!make_request))
591 q = __blk_alloc_queue(node_id);
594 q->make_request_fn = make_request;
595 q->nr_requests = BLKDEV_MAX_RQ;
598 EXPORT_SYMBOL(blk_alloc_queue);
600 bool blk_get_queue(struct request_queue *q)
602 if (likely(!blk_queue_dying(q))) {
609 EXPORT_SYMBOL(blk_get_queue);
612 * blk_get_request - allocate a request
613 * @q: request queue to allocate a request for
614 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
615 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
617 struct request *blk_get_request(struct request_queue *q, unsigned int op,
618 blk_mq_req_flags_t flags)
622 WARN_ON_ONCE(op & REQ_NOWAIT);
623 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
625 req = blk_mq_alloc_request(q, op, flags);
626 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
627 q->mq_ops->initialize_rq_fn(req);
631 EXPORT_SYMBOL(blk_get_request);
633 void blk_put_request(struct request *req)
635 blk_mq_free_request(req);
637 EXPORT_SYMBOL(blk_put_request);
639 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
640 unsigned int nr_segs)
642 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
644 if (!ll_back_merge_fn(req, bio, nr_segs))
647 trace_block_bio_backmerge(req->q, req, bio);
648 rq_qos_merge(req->q, req, bio);
650 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
651 blk_rq_set_mixed_merge(req);
653 req->biotail->bi_next = bio;
655 req->__data_len += bio->bi_iter.bi_size;
657 bio_crypt_free_ctx(bio);
659 blk_account_io_start(req, false);
663 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
664 unsigned int nr_segs)
666 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
668 if (!ll_front_merge_fn(req, bio, nr_segs))
671 trace_block_bio_frontmerge(req->q, req, bio);
672 rq_qos_merge(req->q, req, bio);
674 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
675 blk_rq_set_mixed_merge(req);
677 bio->bi_next = req->bio;
680 req->__sector = bio->bi_iter.bi_sector;
681 req->__data_len += bio->bi_iter.bi_size;
683 bio_crypt_do_front_merge(req, bio);
685 blk_account_io_start(req, false);
689 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
692 unsigned short segments = blk_rq_nr_discard_segments(req);
694 if (segments >= queue_max_discard_segments(q))
696 if (blk_rq_sectors(req) + bio_sectors(bio) >
697 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
700 rq_qos_merge(q, req, bio);
702 req->biotail->bi_next = bio;
704 req->__data_len += bio->bi_iter.bi_size;
705 req->nr_phys_segments = segments + 1;
707 blk_account_io_start(req, false);
710 req_set_nomerge(q, req);
715 * blk_attempt_plug_merge - try to merge with %current's plugged list
716 * @q: request_queue new bio is being queued at
717 * @bio: new bio being queued
718 * @nr_segs: number of segments in @bio
719 * @same_queue_rq: pointer to &struct request that gets filled in when
720 * another request associated with @q is found on the plug list
721 * (optional, may be %NULL)
723 * Determine whether @bio being queued on @q can be merged with a request
724 * on %current's plugged list. Returns %true if merge was successful,
727 * Plugging coalesces IOs from the same issuer for the same purpose without
728 * going through @q->queue_lock. As such it's more of an issuing mechanism
729 * than scheduling, and the request, while may have elvpriv data, is not
730 * added on the elevator at this point. In addition, we don't have
731 * reliable access to the elevator outside queue lock. Only check basic
732 * merging parameters without querying the elevator.
734 * Caller must ensure !blk_queue_nomerges(q) beforehand.
736 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
737 unsigned int nr_segs, struct request **same_queue_rq)
739 struct blk_plug *plug;
741 struct list_head *plug_list;
743 plug = blk_mq_plug(q, bio);
747 plug_list = &plug->mq_list;
749 list_for_each_entry_reverse(rq, plug_list, queuelist) {
752 if (rq->q == q && same_queue_rq) {
754 * Only blk-mq multiple hardware queues case checks the
755 * rq in the same queue, there should be only one such
761 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
764 switch (blk_try_merge(rq, bio)) {
765 case ELEVATOR_BACK_MERGE:
766 merged = bio_attempt_back_merge(rq, bio, nr_segs);
768 case ELEVATOR_FRONT_MERGE:
769 merged = bio_attempt_front_merge(rq, bio, nr_segs);
771 case ELEVATOR_DISCARD_MERGE:
772 merged = bio_attempt_discard_merge(q, rq, bio);
785 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
787 char b[BDEVNAME_SIZE];
789 printk(KERN_INFO "attempt to access beyond end of device\n");
790 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
791 bio_devname(bio, b), bio->bi_opf,
792 (unsigned long long)bio_end_sector(bio),
793 (long long)maxsector);
796 #ifdef CONFIG_FAIL_MAKE_REQUEST
798 static DECLARE_FAULT_ATTR(fail_make_request);
800 static int __init setup_fail_make_request(char *str)
802 return setup_fault_attr(&fail_make_request, str);
804 __setup("fail_make_request=", setup_fail_make_request);
806 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
808 return part->make_it_fail && should_fail(&fail_make_request, bytes);
811 static int __init fail_make_request_debugfs(void)
813 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
814 NULL, &fail_make_request);
816 return PTR_ERR_OR_ZERO(dir);
819 late_initcall(fail_make_request_debugfs);
821 #else /* CONFIG_FAIL_MAKE_REQUEST */
823 static inline bool should_fail_request(struct hd_struct *part,
829 #endif /* CONFIG_FAIL_MAKE_REQUEST */
831 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
833 const int op = bio_op(bio);
835 if (part->policy && op_is_write(op)) {
836 char b[BDEVNAME_SIZE];
838 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
842 "generic_make_request: Trying to write "
843 "to read-only block-device %s (partno %d)\n",
844 bio_devname(bio, b), part->partno);
845 /* Older lvm-tools actually trigger this */
852 static noinline int should_fail_bio(struct bio *bio)
854 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
858 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
861 * Check whether this bio extends beyond the end of the device or partition.
862 * This may well happen - the kernel calls bread() without checking the size of
863 * the device, e.g., when mounting a file system.
865 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
867 unsigned int nr_sectors = bio_sectors(bio);
869 if (nr_sectors && maxsector &&
870 (nr_sectors > maxsector ||
871 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
872 handle_bad_sector(bio, maxsector);
879 * Remap block n of partition p to block n+start(p) of the disk.
881 static inline int blk_partition_remap(struct bio *bio)
887 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
890 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
892 if (unlikely(bio_check_ro(bio, p)))
895 if (bio_sectors(bio)) {
896 if (bio_check_eod(bio, part_nr_sects_read(p)))
898 bio->bi_iter.bi_sector += p->start_sect;
899 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
900 bio->bi_iter.bi_sector - p->start_sect);
910 * Check write append to a zoned block device.
912 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
915 sector_t pos = bio->bi_iter.bi_sector;
916 int nr_sectors = bio_sectors(bio);
918 /* Only applicable to zoned block devices */
919 if (!blk_queue_is_zoned(q))
920 return BLK_STS_NOTSUPP;
922 /* The bio sector must point to the start of a sequential zone */
923 if (pos & (blk_queue_zone_sectors(q) - 1) ||
924 !blk_queue_zone_is_seq(q, pos))
925 return BLK_STS_IOERR;
928 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
929 * split and could result in non-contiguous sectors being written in
932 if (nr_sectors > q->limits.chunk_sectors)
933 return BLK_STS_IOERR;
935 /* Make sure the BIO is small enough and will not get split */
936 if (nr_sectors > q->limits.max_zone_append_sectors)
937 return BLK_STS_IOERR;
939 bio->bi_opf |= REQ_NOMERGE;
944 static noinline_for_stack bool
945 generic_make_request_checks(struct bio *bio)
947 struct request_queue *q;
948 int nr_sectors = bio_sectors(bio);
949 blk_status_t status = BLK_STS_IOERR;
950 char b[BDEVNAME_SIZE];
954 q = bio->bi_disk->queue;
957 "generic_make_request: Trying to access "
958 "nonexistent block-device %s (%Lu)\n",
959 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
964 * Non-mq queues do not honor REQ_NOWAIT, so complete a bio
965 * with BLK_STS_AGAIN status in order to catch -EAGAIN and
966 * to give a chance to the caller to repeat request gracefully.
968 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) {
969 status = BLK_STS_AGAIN;
973 if (should_fail_bio(bio))
976 if (bio->bi_partno) {
977 if (unlikely(blk_partition_remap(bio)))
980 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
982 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
987 * Filter flush bio's early so that make_request based
988 * drivers without flush support don't have to worry
991 if (op_is_flush(bio->bi_opf) &&
992 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
993 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1000 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
1001 bio->bi_opf &= ~REQ_HIPRI;
1003 switch (bio_op(bio)) {
1004 case REQ_OP_DISCARD:
1005 if (!blk_queue_discard(q))
1008 case REQ_OP_SECURE_ERASE:
1009 if (!blk_queue_secure_erase(q))
1012 case REQ_OP_WRITE_SAME:
1013 if (!q->limits.max_write_same_sectors)
1016 case REQ_OP_ZONE_APPEND:
1017 status = blk_check_zone_append(q, bio);
1018 if (status != BLK_STS_OK)
1021 case REQ_OP_ZONE_RESET:
1022 case REQ_OP_ZONE_OPEN:
1023 case REQ_OP_ZONE_CLOSE:
1024 case REQ_OP_ZONE_FINISH:
1025 if (!blk_queue_is_zoned(q))
1028 case REQ_OP_ZONE_RESET_ALL:
1029 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
1032 case REQ_OP_WRITE_ZEROES:
1033 if (!q->limits.max_write_zeroes_sectors)
1041 * Various block parts want %current->io_context, so allocate it up
1042 * front rather than dealing with lots of pain to allocate it only
1043 * where needed. This may fail and the block layer knows how to live
1046 if (unlikely(!current->io_context))
1047 create_task_io_context(current, GFP_ATOMIC, q->node);
1049 if (!blkcg_bio_issue_check(q, bio))
1052 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1053 trace_block_bio_queue(q, bio);
1054 /* Now that enqueuing has been traced, we need to trace
1055 * completion as well.
1057 bio_set_flag(bio, BIO_TRACE_COMPLETION);
1062 status = BLK_STS_NOTSUPP;
1064 bio->bi_status = status;
1070 * generic_make_request - re-submit a bio to the block device layer for I/O
1071 * @bio: The bio describing the location in memory and on the device.
1073 * This is a version of submit_bio() that shall only be used for I/O that is
1074 * resubmitted to lower level drivers by stacking block drivers. All file
1075 * systems and other upper level users of the block layer should use
1076 * submit_bio() instead.
1078 blk_qc_t generic_make_request(struct bio *bio)
1081 * bio_list_on_stack[0] contains bios submitted by the current
1083 * bio_list_on_stack[1] contains bios that were submitted before
1084 * the current make_request_fn, but that haven't been processed
1087 struct bio_list bio_list_on_stack[2];
1088 blk_qc_t ret = BLK_QC_T_NONE;
1090 if (!generic_make_request_checks(bio))
1094 * We only want one ->make_request_fn to be active at a time, else
1095 * stack usage with stacked devices could be a problem. So use
1096 * current->bio_list to keep a list of requests submited by a
1097 * make_request_fn function. current->bio_list is also used as a
1098 * flag to say if generic_make_request is currently active in this
1099 * task or not. If it is NULL, then no make_request is active. If
1100 * it is non-NULL, then a make_request is active, and new requests
1101 * should be added at the tail
1103 if (current->bio_list) {
1104 bio_list_add(¤t->bio_list[0], bio);
1108 /* following loop may be a bit non-obvious, and so deserves some
1110 * Before entering the loop, bio->bi_next is NULL (as all callers
1111 * ensure that) so we have a list with a single bio.
1112 * We pretend that we have just taken it off a longer list, so
1113 * we assign bio_list to a pointer to the bio_list_on_stack,
1114 * thus initialising the bio_list of new bios to be
1115 * added. ->make_request() may indeed add some more bios
1116 * through a recursive call to generic_make_request. If it
1117 * did, we find a non-NULL value in bio_list and re-enter the loop
1118 * from the top. In this case we really did just take the bio
1119 * of the top of the list (no pretending) and so remove it from
1120 * bio_list, and call into ->make_request() again.
1122 BUG_ON(bio->bi_next);
1123 bio_list_init(&bio_list_on_stack[0]);
1124 current->bio_list = bio_list_on_stack;
1126 struct request_queue *q = bio->bi_disk->queue;
1128 if (likely(bio_queue_enter(bio) == 0)) {
1129 struct bio_list lower, same;
1131 /* Create a fresh bio_list for all subordinate requests */
1132 bio_list_on_stack[1] = bio_list_on_stack[0];
1133 bio_list_init(&bio_list_on_stack[0]);
1134 if (blk_crypto_bio_prep(&bio)) {
1135 if (q->make_request_fn)
1136 ret = q->make_request_fn(q, bio);
1138 ret = blk_mq_make_request(q, bio);
1143 /* sort new bios into those for a lower level
1144 * and those for the same level
1146 bio_list_init(&lower);
1147 bio_list_init(&same);
1148 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1149 if (q == bio->bi_disk->queue)
1150 bio_list_add(&same, bio);
1152 bio_list_add(&lower, bio);
1153 /* now assemble so we handle the lowest level first */
1154 bio_list_merge(&bio_list_on_stack[0], &lower);
1155 bio_list_merge(&bio_list_on_stack[0], &same);
1156 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1158 bio = bio_list_pop(&bio_list_on_stack[0]);
1160 current->bio_list = NULL; /* deactivate */
1165 EXPORT_SYMBOL(generic_make_request);
1168 * direct_make_request - hand a buffer directly to its device driver for I/O
1169 * @bio: The bio describing the location in memory and on the device.
1171 * This function behaves like generic_make_request(), but does not protect
1172 * against recursion. Must only be used if the called driver is known
1173 * to be blk-mq based.
1175 blk_qc_t direct_make_request(struct bio *bio)
1177 struct request_queue *q = bio->bi_disk->queue;
1178 blk_qc_t ret = BLK_QC_T_NONE;
1180 if (WARN_ON_ONCE(q->make_request_fn)) {
1182 return BLK_QC_T_NONE;
1184 if (!generic_make_request_checks(bio))
1185 return BLK_QC_T_NONE;
1186 if (unlikely(bio_queue_enter(bio)))
1187 return BLK_QC_T_NONE;
1188 if (blk_crypto_bio_prep(&bio))
1189 ret = blk_mq_make_request(q, bio);
1193 EXPORT_SYMBOL_GPL(direct_make_request);
1196 * submit_bio - submit a bio to the block device layer for I/O
1197 * @bio: The &struct bio which describes the I/O
1199 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1200 * fully set up &struct bio that describes the I/O that needs to be done. The
1201 * bio will be send to the device described by the bi_disk and bi_partno fields.
1203 * The success/failure status of the request, along with notification of
1204 * completion, is delivered asynchronously through the ->bi_end_io() callback
1205 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1208 blk_qc_t submit_bio(struct bio *bio)
1210 if (blkcg_punt_bio_submit(bio))
1211 return BLK_QC_T_NONE;
1214 * If it's a regular read/write or a barrier with data attached,
1215 * go through the normal accounting stuff before submission.
1217 if (bio_has_data(bio)) {
1220 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1221 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1223 count = bio_sectors(bio);
1225 if (op_is_write(bio_op(bio))) {
1226 count_vm_events(PGPGOUT, count);
1228 task_io_account_read(bio->bi_iter.bi_size);
1229 count_vm_events(PGPGIN, count);
1232 if (unlikely(block_dump)) {
1233 char b[BDEVNAME_SIZE];
1234 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1235 current->comm, task_pid_nr(current),
1236 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1237 (unsigned long long)bio->bi_iter.bi_sector,
1238 bio_devname(bio, b), count);
1243 * If we're reading data that is part of the userspace workingset, count
1244 * submission time as memory stall. When the device is congested, or
1245 * the submitting cgroup IO-throttled, submission can be a significant
1246 * part of overall IO time.
1248 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1249 bio_flagged(bio, BIO_WORKINGSET))) {
1250 unsigned long pflags;
1253 psi_memstall_enter(&pflags);
1254 ret = generic_make_request(bio);
1255 psi_memstall_leave(&pflags);
1260 return generic_make_request(bio);
1262 EXPORT_SYMBOL(submit_bio);
1265 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1266 * for the new queue limits
1268 * @rq: the request being checked
1271 * @rq may have been made based on weaker limitations of upper-level queues
1272 * in request stacking drivers, and it may violate the limitation of @q.
1273 * Since the block layer and the underlying device driver trust @rq
1274 * after it is inserted to @q, it should be checked against @q before
1275 * the insertion using this generic function.
1277 * Request stacking drivers like request-based dm may change the queue
1278 * limits when retrying requests on other queues. Those requests need
1279 * to be checked against the new queue limits again during dispatch.
1281 static int blk_cloned_rq_check_limits(struct request_queue *q,
1284 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1285 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1286 __func__, blk_rq_sectors(rq),
1287 blk_queue_get_max_sectors(q, req_op(rq)));
1292 * queue's settings related to segment counting like q->bounce_pfn
1293 * may differ from that of other stacking queues.
1294 * Recalculate it to check the request correctly on this queue's
1297 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1298 if (rq->nr_phys_segments > queue_max_segments(q)) {
1299 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1300 __func__, rq->nr_phys_segments, queue_max_segments(q));
1308 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1309 * @q: the queue to submit the request
1310 * @rq: the request being queued
1312 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1314 if (blk_cloned_rq_check_limits(q, rq))
1315 return BLK_STS_IOERR;
1318 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1319 return BLK_STS_IOERR;
1321 if (blk_crypto_insert_cloned_request(rq))
1322 return BLK_STS_IOERR;
1324 if (blk_queue_io_stat(q))
1325 blk_account_io_start(rq, true);
1328 * Since we have a scheduler attached on the top device,
1329 * bypass a potential scheduler on the bottom device for
1332 return blk_mq_request_issue_directly(rq, true);
1334 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1337 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1338 * @rq: request to examine
1341 * A request could be merge of IOs which require different failure
1342 * handling. This function determines the number of bytes which
1343 * can be failed from the beginning of the request without
1344 * crossing into area which need to be retried further.
1347 * The number of bytes to fail.
1349 unsigned int blk_rq_err_bytes(const struct request *rq)
1351 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1352 unsigned int bytes = 0;
1355 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1356 return blk_rq_bytes(rq);
1359 * Currently the only 'mixing' which can happen is between
1360 * different fastfail types. We can safely fail portions
1361 * which have all the failfast bits that the first one has -
1362 * the ones which are at least as eager to fail as the first
1365 for (bio = rq->bio; bio; bio = bio->bi_next) {
1366 if ((bio->bi_opf & ff) != ff)
1368 bytes += bio->bi_iter.bi_size;
1371 /* this could lead to infinite loop */
1372 BUG_ON(blk_rq_bytes(rq) && !bytes);
1375 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1377 void blk_account_io_completion(struct request *req, unsigned int bytes)
1379 if (req->part && blk_do_io_stat(req)) {
1380 const int sgrp = op_stat_group(req_op(req));
1381 struct hd_struct *part;
1385 part_stat_add(part, sectors[sgrp], bytes >> 9);
1390 void blk_account_io_done(struct request *req, u64 now)
1393 * Account IO completion. flush_rq isn't accounted as a
1394 * normal IO on queueing nor completion. Accounting the
1395 * containing request is enough.
1397 if (req->part && blk_do_io_stat(req) &&
1398 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1399 const int sgrp = op_stat_group(req_op(req));
1400 struct hd_struct *part;
1405 update_io_ticks(part, jiffies, true);
1406 part_stat_inc(part, ios[sgrp]);
1407 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1408 part_dec_in_flight(req->q, part, rq_data_dir(req));
1410 hd_struct_put(part);
1415 void blk_account_io_start(struct request *rq, bool new_io)
1417 struct hd_struct *part;
1418 int rw = rq_data_dir(rq);
1420 if (!blk_do_io_stat(rq))
1427 part_stat_inc(part, merges[rw]);
1429 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1430 part_inc_in_flight(rq->q, part, rw);
1434 update_io_ticks(part, jiffies, false);
1440 * Steal bios from a request and add them to a bio list.
1441 * The request must not have been partially completed before.
1443 void blk_steal_bios(struct bio_list *list, struct request *rq)
1447 list->tail->bi_next = rq->bio;
1449 list->head = rq->bio;
1450 list->tail = rq->biotail;
1458 EXPORT_SYMBOL_GPL(blk_steal_bios);
1461 * blk_update_request - Special helper function for request stacking drivers
1462 * @req: the request being processed
1463 * @error: block status code
1464 * @nr_bytes: number of bytes to complete @req
1467 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1468 * the request structure even if @req doesn't have leftover.
1469 * If @req has leftover, sets it up for the next range of segments.
1471 * This special helper function is only for request stacking drivers
1472 * (e.g. request-based dm) so that they can handle partial completion.
1473 * Actual device drivers should use blk_mq_end_request instead.
1475 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1476 * %false return from this function.
1479 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1480 * blk_rq_bytes() and in blk_update_request().
1483 * %false - this request doesn't have any more data
1484 * %true - this request has more data
1486 bool blk_update_request(struct request *req, blk_status_t error,
1487 unsigned int nr_bytes)
1491 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1496 #ifdef CONFIG_BLK_DEV_INTEGRITY
1497 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1498 error == BLK_STS_OK)
1499 req->q->integrity.profile->complete_fn(req, nr_bytes);
1502 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1503 !(req->rq_flags & RQF_QUIET)))
1504 print_req_error(req, error, __func__);
1506 blk_account_io_completion(req, nr_bytes);
1510 struct bio *bio = req->bio;
1511 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1513 if (bio_bytes == bio->bi_iter.bi_size)
1514 req->bio = bio->bi_next;
1516 /* Completion has already been traced */
1517 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1518 req_bio_endio(req, bio, bio_bytes, error);
1520 total_bytes += bio_bytes;
1521 nr_bytes -= bio_bytes;
1532 * Reset counters so that the request stacking driver
1533 * can find how many bytes remain in the request
1536 req->__data_len = 0;
1540 req->__data_len -= total_bytes;
1542 /* update sector only for requests with clear definition of sector */
1543 if (!blk_rq_is_passthrough(req))
1544 req->__sector += total_bytes >> 9;
1546 /* mixed attributes always follow the first bio */
1547 if (req->rq_flags & RQF_MIXED_MERGE) {
1548 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1549 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1552 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1554 * If total number of sectors is less than the first segment
1555 * size, something has gone terribly wrong.
1557 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1558 blk_dump_rq_flags(req, "request botched");
1559 req->__data_len = blk_rq_cur_bytes(req);
1562 /* recalculate the number of segments */
1563 req->nr_phys_segments = blk_recalc_rq_segments(req);
1568 EXPORT_SYMBOL_GPL(blk_update_request);
1570 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1572 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1573 * @rq: the request to be flushed
1576 * Flush all pages in @rq.
1578 void rq_flush_dcache_pages(struct request *rq)
1580 struct req_iterator iter;
1581 struct bio_vec bvec;
1583 rq_for_each_segment(bvec, rq, iter)
1584 flush_dcache_page(bvec.bv_page);
1586 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1590 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1591 * @q : the queue of the device being checked
1594 * Check if underlying low-level drivers of a device are busy.
1595 * If the drivers want to export their busy state, they must set own
1596 * exporting function using blk_queue_lld_busy() first.
1598 * Basically, this function is used only by request stacking drivers
1599 * to stop dispatching requests to underlying devices when underlying
1600 * devices are busy. This behavior helps more I/O merging on the queue
1601 * of the request stacking driver and prevents I/O throughput regression
1602 * on burst I/O load.
1605 * 0 - Not busy (The request stacking driver should dispatch request)
1606 * 1 - Busy (The request stacking driver should stop dispatching request)
1608 int blk_lld_busy(struct request_queue *q)
1610 if (queue_is_mq(q) && q->mq_ops->busy)
1611 return q->mq_ops->busy(q);
1615 EXPORT_SYMBOL_GPL(blk_lld_busy);
1618 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1619 * @rq: the clone request to be cleaned up
1622 * Free all bios in @rq for a cloned request.
1624 void blk_rq_unprep_clone(struct request *rq)
1628 while ((bio = rq->bio) != NULL) {
1629 rq->bio = bio->bi_next;
1634 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1637 * blk_rq_prep_clone - Helper function to setup clone request
1638 * @rq: the request to be setup
1639 * @rq_src: original request to be cloned
1640 * @bs: bio_set that bios for clone are allocated from
1641 * @gfp_mask: memory allocation mask for bio
1642 * @bio_ctr: setup function to be called for each clone bio.
1643 * Returns %0 for success, non %0 for failure.
1644 * @data: private data to be passed to @bio_ctr
1647 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1648 * Also, pages which the original bios are pointing to are not copied
1649 * and the cloned bios just point same pages.
1650 * So cloned bios must be completed before original bios, which means
1651 * the caller must complete @rq before @rq_src.
1653 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1654 struct bio_set *bs, gfp_t gfp_mask,
1655 int (*bio_ctr)(struct bio *, struct bio *, void *),
1658 struct bio *bio, *bio_src;
1663 __rq_for_each_bio(bio_src, rq_src) {
1664 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1668 if (bio_ctr && bio_ctr(bio, bio_src, data))
1672 rq->biotail->bi_next = bio;
1675 rq->bio = rq->biotail = bio;
1678 /* Copy attributes of the original request to the clone request. */
1679 rq->__sector = blk_rq_pos(rq_src);
1680 rq->__data_len = blk_rq_bytes(rq_src);
1681 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1682 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1683 rq->special_vec = rq_src->special_vec;
1685 rq->nr_phys_segments = rq_src->nr_phys_segments;
1686 rq->ioprio = rq_src->ioprio;
1689 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1696 blk_rq_unprep_clone(rq);
1700 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1702 int kblockd_schedule_work(struct work_struct *work)
1704 return queue_work(kblockd_workqueue, work);
1706 EXPORT_SYMBOL(kblockd_schedule_work);
1708 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1709 unsigned long delay)
1711 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1713 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1716 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1717 * @plug: The &struct blk_plug that needs to be initialized
1720 * blk_start_plug() indicates to the block layer an intent by the caller
1721 * to submit multiple I/O requests in a batch. The block layer may use
1722 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1723 * is called. However, the block layer may choose to submit requests
1724 * before a call to blk_finish_plug() if the number of queued I/Os
1725 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1726 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1727 * the task schedules (see below).
1729 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1730 * pending I/O should the task end up blocking between blk_start_plug() and
1731 * blk_finish_plug(). This is important from a performance perspective, but
1732 * also ensures that we don't deadlock. For instance, if the task is blocking
1733 * for a memory allocation, memory reclaim could end up wanting to free a
1734 * page belonging to that request that is currently residing in our private
1735 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1736 * this kind of deadlock.
1738 void blk_start_plug(struct blk_plug *plug)
1740 struct task_struct *tsk = current;
1743 * If this is a nested plug, don't actually assign it.
1748 INIT_LIST_HEAD(&plug->mq_list);
1749 INIT_LIST_HEAD(&plug->cb_list);
1751 plug->multiple_queues = false;
1754 * Store ordering should not be needed here, since a potential
1755 * preempt will imply a full memory barrier
1759 EXPORT_SYMBOL(blk_start_plug);
1761 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1763 LIST_HEAD(callbacks);
1765 while (!list_empty(&plug->cb_list)) {
1766 list_splice_init(&plug->cb_list, &callbacks);
1768 while (!list_empty(&callbacks)) {
1769 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1772 list_del(&cb->list);
1773 cb->callback(cb, from_schedule);
1778 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1781 struct blk_plug *plug = current->plug;
1782 struct blk_plug_cb *cb;
1787 list_for_each_entry(cb, &plug->cb_list, list)
1788 if (cb->callback == unplug && cb->data == data)
1791 /* Not currently on the callback list */
1792 BUG_ON(size < sizeof(*cb));
1793 cb = kzalloc(size, GFP_ATOMIC);
1796 cb->callback = unplug;
1797 list_add(&cb->list, &plug->cb_list);
1801 EXPORT_SYMBOL(blk_check_plugged);
1803 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1805 flush_plug_callbacks(plug, from_schedule);
1807 if (!list_empty(&plug->mq_list))
1808 blk_mq_flush_plug_list(plug, from_schedule);
1812 * blk_finish_plug - mark the end of a batch of submitted I/O
1813 * @plug: The &struct blk_plug passed to blk_start_plug()
1816 * Indicate that a batch of I/O submissions is complete. This function
1817 * must be paired with an initial call to blk_start_plug(). The intent
1818 * is to allow the block layer to optimize I/O submission. See the
1819 * documentation for blk_start_plug() for more information.
1821 void blk_finish_plug(struct blk_plug *plug)
1823 if (plug != current->plug)
1825 blk_flush_plug_list(plug, false);
1827 current->plug = NULL;
1829 EXPORT_SYMBOL(blk_finish_plug);
1831 void blk_io_schedule(void)
1833 /* Prevent hang_check timer from firing at us during very long I/O */
1834 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1837 io_schedule_timeout(timeout);
1841 EXPORT_SYMBOL_GPL(blk_io_schedule);
1843 int __init blk_dev_init(void)
1845 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1846 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1847 sizeof_field(struct request, cmd_flags));
1848 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1849 sizeof_field(struct bio, bi_opf));
1851 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1852 kblockd_workqueue = alloc_workqueue("kblockd",
1853 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1854 if (!kblockd_workqueue)
1855 panic("Failed to create kblockd\n");
1857 blk_requestq_cachep = kmem_cache_create("request_queue",
1858 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1860 #ifdef CONFIG_DEBUG_FS
1861 blk_debugfs_root = debugfs_create_dir("block", NULL);