2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
37 #include <linux/bpf.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/block.h>
44 #include "blk-mq-sched.h"
46 #include "blk-rq-qos.h"
48 #ifdef CONFIG_DEBUG_FS
49 struct dentry *blk_debugfs_root;
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
58 DEFINE_IDA(blk_queue_ida);
61 * For queue allocation
63 struct kmem_cache *blk_requestq_cachep;
66 * Controlling structure to kblockd
68 static struct workqueue_struct *kblockd_workqueue;
71 * blk_queue_flag_set - atomically set a queue flag
72 * @flag: flag to be set
75 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
77 set_bit(flag, &q->queue_flags);
79 EXPORT_SYMBOL(blk_queue_flag_set);
82 * blk_queue_flag_clear - atomically clear a queue flag
83 * @flag: flag to be cleared
86 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
88 clear_bit(flag, &q->queue_flags);
90 EXPORT_SYMBOL(blk_queue_flag_clear);
93 * blk_queue_flag_test_and_set - atomically test and set a queue flag
94 * @flag: flag to be set
97 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
98 * the flag was already set.
100 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
102 return test_and_set_bit(flag, &q->queue_flags);
104 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
106 void blk_rq_init(struct request_queue *q, struct request *rq)
108 memset(rq, 0, sizeof(*rq));
110 INIT_LIST_HEAD(&rq->queuelist);
112 rq->__sector = (sector_t) -1;
113 INIT_HLIST_NODE(&rq->hash);
114 RB_CLEAR_NODE(&rq->rb_node);
116 rq->internal_tag = -1;
117 rq->start_time_ns = ktime_get_ns();
120 EXPORT_SYMBOL(blk_rq_init);
122 static const struct {
126 [BLK_STS_OK] = { 0, "" },
127 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
128 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
129 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
130 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
131 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
132 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
133 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
134 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
135 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
136 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
137 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
139 /* device mapper special case, should not leak out: */
140 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
142 /* everything else not covered above: */
143 [BLK_STS_IOERR] = { -EIO, "I/O" },
146 blk_status_t errno_to_blk_status(int errno)
150 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
151 if (blk_errors[i].errno == errno)
152 return (__force blk_status_t)i;
155 return BLK_STS_IOERR;
157 EXPORT_SYMBOL_GPL(errno_to_blk_status);
159 int blk_status_to_errno(blk_status_t status)
161 int idx = (__force int)status;
163 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
165 return blk_errors[idx].errno;
167 EXPORT_SYMBOL_GPL(blk_status_to_errno);
169 static void print_req_error(struct request *req, blk_status_t status)
171 int idx = (__force int)status;
173 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
176 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu flags %x\n",
177 __func__, blk_errors[idx].name,
178 req->rq_disk ? req->rq_disk->disk_name : "?",
179 (unsigned long long)blk_rq_pos(req),
183 static void req_bio_endio(struct request *rq, struct bio *bio,
184 unsigned int nbytes, blk_status_t error)
187 bio->bi_status = error;
189 if (unlikely(rq->rq_flags & RQF_QUIET))
190 bio_set_flag(bio, BIO_QUIET);
192 bio_advance(bio, nbytes);
194 /* don't actually finish bio if it's part of flush sequence */
195 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
199 void blk_dump_rq_flags(struct request *rq, char *msg)
201 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
202 rq->rq_disk ? rq->rq_disk->disk_name : "?",
203 (unsigned long long) rq->cmd_flags);
205 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
206 (unsigned long long)blk_rq_pos(rq),
207 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
208 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
209 rq->bio, rq->biotail, blk_rq_bytes(rq));
211 EXPORT_SYMBOL(blk_dump_rq_flags);
214 * blk_sync_queue - cancel any pending callbacks on a queue
218 * The block layer may perform asynchronous callback activity
219 * on a queue, such as calling the unplug function after a timeout.
220 * A block device may call blk_sync_queue to ensure that any
221 * such activity is cancelled, thus allowing it to release resources
222 * that the callbacks might use. The caller must already have made sure
223 * that its ->make_request_fn will not re-add plugging prior to calling
226 * This function does not cancel any asynchronous activity arising
227 * out of elevator or throttling code. That would require elevator_exit()
228 * and blkcg_exit_queue() to be called with queue lock initialized.
231 void blk_sync_queue(struct request_queue *q)
233 del_timer_sync(&q->timeout);
234 cancel_work_sync(&q->timeout_work);
236 if (queue_is_mq(q)) {
237 struct blk_mq_hw_ctx *hctx;
240 cancel_delayed_work_sync(&q->requeue_work);
241 queue_for_each_hw_ctx(q, hctx, i)
242 cancel_delayed_work_sync(&hctx->run_work);
245 EXPORT_SYMBOL(blk_sync_queue);
248 * blk_set_pm_only - increment pm_only counter
249 * @q: request queue pointer
251 void blk_set_pm_only(struct request_queue *q)
253 atomic_inc(&q->pm_only);
255 EXPORT_SYMBOL_GPL(blk_set_pm_only);
257 void blk_clear_pm_only(struct request_queue *q)
261 pm_only = atomic_dec_return(&q->pm_only);
262 WARN_ON_ONCE(pm_only < 0);
264 wake_up_all(&q->mq_freeze_wq);
266 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
268 void blk_put_queue(struct request_queue *q)
270 kobject_put(&q->kobj);
272 EXPORT_SYMBOL(blk_put_queue);
274 void blk_set_queue_dying(struct request_queue *q)
276 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
279 * When queue DYING flag is set, we need to block new req
280 * entering queue, so we call blk_freeze_queue_start() to
281 * prevent I/O from crossing blk_queue_enter().
283 blk_freeze_queue_start(q);
286 blk_mq_wake_waiters(q);
288 /* Make blk_queue_enter() reexamine the DYING flag. */
289 wake_up_all(&q->mq_freeze_wq);
291 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
293 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
294 void blk_exit_queue(struct request_queue *q)
297 * Since the I/O scheduler exit code may access cgroup information,
298 * perform I/O scheduler exit before disassociating from the block
303 elevator_exit(q, q->elevator);
308 * Remove all references to @q from the block cgroup controller before
309 * restoring @q->queue_lock to avoid that restoring this pointer causes
310 * e.g. blkcg_print_blkgs() to crash.
315 * Since the cgroup code may dereference the @q->backing_dev_info
316 * pointer, only decrease its reference count after having removed the
317 * association with the block cgroup controller.
319 bdi_put(q->backing_dev_info);
323 * blk_cleanup_queue - shutdown a request queue
324 * @q: request queue to shutdown
326 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
327 * put it. All future requests will be failed immediately with -ENODEV.
329 void blk_cleanup_queue(struct request_queue *q)
331 /* mark @q DYING, no new request or merges will be allowed afterwards */
332 mutex_lock(&q->sysfs_lock);
333 blk_set_queue_dying(q);
335 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
336 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
337 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
338 mutex_unlock(&q->sysfs_lock);
341 * Drain all requests queued before DYING marking. Set DEAD flag to
342 * prevent that q->request_fn() gets invoked after draining finished.
348 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
351 * make sure all in-progress dispatch are completed because
352 * blk_freeze_queue() can only complete all requests, and
353 * dispatch may still be in-progress since we dispatch requests
354 * from more than one contexts.
356 * We rely on driver to deal with the race in case that queue
357 * initialization isn't done.
359 if (queue_is_mq(q) && blk_queue_init_done(q))
360 blk_mq_quiesce_queue(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 * I/O scheduler exit is only safe after the sysfs scheduler attribute
373 WARN_ON_ONCE(q->kobj.state_in_sysfs);
378 blk_mq_free_queue(q);
380 percpu_ref_exit(&q->q_usage_counter);
382 /* @q is and will stay empty, shutdown and put */
385 EXPORT_SYMBOL(blk_cleanup_queue);
387 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
389 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
391 EXPORT_SYMBOL(blk_alloc_queue);
394 * blk_queue_enter() - try to increase q->q_usage_counter
395 * @q: request queue pointer
396 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
398 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
400 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
403 bool success = false;
406 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
408 * The code that increments the pm_only counter is
409 * responsible for ensuring that that counter is
410 * globally visible before the queue is unfrozen.
412 if (pm || !blk_queue_pm_only(q)) {
415 percpu_ref_put(&q->q_usage_counter);
423 if (flags & BLK_MQ_REQ_NOWAIT)
427 * read pair of barrier in blk_freeze_queue_start(),
428 * we need to order reading __PERCPU_REF_DEAD flag of
429 * .q_usage_counter and reading .mq_freeze_depth or
430 * queue dying flag, otherwise the following wait may
431 * never return if the two reads are reordered.
435 wait_event(q->mq_freeze_wq,
436 (atomic_read(&q->mq_freeze_depth) == 0 &&
437 (pm || (blk_pm_request_resume(q),
438 !blk_queue_pm_only(q)))) ||
440 if (blk_queue_dying(q))
445 void blk_queue_exit(struct request_queue *q)
447 percpu_ref_put(&q->q_usage_counter);
450 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
452 struct request_queue *q =
453 container_of(ref, struct request_queue, q_usage_counter);
455 wake_up_all(&q->mq_freeze_wq);
458 static void blk_rq_timed_out_timer(struct timer_list *t)
460 struct request_queue *q = from_timer(q, t, timeout);
462 kblockd_schedule_work(&q->timeout_work);
466 * blk_alloc_queue_node - allocate a request queue
467 * @gfp_mask: memory allocation flags
468 * @node_id: NUMA node to allocate memory from
470 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
472 struct request_queue *q;
475 q = kmem_cache_alloc_node(blk_requestq_cachep,
476 gfp_mask | __GFP_ZERO, node_id);
480 INIT_LIST_HEAD(&q->queue_head);
481 q->last_merge = NULL;
483 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
487 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
491 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
492 if (!q->backing_dev_info)
495 q->stats = blk_alloc_queue_stats();
499 q->backing_dev_info->ra_pages =
500 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
501 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
502 q->backing_dev_info->name = "block";
505 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
506 laptop_mode_timer_fn, 0);
507 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
508 INIT_WORK(&q->timeout_work, NULL);
509 INIT_LIST_HEAD(&q->icq_list);
510 #ifdef CONFIG_BLK_CGROUP
511 INIT_LIST_HEAD(&q->blkg_list);
514 kobject_init(&q->kobj, &blk_queue_ktype);
516 #ifdef CONFIG_BLK_DEV_IO_TRACE
517 mutex_init(&q->blk_trace_mutex);
519 mutex_init(&q->sysfs_lock);
520 spin_lock_init(&q->queue_lock);
522 init_waitqueue_head(&q->mq_freeze_wq);
525 * Init percpu_ref in atomic mode so that it's faster to shutdown.
526 * See blk_register_queue() for details.
528 if (percpu_ref_init(&q->q_usage_counter,
529 blk_queue_usage_counter_release,
530 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
533 if (blkcg_init_queue(q))
539 percpu_ref_exit(&q->q_usage_counter);
541 blk_free_queue_stats(q->stats);
543 bdi_put(q->backing_dev_info);
545 bioset_exit(&q->bio_split);
547 ida_simple_remove(&blk_queue_ida, q->id);
549 kmem_cache_free(blk_requestq_cachep, q);
552 EXPORT_SYMBOL(blk_alloc_queue_node);
554 bool blk_get_queue(struct request_queue *q)
556 if (likely(!blk_queue_dying(q))) {
563 EXPORT_SYMBOL(blk_get_queue);
566 * blk_get_request - allocate a request
567 * @q: request queue to allocate a request for
568 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
569 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
571 struct request *blk_get_request(struct request_queue *q, unsigned int op,
572 blk_mq_req_flags_t flags)
576 WARN_ON_ONCE(op & REQ_NOWAIT);
577 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
579 req = blk_mq_alloc_request(q, op, flags);
580 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
581 q->mq_ops->initialize_rq_fn(req);
585 EXPORT_SYMBOL(blk_get_request);
587 static void part_round_stats_single(struct request_queue *q, int cpu,
588 struct hd_struct *part, unsigned long now,
589 unsigned int inflight)
592 __part_stat_add(cpu, part, time_in_queue,
593 inflight * (now - part->stamp));
594 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
600 * part_round_stats() - Round off the performance stats on a struct disk_stats.
601 * @q: target block queue
602 * @cpu: cpu number for stats access
603 * @part: target partition
605 * The average IO queue length and utilisation statistics are maintained
606 * by observing the current state of the queue length and the amount of
607 * time it has been in this state for.
609 * Normally, that accounting is done on IO completion, but that can result
610 * in more than a second's worth of IO being accounted for within any one
611 * second, leading to >100% utilisation. To deal with that, we call this
612 * function to do a round-off before returning the results when reading
613 * /proc/diskstats. This accounts immediately for all queue usage up to
614 * the current jiffies and restarts the counters again.
616 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
618 struct hd_struct *part2 = NULL;
619 unsigned long now = jiffies;
620 unsigned int inflight[2];
623 if (part->stamp != now)
627 part2 = &part_to_disk(part)->part0;
628 if (part2->stamp != now)
635 part_in_flight(q, part, inflight);
638 part_round_stats_single(q, cpu, part2, now, inflight[1]);
640 part_round_stats_single(q, cpu, part, now, inflight[0]);
642 EXPORT_SYMBOL_GPL(part_round_stats);
644 void blk_put_request(struct request *req)
646 blk_mq_free_request(req);
648 EXPORT_SYMBOL(blk_put_request);
650 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
653 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
655 if (!ll_back_merge_fn(q, req, bio))
658 trace_block_bio_backmerge(q, req, bio);
660 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
661 blk_rq_set_mixed_merge(req);
663 req->biotail->bi_next = bio;
665 req->__data_len += bio->bi_iter.bi_size;
667 blk_account_io_start(req, false);
671 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
674 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
676 if (!ll_front_merge_fn(q, req, bio))
679 trace_block_bio_frontmerge(q, req, bio);
681 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
682 blk_rq_set_mixed_merge(req);
684 bio->bi_next = req->bio;
687 req->__sector = bio->bi_iter.bi_sector;
688 req->__data_len += bio->bi_iter.bi_size;
690 blk_account_io_start(req, false);
694 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
697 unsigned short segments = blk_rq_nr_discard_segments(req);
699 if (segments >= queue_max_discard_segments(q))
701 if (blk_rq_sectors(req) + bio_sectors(bio) >
702 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
705 req->biotail->bi_next = bio;
707 req->__data_len += bio->bi_iter.bi_size;
708 req->nr_phys_segments = segments + 1;
710 blk_account_io_start(req, false);
713 req_set_nomerge(q, req);
718 * blk_attempt_plug_merge - try to merge with %current's plugged list
719 * @q: request_queue new bio is being queued at
720 * @bio: new bio being queued
721 * @request_count: out parameter for number of traversed plugged requests
722 * @same_queue_rq: pointer to &struct request that gets filled in when
723 * another request associated with @q is found on the plug list
724 * (optional, may be %NULL)
726 * Determine whether @bio being queued on @q can be merged with a request
727 * on %current's plugged list. Returns %true if merge was successful,
730 * Plugging coalesces IOs from the same issuer for the same purpose without
731 * going through @q->queue_lock. As such it's more of an issuing mechanism
732 * than scheduling, and the request, while may have elvpriv data, is not
733 * added on the elevator at this point. In addition, we don't have
734 * reliable access to the elevator outside queue lock. Only check basic
735 * merging parameters without querying the elevator.
737 * Caller must ensure !blk_queue_nomerges(q) beforehand.
739 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
740 struct request **same_queue_rq)
742 struct blk_plug *plug;
744 struct list_head *plug_list;
746 plug = current->plug;
750 plug_list = &plug->mq_list;
752 list_for_each_entry_reverse(rq, plug_list, queuelist) {
755 if (rq->q == q && same_queue_rq) {
757 * Only blk-mq multiple hardware queues case checks the
758 * rq in the same queue, there should be only one such
764 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
767 switch (blk_try_merge(rq, bio)) {
768 case ELEVATOR_BACK_MERGE:
769 merged = bio_attempt_back_merge(q, rq, bio);
771 case ELEVATOR_FRONT_MERGE:
772 merged = bio_attempt_front_merge(q, rq, bio);
774 case ELEVATOR_DISCARD_MERGE:
775 merged = bio_attempt_discard_merge(q, rq, bio);
788 void blk_init_request_from_bio(struct request *req, struct bio *bio)
790 if (bio->bi_opf & REQ_RAHEAD)
791 req->cmd_flags |= REQ_FAILFAST_MASK;
793 req->__sector = bio->bi_iter.bi_sector;
794 req->ioprio = bio_prio(bio);
795 req->write_hint = bio->bi_write_hint;
796 blk_rq_bio_prep(req->q, req, bio);
798 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
800 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
802 char b[BDEVNAME_SIZE];
804 printk(KERN_INFO "attempt to access beyond end of device\n");
805 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
806 bio_devname(bio, b), bio->bi_opf,
807 (unsigned long long)bio_end_sector(bio),
808 (long long)maxsector);
811 #ifdef CONFIG_FAIL_MAKE_REQUEST
813 static DECLARE_FAULT_ATTR(fail_make_request);
815 static int __init setup_fail_make_request(char *str)
817 return setup_fault_attr(&fail_make_request, str);
819 __setup("fail_make_request=", setup_fail_make_request);
821 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
823 return part->make_it_fail && should_fail(&fail_make_request, bytes);
826 static int __init fail_make_request_debugfs(void)
828 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
829 NULL, &fail_make_request);
831 return PTR_ERR_OR_ZERO(dir);
834 late_initcall(fail_make_request_debugfs);
836 #else /* CONFIG_FAIL_MAKE_REQUEST */
838 static inline bool should_fail_request(struct hd_struct *part,
844 #endif /* CONFIG_FAIL_MAKE_REQUEST */
846 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
848 const int op = bio_op(bio);
850 if (part->policy && op_is_write(op)) {
851 char b[BDEVNAME_SIZE];
853 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
857 "generic_make_request: Trying to write "
858 "to read-only block-device %s (partno %d)\n",
859 bio_devname(bio, b), part->partno);
860 /* Older lvm-tools actually trigger this */
867 static noinline int should_fail_bio(struct bio *bio)
869 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
873 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
876 * Check whether this bio extends beyond the end of the device or partition.
877 * This may well happen - the kernel calls bread() without checking the size of
878 * the device, e.g., when mounting a file system.
880 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
882 unsigned int nr_sectors = bio_sectors(bio);
884 if (nr_sectors && maxsector &&
885 (nr_sectors > maxsector ||
886 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
887 handle_bad_sector(bio, maxsector);
894 * Remap block n of partition p to block n+start(p) of the disk.
896 static inline int blk_partition_remap(struct bio *bio)
902 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
905 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
907 if (unlikely(bio_check_ro(bio, p)))
911 * Zone reset does not include bi_size so bio_sectors() is always 0.
912 * Include a test for the reset op code and perform the remap if needed.
914 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
915 if (bio_check_eod(bio, part_nr_sects_read(p)))
917 bio->bi_iter.bi_sector += p->start_sect;
918 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
919 bio->bi_iter.bi_sector - p->start_sect);
928 static noinline_for_stack bool
929 generic_make_request_checks(struct bio *bio)
931 struct request_queue *q;
932 int nr_sectors = bio_sectors(bio);
933 blk_status_t status = BLK_STS_IOERR;
934 char b[BDEVNAME_SIZE];
938 q = bio->bi_disk->queue;
941 "generic_make_request: Trying to access "
942 "nonexistent block-device %s (%Lu)\n",
943 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
948 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
949 * if queue is not a request based queue.
951 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
954 if (should_fail_bio(bio))
957 if (bio->bi_partno) {
958 if (unlikely(blk_partition_remap(bio)))
961 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
963 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
968 * Filter flush bio's early so that make_request based
969 * drivers without flush support don't have to worry
972 if (op_is_flush(bio->bi_opf) &&
973 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
974 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
981 switch (bio_op(bio)) {
983 if (!blk_queue_discard(q))
986 case REQ_OP_SECURE_ERASE:
987 if (!blk_queue_secure_erase(q))
990 case REQ_OP_WRITE_SAME:
991 if (!q->limits.max_write_same_sectors)
994 case REQ_OP_ZONE_RESET:
995 if (!blk_queue_is_zoned(q))
998 case REQ_OP_WRITE_ZEROES:
999 if (!q->limits.max_write_zeroes_sectors)
1007 * Various block parts want %current->io_context and lazy ioc
1008 * allocation ends up trading a lot of pain for a small amount of
1009 * memory. Just allocate it upfront. This may fail and block
1010 * layer knows how to live with it.
1012 create_io_context(GFP_ATOMIC, q->node);
1014 if (!blkcg_bio_issue_check(q, bio))
1017 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1018 trace_block_bio_queue(q, bio);
1019 /* Now that enqueuing has been traced, we need to trace
1020 * completion as well.
1022 bio_set_flag(bio, BIO_TRACE_COMPLETION);
1027 status = BLK_STS_NOTSUPP;
1029 bio->bi_status = status;
1035 * generic_make_request - hand a buffer to its device driver for I/O
1036 * @bio: The bio describing the location in memory and on the device.
1038 * generic_make_request() is used to make I/O requests of block
1039 * devices. It is passed a &struct bio, which describes the I/O that needs
1042 * generic_make_request() does not return any status. The
1043 * success/failure status of the request, along with notification of
1044 * completion, is delivered asynchronously through the bio->bi_end_io
1045 * function described (one day) else where.
1047 * The caller of generic_make_request must make sure that bi_io_vec
1048 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1049 * set to describe the device address, and the
1050 * bi_end_io and optionally bi_private are set to describe how
1051 * completion notification should be signaled.
1053 * generic_make_request and the drivers it calls may use bi_next if this
1054 * bio happens to be merged with someone else, and may resubmit the bio to
1055 * a lower device by calling into generic_make_request recursively, which
1056 * means the bio should NOT be touched after the call to ->make_request_fn.
1058 blk_qc_t generic_make_request(struct bio *bio)
1061 * bio_list_on_stack[0] contains bios submitted by the current
1063 * bio_list_on_stack[1] contains bios that were submitted before
1064 * the current make_request_fn, but that haven't been processed
1067 struct bio_list bio_list_on_stack[2];
1068 blk_mq_req_flags_t flags = 0;
1069 struct request_queue *q = bio->bi_disk->queue;
1070 blk_qc_t ret = BLK_QC_T_NONE;
1072 if (bio->bi_opf & REQ_NOWAIT)
1073 flags = BLK_MQ_REQ_NOWAIT;
1074 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
1075 blk_queue_enter_live(q);
1076 else if (blk_queue_enter(q, flags) < 0) {
1077 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
1078 bio_wouldblock_error(bio);
1084 if (!generic_make_request_checks(bio))
1088 * We only want one ->make_request_fn to be active at a time, else
1089 * stack usage with stacked devices could be a problem. So use
1090 * current->bio_list to keep a list of requests submited by a
1091 * make_request_fn function. current->bio_list is also used as a
1092 * flag to say if generic_make_request is currently active in this
1093 * task or not. If it is NULL, then no make_request is active. If
1094 * it is non-NULL, then a make_request is active, and new requests
1095 * should be added at the tail
1097 if (current->bio_list) {
1098 bio_list_add(¤t->bio_list[0], bio);
1102 /* following loop may be a bit non-obvious, and so deserves some
1104 * Before entering the loop, bio->bi_next is NULL (as all callers
1105 * ensure that) so we have a list with a single bio.
1106 * We pretend that we have just taken it off a longer list, so
1107 * we assign bio_list to a pointer to the bio_list_on_stack,
1108 * thus initialising the bio_list of new bios to be
1109 * added. ->make_request() may indeed add some more bios
1110 * through a recursive call to generic_make_request. If it
1111 * did, we find a non-NULL value in bio_list and re-enter the loop
1112 * from the top. In this case we really did just take the bio
1113 * of the top of the list (no pretending) and so remove it from
1114 * bio_list, and call into ->make_request() again.
1116 BUG_ON(bio->bi_next);
1117 bio_list_init(&bio_list_on_stack[0]);
1118 current->bio_list = bio_list_on_stack;
1120 bool enter_succeeded = true;
1122 if (unlikely(q != bio->bi_disk->queue)) {
1125 q = bio->bi_disk->queue;
1127 if (bio->bi_opf & REQ_NOWAIT)
1128 flags = BLK_MQ_REQ_NOWAIT;
1129 if (blk_queue_enter(q, flags) < 0) {
1130 enter_succeeded = false;
1135 if (enter_succeeded) {
1136 struct bio_list lower, same;
1138 /* Create a fresh bio_list for all subordinate requests */
1139 bio_list_on_stack[1] = bio_list_on_stack[0];
1140 bio_list_init(&bio_list_on_stack[0]);
1141 ret = q->make_request_fn(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 if (unlikely(!blk_queue_dying(q) &&
1159 (bio->bi_opf & REQ_NOWAIT)))
1160 bio_wouldblock_error(bio);
1164 bio = bio_list_pop(&bio_list_on_stack[0]);
1166 current->bio_list = NULL; /* deactivate */
1173 EXPORT_SYMBOL(generic_make_request);
1176 * direct_make_request - hand a buffer directly to its device driver for I/O
1177 * @bio: The bio describing the location in memory and on the device.
1179 * This function behaves like generic_make_request(), but does not protect
1180 * against recursion. Must only be used if the called driver is known
1181 * to not call generic_make_request (or direct_make_request) again from
1182 * its make_request function. (Calling direct_make_request again from
1183 * a workqueue is perfectly fine as that doesn't recurse).
1185 blk_qc_t direct_make_request(struct bio *bio)
1187 struct request_queue *q = bio->bi_disk->queue;
1188 bool nowait = bio->bi_opf & REQ_NOWAIT;
1191 if (!generic_make_request_checks(bio))
1192 return BLK_QC_T_NONE;
1194 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1195 if (nowait && !blk_queue_dying(q))
1196 bio->bi_status = BLK_STS_AGAIN;
1198 bio->bi_status = BLK_STS_IOERR;
1200 return BLK_QC_T_NONE;
1203 ret = q->make_request_fn(q, bio);
1207 EXPORT_SYMBOL_GPL(direct_make_request);
1210 * submit_bio - submit a bio to the block device layer for I/O
1211 * @bio: The &struct bio which describes the I/O
1213 * submit_bio() is very similar in purpose to generic_make_request(), and
1214 * uses that function to do most of the work. Both are fairly rough
1215 * interfaces; @bio must be presetup and ready for I/O.
1218 blk_qc_t submit_bio(struct bio *bio)
1221 * If it's a regular read/write or a barrier with data attached,
1222 * go through the normal accounting stuff before submission.
1224 if (bio_has_data(bio)) {
1227 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1228 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1230 count = bio_sectors(bio);
1232 if (op_is_write(bio_op(bio))) {
1233 count_vm_events(PGPGOUT, count);
1235 task_io_account_read(bio->bi_iter.bi_size);
1236 count_vm_events(PGPGIN, count);
1239 if (unlikely(block_dump)) {
1240 char b[BDEVNAME_SIZE];
1241 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1242 current->comm, task_pid_nr(current),
1243 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1244 (unsigned long long)bio->bi_iter.bi_sector,
1245 bio_devname(bio, b), count);
1249 return generic_make_request(bio);
1251 EXPORT_SYMBOL(submit_bio);
1254 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1255 * for new the queue limits
1257 * @rq: the request being checked
1260 * @rq may have been made based on weaker limitations of upper-level queues
1261 * in request stacking drivers, and it may violate the limitation of @q.
1262 * Since the block layer and the underlying device driver trust @rq
1263 * after it is inserted to @q, it should be checked against @q before
1264 * the insertion using this generic function.
1266 * Request stacking drivers like request-based dm may change the queue
1267 * limits when retrying requests on other queues. Those requests need
1268 * to be checked against the new queue limits again during dispatch.
1270 static int blk_cloned_rq_check_limits(struct request_queue *q,
1273 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1274 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1279 * queue's settings related to segment counting like q->bounce_pfn
1280 * may differ from that of other stacking queues.
1281 * Recalculate it to check the request correctly on this queue's
1284 blk_recalc_rq_segments(rq);
1285 if (rq->nr_phys_segments > queue_max_segments(q)) {
1286 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1294 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1295 * @q: the queue to submit the request
1296 * @rq: the request being queued
1298 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1300 if (blk_cloned_rq_check_limits(q, rq))
1301 return BLK_STS_IOERR;
1304 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1305 return BLK_STS_IOERR;
1307 if (blk_queue_io_stat(q))
1308 blk_account_io_start(rq, true);
1311 * Since we have a scheduler attached on the top device,
1312 * bypass a potential scheduler on the bottom device for
1315 return blk_mq_request_issue_directly(rq, true);
1317 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1320 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1321 * @rq: request to examine
1324 * A request could be merge of IOs which require different failure
1325 * handling. This function determines the number of bytes which
1326 * can be failed from the beginning of the request without
1327 * crossing into area which need to be retried further.
1330 * The number of bytes to fail.
1332 unsigned int blk_rq_err_bytes(const struct request *rq)
1334 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1335 unsigned int bytes = 0;
1338 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1339 return blk_rq_bytes(rq);
1342 * Currently the only 'mixing' which can happen is between
1343 * different fastfail types. We can safely fail portions
1344 * which have all the failfast bits that the first one has -
1345 * the ones which are at least as eager to fail as the first
1348 for (bio = rq->bio; bio; bio = bio->bi_next) {
1349 if ((bio->bi_opf & ff) != ff)
1351 bytes += bio->bi_iter.bi_size;
1354 /* this could lead to infinite loop */
1355 BUG_ON(blk_rq_bytes(rq) && !bytes);
1358 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1360 void blk_account_io_completion(struct request *req, unsigned int bytes)
1362 if (blk_do_io_stat(req)) {
1363 const int sgrp = op_stat_group(req_op(req));
1364 struct hd_struct *part;
1367 cpu = part_stat_lock();
1369 part_stat_add(cpu, part, sectors[sgrp], bytes >> 9);
1374 void blk_account_io_done(struct request *req, u64 now)
1377 * Account IO completion. flush_rq isn't accounted as a
1378 * normal IO on queueing nor completion. Accounting the
1379 * containing request is enough.
1381 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1382 const int sgrp = op_stat_group(req_op(req));
1383 struct hd_struct *part;
1386 cpu = part_stat_lock();
1389 part_stat_inc(cpu, part, ios[sgrp]);
1390 part_stat_add(cpu, part, nsecs[sgrp], now - req->start_time_ns);
1391 part_round_stats(req->q, cpu, part);
1392 part_dec_in_flight(req->q, part, rq_data_dir(req));
1394 hd_struct_put(part);
1399 void blk_account_io_start(struct request *rq, bool new_io)
1401 struct hd_struct *part;
1402 int rw = rq_data_dir(rq);
1405 if (!blk_do_io_stat(rq))
1408 cpu = part_stat_lock();
1412 part_stat_inc(cpu, part, merges[rw]);
1414 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1415 if (!hd_struct_try_get(part)) {
1417 * The partition is already being removed,
1418 * the request will be accounted on the disk only
1420 * We take a reference on disk->part0 although that
1421 * partition will never be deleted, so we can treat
1422 * it as any other partition.
1424 part = &rq->rq_disk->part0;
1425 hd_struct_get(part);
1427 part_round_stats(rq->q, cpu, part);
1428 part_inc_in_flight(rq->q, part, rw);
1436 * Steal bios from a request and add them to a bio list.
1437 * The request must not have been partially completed before.
1439 void blk_steal_bios(struct bio_list *list, struct request *rq)
1443 list->tail->bi_next = rq->bio;
1445 list->head = rq->bio;
1446 list->tail = rq->biotail;
1454 EXPORT_SYMBOL_GPL(blk_steal_bios);
1457 * blk_update_request - Special helper function for request stacking drivers
1458 * @req: the request being processed
1459 * @error: block status code
1460 * @nr_bytes: number of bytes to complete @req
1463 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1464 * the request structure even if @req doesn't have leftover.
1465 * If @req has leftover, sets it up for the next range of segments.
1467 * This special helper function is only for request stacking drivers
1468 * (e.g. request-based dm) so that they can handle partial completion.
1469 * Actual device drivers should use blk_end_request instead.
1471 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1472 * %false return from this function.
1475 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1476 * blk_rq_bytes() and in blk_update_request().
1479 * %false - this request doesn't have any more data
1480 * %true - this request has more data
1482 bool blk_update_request(struct request *req, blk_status_t error,
1483 unsigned int nr_bytes)
1487 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1492 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1493 !(req->rq_flags & RQF_QUIET)))
1494 print_req_error(req, error);
1496 blk_account_io_completion(req, nr_bytes);
1500 struct bio *bio = req->bio;
1501 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1503 if (bio_bytes == bio->bi_iter.bi_size)
1504 req->bio = bio->bi_next;
1506 /* Completion has already been traced */
1507 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1508 req_bio_endio(req, bio, bio_bytes, error);
1510 total_bytes += bio_bytes;
1511 nr_bytes -= bio_bytes;
1522 * Reset counters so that the request stacking driver
1523 * can find how many bytes remain in the request
1526 req->__data_len = 0;
1530 req->__data_len -= total_bytes;
1532 /* update sector only for requests with clear definition of sector */
1533 if (!blk_rq_is_passthrough(req))
1534 req->__sector += total_bytes >> 9;
1536 /* mixed attributes always follow the first bio */
1537 if (req->rq_flags & RQF_MIXED_MERGE) {
1538 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1539 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1542 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1544 * If total number of sectors is less than the first segment
1545 * size, something has gone terribly wrong.
1547 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1548 blk_dump_rq_flags(req, "request botched");
1549 req->__data_len = blk_rq_cur_bytes(req);
1552 /* recalculate the number of segments */
1553 blk_recalc_rq_segments(req);
1558 EXPORT_SYMBOL_GPL(blk_update_request);
1560 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1563 if (bio_has_data(bio))
1564 rq->nr_phys_segments = bio_phys_segments(q, bio);
1565 else if (bio_op(bio) == REQ_OP_DISCARD)
1566 rq->nr_phys_segments = 1;
1568 rq->__data_len = bio->bi_iter.bi_size;
1569 rq->bio = rq->biotail = bio;
1572 rq->rq_disk = bio->bi_disk;
1575 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1577 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1578 * @rq: the request to be flushed
1581 * Flush all pages in @rq.
1583 void rq_flush_dcache_pages(struct request *rq)
1585 struct req_iterator iter;
1586 struct bio_vec bvec;
1588 rq_for_each_segment(bvec, rq, iter)
1589 flush_dcache_page(bvec.bv_page);
1591 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1595 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1596 * @q : the queue of the device being checked
1599 * Check if underlying low-level drivers of a device are busy.
1600 * If the drivers want to export their busy state, they must set own
1601 * exporting function using blk_queue_lld_busy() first.
1603 * Basically, this function is used only by request stacking drivers
1604 * to stop dispatching requests to underlying devices when underlying
1605 * devices are busy. This behavior helps more I/O merging on the queue
1606 * of the request stacking driver and prevents I/O throughput regression
1607 * on burst I/O load.
1610 * 0 - Not busy (The request stacking driver should dispatch request)
1611 * 1 - Busy (The request stacking driver should stop dispatching request)
1613 int blk_lld_busy(struct request_queue *q)
1615 if (queue_is_mq(q) && q->mq_ops->busy)
1616 return q->mq_ops->busy(q);
1620 EXPORT_SYMBOL_GPL(blk_lld_busy);
1623 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1624 * @rq: the clone request to be cleaned up
1627 * Free all bios in @rq for a cloned request.
1629 void blk_rq_unprep_clone(struct request *rq)
1633 while ((bio = rq->bio) != NULL) {
1634 rq->bio = bio->bi_next;
1639 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1642 * Copy attributes of the original request to the clone request.
1643 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1645 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1647 dst->__sector = blk_rq_pos(src);
1648 dst->__data_len = blk_rq_bytes(src);
1649 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1650 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1651 dst->special_vec = src->special_vec;
1653 dst->nr_phys_segments = src->nr_phys_segments;
1654 dst->ioprio = src->ioprio;
1655 dst->extra_len = src->extra_len;
1659 * blk_rq_prep_clone - Helper function to setup clone request
1660 * @rq: the request to be setup
1661 * @rq_src: original request to be cloned
1662 * @bs: bio_set that bios for clone are allocated from
1663 * @gfp_mask: memory allocation mask for bio
1664 * @bio_ctr: setup function to be called for each clone bio.
1665 * Returns %0 for success, non %0 for failure.
1666 * @data: private data to be passed to @bio_ctr
1669 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1670 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1671 * are not copied, and copying such parts is the caller's responsibility.
1672 * Also, pages which the original bios are pointing to are not copied
1673 * and the cloned bios just point same pages.
1674 * So cloned bios must be completed before original bios, which means
1675 * the caller must complete @rq before @rq_src.
1677 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1678 struct bio_set *bs, gfp_t gfp_mask,
1679 int (*bio_ctr)(struct bio *, struct bio *, void *),
1682 struct bio *bio, *bio_src;
1687 __rq_for_each_bio(bio_src, rq_src) {
1688 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1692 if (bio_ctr && bio_ctr(bio, bio_src, data))
1696 rq->biotail->bi_next = bio;
1699 rq->bio = rq->biotail = bio;
1702 __blk_rq_prep_clone(rq, rq_src);
1709 blk_rq_unprep_clone(rq);
1713 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1715 int kblockd_schedule_work(struct work_struct *work)
1717 return queue_work(kblockd_workqueue, work);
1719 EXPORT_SYMBOL(kblockd_schedule_work);
1721 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1723 return queue_work_on(cpu, kblockd_workqueue, work);
1725 EXPORT_SYMBOL(kblockd_schedule_work_on);
1727 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1728 unsigned long delay)
1730 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1732 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1735 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1736 * @plug: The &struct blk_plug that needs to be initialized
1739 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1740 * pending I/O should the task end up blocking between blk_start_plug() and
1741 * blk_finish_plug(). This is important from a performance perspective, but
1742 * also ensures that we don't deadlock. For instance, if the task is blocking
1743 * for a memory allocation, memory reclaim could end up wanting to free a
1744 * page belonging to that request that is currently residing in our private
1745 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1746 * this kind of deadlock.
1748 void blk_start_plug(struct blk_plug *plug)
1750 struct task_struct *tsk = current;
1753 * If this is a nested plug, don't actually assign it.
1758 INIT_LIST_HEAD(&plug->mq_list);
1759 INIT_LIST_HEAD(&plug->cb_list);
1761 plug->multiple_queues = false;
1764 * Store ordering should not be needed here, since a potential
1765 * preempt will imply a full memory barrier
1769 EXPORT_SYMBOL(blk_start_plug);
1771 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1773 LIST_HEAD(callbacks);
1775 while (!list_empty(&plug->cb_list)) {
1776 list_splice_init(&plug->cb_list, &callbacks);
1778 while (!list_empty(&callbacks)) {
1779 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1782 list_del(&cb->list);
1783 cb->callback(cb, from_schedule);
1788 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1791 struct blk_plug *plug = current->plug;
1792 struct blk_plug_cb *cb;
1797 list_for_each_entry(cb, &plug->cb_list, list)
1798 if (cb->callback == unplug && cb->data == data)
1801 /* Not currently on the callback list */
1802 BUG_ON(size < sizeof(*cb));
1803 cb = kzalloc(size, GFP_ATOMIC);
1806 cb->callback = unplug;
1807 list_add(&cb->list, &plug->cb_list);
1811 EXPORT_SYMBOL(blk_check_plugged);
1813 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1815 flush_plug_callbacks(plug, from_schedule);
1817 if (!list_empty(&plug->mq_list))
1818 blk_mq_flush_plug_list(plug, from_schedule);
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 int __init blk_dev_init(void)
1833 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1834 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1835 FIELD_SIZEOF(struct request, cmd_flags));
1836 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1837 FIELD_SIZEOF(struct bio, bi_opf));
1839 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1840 kblockd_workqueue = alloc_workqueue("kblockd",
1841 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1842 if (!kblockd_workqueue)
1843 panic("Failed to create kblockd\n");
1845 blk_requestq_cachep = kmem_cache_create("request_queue",
1846 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1848 #ifdef CONFIG_DEBUG_FS
1849 blk_debugfs_root = debugfs_create_dir("block", NULL);